Blog

AI job matching tools promise to find roles that fit you, but nobody can tell you in advance whether a given tool will work for your field, your level, and your location, so the only reliable answer is a short test you run yourself.

That’s the problem with most advice on these tools. A roundup can list names and features, but the thing you actually need to know is whether a tool’s idea of “a good match” agrees with yours. Matching tools differ in what they look at, how they score, and how much they explain. Some suggest roles, some rank them, and some apply for you. I don’t cite any tool’s accuracy claims here; the test below checks a tool against your own judgment.

This guide gives you a short test that works on any matching tool, free or paid. You’ll pick a handful of postings you already understand, see how the tool rates them, compare its opinion with yours, check how it treats your data, and decide with a simple scorecard. A chatbot can do the comparison for you if you stop it from making judgments it can’t support. The tool and results in the examples are invented. This is general information, not career or legal advice, and every site has its own terms, so read them.

Two Kinds of Tools, and Why It Matters

Before testing anything, know which kind you’re looking at.

• Discovery: Suggests and scores roles.
• Auto-apply: Fills and submits for you.

Discovery tools leave the decision with you. Auto-apply tools act on your behalf, which raises the stakes: a wrong answer in a form goes out under your name. Our guide to auto apply job applications covers the safety checks for that second kind. This guide focuses on testing how well any tool’s matching matches reality.

What You’re Actually Testing

A matching tool can fail in a few distinct ways. Test for each.

• Fit. Does it recommend roles you could realistically get?
• Level. Does it get seniority right, or mix up junior and senior roles?
• Location and terms. Does it respect your location, remote preference, and other limits?
• Explanation. Does it say why a role scored the way it did?
• Duplicates and junk. Does it repeat postings or show ones that are closed?
• Your data. What does it ask for, and what does it say it does with it?

You don’t need to test everything deeply. A quick pass on each will tell you a lot.

Six Steps to Test Any Tool

Step One: Build Your Benchmark

Pick five postings you can judge yourself. Use the method from our job posting requirements checklist to mark each one beforehand, without looking at the tool.

• Posting 1: Strong fit.
• Posting 2: Strong fit.
• Posting 3: Stretch.
• Posting 4: Stretch.
• Posting 5: Not a fit.

Choose a mix on purpose. Two clear fits, two stretches, and one that clearly isn’t right. The tool should score the fits higher than the non-fit. If it can’t tell the difference between your best and worst posting, that’s useful to know.

Write your marks down before you run the tool. If you decide after seeing its scores, you’ll be tempted to agree with it.

Step Two: Run the Tool

Give the tool your resume or profile, and have it evaluate the same five postings. Some tools show a score, a ranking, or a label such as “great match.” Record exactly what it says for each posting, including any explanation.

Posting	Tool’s result
1	High
2	High
3	High
4	Low
5	Medium

These results are invented. Notice what you can already see: the tool agrees on 1 and 2, rates the stretch at 3 as high, rates stretch 4 as low, and gives a medium rating to the one that isn’t a fit.

Step Three: Compare With Your Marks

Count agreements. Map your marks to rough levels: strong fit to high, stretch to medium or low, not a fit to low.

In the invented example:

• Posting 1: you said strong fit, tool said high. Agree.
• Posting 2: strong fit, high. Agree.
• Posting 3: stretch, high. Disagree (the tool is more optimistic than you).
• Posting 4: stretch, low. Agree, since a stretch can reasonably score low.
• Posting 5: not a fit, medium. Disagree (the tool gave a non-fit a medium rating).

That’s 3 agreements out of 5, or 60%. With five postings, that number is rough. One disagreement changes it by 20 points. Treat it as a first impression, not a measurement, and increase the sample if the tool looks promising.

Step Four: Check the Explanations

A score without a reason is hard to trust. For the postings where the tool disagreed with you, look for any explanation. Does it point to specific skills, years, or keywords? Are they correct? If the tool says “strong match because of your project management experience” and you have none, that’s an error in its reading of your profile, and you should fix your profile or note the tool’s misreading.

If the tool gives no explanation at all, note that. You can still use it for discovery, but you shouldn’t lean on its scores.

Step Five: Look at How It Handles Your Data

Your resume contains personal information: your name, contact details, work history, and maybe your address. Before you commit, read the tool’s privacy policy and look for:

• What data does it collect?: Privacy policy.
• Is it shared with others?: Sharing section.
• Can I delete my data?: Account or policy.
• Does it need my passwords?: Setup steps.

If a tool asks for passwords to other accounts, think carefully. Use unique passwords, and consider whether the tool is worth the risk. If something is unclear, treat it as unknown, not as safe.

Step Six: Read the Cost and Cancellation Terms

Free trials can turn into paid plans. Before you enter payment details, find out what the price is after the trial, when it renews, and how to cancel. Write the answers down, and set a calendar reminder a few days before the trial ends. You’re checking terms, not trusting claims, so use the tool’s own pages and screenshots.

Let a Chatbot Compare the Results

A chatbot can build the comparison table, as long as it doesn’t judge the tool. Remove personal details first. Use this prompt:

I tested a job matching tool on five postings. Below are my own marks (made before running the tool) and the tool's results.

My marks:
[paste]

Tool's results (exactly as shown, including any explanation):
[paste]

Please:
1. Make a table with one row per posting: my mark, the tool's result, and whether they AGREE or DISAGREE. Treat "strong fit" as matching "high," "not a fit" as matching "low," and a "stretch" as matching either "medium" or "low."
2. Count agreements and show the fraction and percentage.
3. For each disagreement, quote any explanation the tool gave, or write "NO EXPLANATION."
4. List anything the tool did not say that I'd need to judge it, marked "NOT STATED."

Rules:
- Use only what I pasted.
- Do not say whether the tool is good, accurate, or worth using.
- Do not guess how the tool works.

I wrote this prompt myself as a suggestion; no source publishes it as a standard.

You get a table, a fraction, and a list of questions. The judgment stays with you.

A Scorecard You Can Reuse

After the test, give the tool a simple score from 0 to 2 on each point, and add them up.

Check	Score (0-2)
Fit agreement
Level accuracy
Respects my limits
Gives reasons
Clear data terms
Clear price terms

A total out of 12 gives you a rough way to compare two tools. It isn’t scientific, but it keeps you from choosing on a hunch. If you test two tools on the same five postings, you’ll see the differences clearly.

When to Retest

Tools change. Run the five-posting test again after a major update, or every few months if you keep using the tool. Keep your benchmark postings in a file, so you can rerun the test quickly. You can also swap in new postings, but note that your results will then be harder to compare.

What a Good Result Looks Like

You don’t need a perfect score. A useful tool might agree with you on most fits, miss some stretches, and offer reasons for its picks. A tool worth dropping might rate everything high, give no reasons, or ask for more access than seems reasonable.

Using a Tool for Just One Job

You don’t have to adopt a tool wholesale. You can use a matching tool only for discovery, copy the postings that look promising, and then run their own checklist before applying. That way the tool’s scores are one input, not the decision. If a tool is good at surfacing roles but weak at explaining them, that arrangement works fine. Judging AI job matching tools by what each one does well, instead of whether it does everything, gives you a more realistic and less expensive setup.

Keep Your Benchmark Honest

Over time, your own situation changes. You gain skills, shift your target level, or move to a new city. Update your benchmark postings when that happens, and re-mark them before you run the tool again. A test built for last year’s search can mislead you this year.

Mistakes That Waste a Free Trial

• Judging by the first screen. Run the five-posting test.
• Marking your answers after seeing the tool’s. Decide first.
• Trusting a high score with no explanation. Ask why.
• Skipping the privacy policy. Your resume is personal.
• Forgetting the end of the trial. Set a reminder.
• Treating one test as proof. Retest, and add postings.

Questions About AI Job Matching Tools

What are AI job matching tools? They’re tools that suggest or rank job openings based on your resume and preferences. Some also fill out applications for you.

How can I tell if a matching tool is accurate? Compare its ratings with your own marks on a handful of postings you’ve judged yourself, and check whether it explains its picks.

Are the accuracy numbers on tool websites reliable? They’re hard to compare, because each company measures differently. A test on your own postings tells you more.

Should I give a tool my passwords? Think carefully. Use a unique password and consider whether the tool is worth the risk.

Can AI compare the results for me? It can build the table and count agreements if you forbid judgments. You decide what the results mean.

How often should I retest? After major updates, or every few months.

Run Your First Five-Posting Test

Pick five postings, mark them yourself, and run one tool on them. AI job matching tools vary a great deal, and a short test gives you a clearer answer than any list could.

This article offers general information, not career or legal advice. Each site and tool has its own terms, which you should read. The example tool, postings, and results are fictional.

A job posting requirements checklist turns a long, wordy listing into a short list of what the employer must have, would like to have, and hasn’t made clear, so you can decide whether to apply before you spend time on a resume.

Getting no reply can begin with the decision to apply. A posting lists six “requirements,” you match three, you apply anyway, and then you wait. Sometimes that works. When it doesn’t, you may never learn why. A checklist won’t guarantee a reply, but it shows you where you’re strong, where you’re stretching, and where tailoring is worth your time.

This guide shows how to break a posting into parts, how to count your real matches with evidence, how to work out “years of experience” without fudging, and how to use a chatbot to do the sorting without letting it decide for you. The posting and person in the examples are invented. This is general information, not career or legal advice.

Why the Decision Comes First

Tailoring a resume is useful, but only if the job is a realistic target. Spending your best effort on a posting you clearly don’t fit leaves less time for ones you do. So the first fix is simple: decide with a checklist, not a feeling.

The checklist does three jobs:

• It separates what’s required from what’s a bonus.
• It shows which requirements you can prove.
• It tells you what to say first in your resume.

How to Build the Checklist

Splitting a Posting Into Four Buckets

Read the posting and sort every line into one of four buckets.

Bucket	Clue words
Must-have	required, must, minimum
Nice-to-have	preferred, a plus, ideally
Daily duties	you will, responsible for
Unclear	vague words, mixed signals

Some postings mix these up. A line like “3+ years of experience preferred” sits between two buckets. Don’t resolve it by guessing. Put it in Unclear and decide later.

Your Facts Sheet Comes First

Before matching anything, write one plain list of your own facts: each job with start and end dates, tools you’ve used, certifications, and results you can back up. Keep it in a file you update. Everything you claim in an application should come from this sheet, not from memory on the day.

This sheet is also what you’ll paste into a chatbot, so it should hold only things that are true and that you could explain in an interview.

Counting Your Matches

For each must-have, write one of three marks next to it.

• Proof: I can show it with a real example.
• Partly: Close, but not exact.
• None: I don’t have it.

Then count. If you have proof for most must-haves and partial matches for the rest, apply. If several are “None,” think hard. That doesn’t mean you can’t be hired, since postings are sometimes wish lists. It means the posting asks for more than you can prove today.

You decide the threshold. There’s no official number, so choose your own threshold and use it consistently, because consistency is what lets you learn from the results.

Working Out Years of Experience Honestly

Experience requirements trip people up, because dates don’t add up the way they feel. Use months.

Take an invented person with two relevant jobs:

• Job 1: March 2019 to August 2022. From March 2019 to March 2022 is 36 months, and to August adds 5 more, so 41 months.
• Job 2: October 2022 to January 2024. From October 2022 to October 2023 is 12 months, and to January 2024 adds 3, so 15 months.

Total: 41 plus 15 equals 56 months. Divide by 12 and you get 4 years and 8 months, or about 4.7 years.

Rules for doing this fairly:

• Count only the months where you actually did the relevant work.
• Don’t count overlapping jobs twice.
• Don’t round up to look better. If a posting asks for “5+ years” and you have 4 years 8 months, that’s a “Partly.”

Write the result in your facts sheet so you don’t have to recalculate for each application.

Let a Chatbot Sort the Posting

A chatbot can split a posting quickly. The risk is that it paraphrases or invents. This prompt prevents that:

Below is a job posting, copied exactly. Please sort it into buckets using only the text.

[paste the posting]

Make four lists:
1. MUST-HAVE: lines that use words like "required," "must," or "minimum." Quote each line.
2. NICE-TO-HAVE: lines that use words like "preferred" or "a plus." Quote each line.
3. DAILY DUTIES: lines that describe what the person will do. Quote each line.
4. UNCLEAR: lines that could be either must-have or nice-to-have. Quote each line and say why it's unclear.

Rules:
- Quote the posting exactly. Do not paraphrase.
- Do not add requirements that are not written.
- Do not tell me whether I should apply.

This prompt is my suggestion, not an official template.

Then a second step, using your facts sheet:

Below are the MUST-HAVE lines from the posting and my facts sheet. For each must-have, write PROOF, PARTLY, or NONE. For PROOF or PARTLY, quote the line from my facts sheet that supports it. If I have no matching line, write NONE.

Rules:
- Use only my facts sheet. Do not assume any skill I did not list.
- For years of experience, calculate in months from my dates and show the arithmetic.
- Do not suggest that I add anything I haven't done.

Must-haves:
[paste the must-haves]

My facts sheet:
[paste my facts sheet]

The most valuable rule is the one against assuming. If the chatbot starts crediting you with skills you didn’t list, the checklist stops being honest.

A Worked Example With an Invented Posting

Suppose an invented “Operations Coordinator” posting has these must-haves: three or more years in operations, scheduling software experience, vendor communication, a bachelor’s degree, and Spanish fluency. The nice-to-haves include a project management certification.

Matching against the invented facts sheet:

Must-have	Mark
3+ years in operations	Proof (56 months)
Scheduling software	Proof
Vendor communication	Proof
Bachelor’s degree	Proof
Spanish fluency	None

Four of five with proof, and one missing. The unclear line, “strong Excel skills,” goes to Unclear. The decision might be: apply, don’t claim Spanish, and put the four proven items near the top. In my view, a missing item that you openly don’t claim is better than a vague claim you can’t back up. The posting, person, and results are all invented.

Using the Checklist Across Your Search

Write Your Resume From the Checklist

Once you’ve marked your matches, the checklist doubles as an outline.

• Put the “Proof” items first, using the posting’s own wording where it honestly fits.
• Mention “Partly” items carefully, saying exactly what you did.
• Leave out “None” items. Don’t stuff words to cover them.

Keeping your language honest also keeps your interview easy. You’ll be asked about what’s on the page.

Set a Time Limit

A checklist should save time, not eat it. Pick a time limit per posting before you start, and stop when it runs out. If you find yourself agonizing, that may be a sign the match is weaker than you hoped.

A Note on What Postings Don’t Say

A posting may not list everything the team values. The checklist is a tool for reading what’s there, and it can’t tell you what the hiring team actually values. When possible, check the company’s own website for clues, and consider asking a contact who works there.

Build a Weekly Routine Around the Checklist

A checklist works best as a habit. Set aside one block each week to collect the postings you’re interested in and run each through the same steps. Sort them into three piles: strong matches, stretches, and skips. Spend your tailoring time on the strong matches first. For stretches, decide whether a short, honest note about what you can offer is worth the effort. Skips go in a folder, and you don’t look at them again.

At the end of the week, count how many postings fell into each pile. In my view, if almost everything lands in “skip,” your search terms may be off, or you may be aiming at a level you aren’t ready for yet. If almost everything is a “strong match,” you may be reading requirements too generously. The counts tell you something, and a job posting requirements checklist makes those counts easy to keep.

When a Requirement Is Missing From Your Resume but True

Sometimes you do have a skill the posting wants, and your resume simply doesn’t say so. The checklist helps here too. When you mark something “None,” ask one more question: have I actually done this, without ever writing it down? If so, add it to your facts sheet with a concrete example, such as what you did, when, and what came of it. Then it can go on your resume honestly.

If you truly haven’t done it, leave it off. A resume that claims everything invites tough questions, and a resume that’s clear about what it offers lets the interview focus on your real strengths.

Mistakes That Waste Applications

• Treating every line as required. Sort first.
• Skipping the facts sheet. You’ll forget real strengths.
• Rounding experience up. Use months and report honestly.
• Letting a chatbot guess your skills. Forbid it.
• Applying to everything. Time spent on poor matches costs good ones.
• Claiming a skill you don’t have. It can show up in the interview.

Questions About Job Posting Requirements Checklist

What is a job posting requirements checklist? It’s a short sort of a posting into must-haves, nice-to-haves, duties, and unclear lines, with a mark for how well you match each.

Should I apply if I don’t meet every requirement? That’s your call. The checklist shows how many you can prove, so you can decide on your own threshold.

How do I count years of experience? Count months of relevant work, avoid double counting overlaps, divide by 12, and report honestly.

Can AI decide whether I should apply? It can sort and quote, but the decision should be yours.

What if a requirement is unclear? Put it in the Unclear bucket and look for hints elsewhere in the posting, or on the company’s site.

Does this guarantee a reply? No. It helps you choose better targets and write a clearer resume.

Run One Posting Through It Today

Pick a posting you’re considering, sort it into the four buckets, and mark your matches against your facts sheet. A job posting requirements checklist turns “should I apply?” from a guess into a count you can explain.

Once you decide to apply, a job application spreadsheet keeps every decision in one place, and a fake job posting checklist helps you screen out postings that are not real.

This article offers general information, not career or legal advice. Job postings and hiring practices vary widely. The example posting, person, and results are fictional.

Per-image pricing vs subscription is a question that can get answered by gut feeling, sometimes after uploading an entire gallery, and the right answer depends far more on your busiest shoot than on your average one.

A per-image price looks tiny. Five cents. Ten cents. That can seem harmless on a pricing page, until a single large event turns it into a bill with a comma in it. A flat monthly fee looks expensive in a slow month and like a bargain in a busy one. Both models can be the cheaper choice. Which one depends on numbers you already have and probably haven’t written down.

This guide shows you how to calculate it: how to find the break-even number of photos, how to compare against your real year instead of your best or worst month, how culling first changes the answer, and how to use a chatbot to check your arithmetic without letting it guess. All prices and photo counts here are invented and don’t describe any real product. This is general information, not financial advice.

Why “Cheap Per Photo” Can Still Be Expensive

Per-image pricing scales with your volume, so your cost rises every time you shoot more. Subscription pricing stays flat until you hit its limits, so your cost per photo falls the more you shoot. That’s the whole difference. Neither is better in general. The question is where your work sits between the two.

A few things make it hard to judge by feel:

• Your shoots vary a lot. A portrait session might be 150 photos, and a wedding might be 1,200 or more.
• Months aren’t equal. Subscriptions bill monthly even when you shoot nothing, though some services may let you pause.
• Free trials can hide the real cost. A trial allowance can feel generous until your first big gallery uses it up in minutes.
• Fine print matters. Does the price count rejected photos, re-runs, or second passes? That changes the math.

Three Ways People Choose

• Uses your real photo counts: Go with the first free trial – No; Compare the headline prices – No; Calculate from your own shoots – Yes.
• Shows the break-even point: Go with the first free trial – No; Compare the headline prices – No; Calculate from your own shoots – Yes.
• Catches the big-gallery problem: Go with the first free trial – No; Compare the headline prices – Rarely; Calculate from your own shoots – Yes.
• Risk of a surprise bill: Go with the first free trial – High; Compare the headline prices – Medium; Calculate from your own shoots – Low.

Four Steps to Your Own Break-Even

Step One: Gather Your Own Numbers

Before opening a pricing page, write down:

1. Photos per shoot, by shoot type. Include your largest typical job, not only the average.
2. Shoots per month, or at least per year, by type.
3. How many photos you keep after culling. If you edit every frame you shoot, use the full count.

You don’t need exact numbers, but they should be honest. Use last year’s records if you have them.

Step Two: Find the Break-Even Number

With any per-image price and any flat monthly fee, the break-even number of photos per month is:

Monthly fee ÷ price per image

Below that many photos in a month, per-image pricing is cheaper. Above it, the subscription is cheaper.

Using invented numbers: a per-image price of $0.05 and a flat fee of $30 a month. The break-even is 30 divided by 0.05, which is 600 photos a month.

Photos in the month	Per-image cost	Subscription cost	Cheaper
150 (one portrait session)	$7.50	$30	Per-image
300 (two portrait sessions)	$15	$30	Per-image
600	$30	$30	Tie
1,200 (one wedding)	$60	$30	Subscription
3,000 (large event)	$150	$30	Subscription

So a month with one wedding costs twice as much per-image as on a flat fee, and a big event costs five times as much. If your work is mostly small sessions, the opposite holds. This is why the question “which is cheaper” has no answer until you know your own pattern.

Step Three: Test Against a Whole Year

Months with a wedding favor the subscription, and quiet months favor per-image. To see which wins overall, add up the year. Suppose an invented photographer shoots ten portrait sessions of 150 photos and four weddings of 1,200 photos in a year:

• Portrait photos: 10 times 150 equals 1,500
• Wedding photos: 4 times 1,200 equals 4,800
• Total: 1,500 plus 4,800 equals 6,300 photos
• Per-image cost: 6,300 times $0.05 equals $315
• Subscription cost: 12 months times $30 equals $360

On these numbers, per-image is $45 cheaper over the year, even though each wedding costs more under it. The subscription pays for months with nothing to edit. If the same photographer shot six weddings, the total would be 1,500 plus 7,200, or 8,700 photos, costing $435 per-image, which is more than the $360 subscription. A small change in your shoot mix flips the answer, so it’s worth doing the arithmetic for your own year, and then testing a busier one.

Step Four: See What Culling First Does

Some photographers upload everything and pay for every frame, including blinks and near-duplicates. If you cull before sending files to be edited, the count falls.

Take the invented 1,200-photo wedding. If you keep 400 after culling, then per-image costs 400 times $0.05, or $20, instead of $60. That’s a $40 saving on one gallery, using the same tool and the same price. The subscription cost doesn’t change at all. Culling first makes per-image pricing look much better, so any comparison should use the number of photos you’d actually upload, not the number you shot.

Let a Chatbot Check Your Math

You can have a chatbot do the arithmetic, as long as it can’t invent prices. Paste the pricing text from the tool’s own page, and use this prompt:

I'm comparing two pricing models for photo editing. Use only the numbers and text below.

Pricing text copied from the tool's page:
[paste]

My shoots in a year:
[list: shoot type, photos uploaded per shoot, number of shoots]

Please:
1. State the per-image price and the monthly subscription price exactly as written in the pasted text. If either is not stated, write "NOT STATED."
2. Calculate the break-even number of photos per month (subscription price divided by per-image price). Show the division.
3. Calculate my total photos per year, my total yearly cost under per-image pricing, and my total yearly cost under the subscription. Show every step.
4. Make a table of the cost of one month with 150, 600, 1,200, and 3,000 photos under each model.
5. List any pricing terms in the text that could change the cost (for example, charges for re-runs, limits, overage, or trial allowances), and quote them.

Rules:
- Use only the pasted text and my numbers.
- Do not assume any price or limit that is not written.
- Do not recommend a tool or say which option is better. Just calculate.
- If something is unclear, write "UNCLEAR" and tell me what to check on the pricing page.

I wrote this prompt myself as a suggestion; no source publishes it as a standard.

Check the arithmetic with a calculator. If the chatbot says “NOT STATED” for a price, that’s a sign to look at the pricing page directly, or to contact the company before uploading anything.

Questions to Ask Before You Upload

Whatever the model, check these on the pricing page or with the company:

• Does the price apply to every photo uploaded, or only to photos I export?
• Are re-edits or re-runs charged?
• Is there a monthly cap, and what happens if I go over it?
• Does the trial allowance carry over, and does it count against my first gallery?
• Can I pause or cancel the subscription in months with no shoots?
• Is there a separate charge for any add-on step?

Write the answers down. If a term is unclear, treat it as unknown, not as free.

Revisit the Choice Once a Year

Your shoot mix changes. A year with more weddings or a new type of work can flip the answer, so set a yearly reminder to redo the calculation with your last twelve months of real photo counts. Keep the numbers in a simple table: shoot type, photos uploaded, and cost under each model.

Comparing per-image pricing vs subscription once a year can also catch price changes, since services update their plans. If you can, check whether you’d be able to switch without losing anything, such as saved editing profiles, before you commit to a longer plan.

Where the Calculation Goes Wrong

• Using your average shoot only. Use your largest typical one too.
• Comparing months, not years. A subscription costs something in slow months.
• Uploading before culling. It inflates the per-image total.
• Forgetting re-runs. If you redo a batch, you may pay again.
• Trusting a headline price. The terms behind it decide the real cost.
• Treating a trial as the price. The trial ends.

Questions About Per-Image Pricing vs Subscription

Is per-image pricing vs subscription pricing better for weddings? Large galleries can favor flat fees, because per-image cost rises with every photo. But it depends on the actual prices, so calculate with your numbers.

How do I find the break-even point? Divide the monthly subscription price by the per-image price. That’s the number of photos per month where the two cost the same.

Should I cull before uploading? It can lower a per-image bill, since you pay for fewer photos.

What if my work varies a lot month to month? Total up a full year and compare it with twelve months of the subscription, then test a busier year.

Can a chatbot compare the prices for me? It can calculate if you paste the pricing text and forbid guessing.

Do prices change? Yes, so take the figures from the tool’s own pricing page on the day you decide.

Run the Numbers on Your Biggest Shoot

Write down your largest typical gallery, your yearly shoots, and how many photos you keep after culling. Then find your break-even and test a full year. Comparing per-image pricing vs subscription is quick, and it replaces a guess with a number you can put next to a pricing page.

For the break-even arithmetic behind this comparison, see our guide to break-even point calculation with your own numbers.

This article offers general information, not financial advice. All prices, photo counts, and results are invented examples and don’t describe any real product.

A renovation quote comparison looks simple until you put two quotes side by side and realize they’re not describing the same job: one lists permits and the other doesn’t, one bundles cabinet installation into a single price, and one is just a lump sum with a total at the bottom.

That’s why the lowest number on the page can be misleading. Quotes are written by different people, in different formats, with different assumptions about what’s included. Comparing the bottom lines is like comparing two grocery receipts when one of them leaves out the milk.

This guide shows how to line up contractor quotes item by item, how to let an AI chatbot do the tedious matching without judging which bidder is better, and how to turn the gaps into a short list of questions for each contractor. The contractors, prices, and project in the examples are invented. This is general information, not legal, financial, or construction advice, and rules for contractors vary by state and city.

Why Totals Mislead

When one quote says $28,500 and another says $32,900, the natural reaction is to lean toward the cheaper one. But a total only means something if you know what’s inside it. Differences can come from a few places:

• Scope. One quote includes debris removal, permits, or installation, and another leaves them out.
• Grouping. One lists “cabinets” as a single line, and another splits supply and installation.
• Allowances. A quote may list a placeholder amount for something you’ll choose later, such as tile, that might end up higher.
• Detail. A lump sum gives you nothing to compare against.

None of these makes a contractor dishonest. They make quotes hard to compare, and the solution is to put them in the same shape.

Why a Written, Itemized Quote Matters

California’s contractor licensing agency publishes a guide that lists what a home improvement contract should contain: contractor information, a description of the work, details on materials and equipment, pricing, a payment schedule, and a timeline. The guide is specific to California, and other states have their own rules, so check yours. But the logic applies anywhere. A quote that spells out the work and materials is easier to compare, and easier to hold someone to, than one that doesn’t.

Three Ways to Compare Bids

• Shows what’s missing: Compare the totals – No; Read each quote and compare by feel – Sometimes; Build one comparison table – Yes.
• Shows where prices differ: Compare the totals – No; Read each quote and compare by feel – Roughly; Build one comparison table – Line by line.
• Basis for follow-up questions: Compare the totals – Weak; Read each quote and compare by feel – Medium; Build one comparison table – Strong.
• Risk of comparing unlike things: Compare the totals – High; Read each quote and compare by feel – Medium; Build one comparison table – Lower.

Building the Comparison Table

Step One: Build the Master List of Line Items

Create a list of every kind of work your project involves. For a kitchen remodel, that might be demolition, cabinets, countertops, flooring, plumbing, electrical, painting, and permits. Start from the quote that has the most detail, and add any lines the others contain.

Then fill in what each contractor wrote. If a quote doesn’t mention an item, write “NOT LISTED,” not zero. A missing line isn’t a free line. It’s an unanswered question.

Step Two: Let AI Do the Matching

Paste the text of each quote into a chatbot and use rules that stop it from guessing:

I'm comparing contractor quotes for a renovation. Below is the text of each quote, copied exactly.

Quote A:
[paste]

Quote B:
[paste]

Quote C:
[paste]

Please:
1. Make a table with one row per line item and one column per quote. Use the amounts exactly as written.
2. If a quote does not mention an item that another quote lists, write "NOT LISTED" in that cell. Do not enter zero and do not estimate.
3. If a quote groups several items into one price, write the grouped price in the first matching row and "INCLUDED IN [row]" in the others.
4. Add a total row using only the amounts shown, and show the arithmetic.
5. For each row where the prices differ, show the difference in dollars.
6. List every item that appears in only some quotes, and for each one write a question I could ask the contractors who left it out.

Rules:
- Use only the pasted text.
- Do not say which quote is better, cheaper, fair, or recommended.
- Do not guess at prices for missing items.
- If a quote is a lump sum without a breakdown, say so and list what I should ask for.

I wrote this prompt myself as a suggestion; no source publishes it as a standard.

The “NOT LISTED” instruction is the heart of it. It keeps the AI from turning a blank into a number.

A Worked Example With Invented Quotes

Suppose three invented contractors quote a kitchen remodel.

Quote A lists demolition $2,500, cabinets (supply and install) $14,000, countertops $5,200, flooring $3,800, plumbing $1,900, electrical $2,300, and painting $1,200. It doesn’t mention permits or debris removal. Total: $30,900.

Quote B lists demolition including debris removal $3,100, cabinets (supply only) $11,500, cabinet installation $3,000, countertops $5,600, flooring $3,500, plumbing $2,100, electrical $2,000, painting $1,500, and permits $600. Total: $32,900.

Quote C says “Full kitchen remodel, $28,500” with no breakdown.

After normalizing, the table looks like this:

Item	Quote A	Quote B	B minus A
Demolition	$2,500	$3,100	+$600
Cabinets, supply and install	$14,000	$14,500	+$500
Countertops	$5,200	$5,600	+$400
Flooring	$3,800	$3,500	−$300
Plumbing	$1,900	$2,100	+$200
Electrical	$2,300	$2,000	−$300
Painting	$1,200	$1,500	+$300
Permits	NOT LISTED	$600	not comparable
Total	$30,900	$32,900	+$2,000

Check the arithmetic. On the lines both quotes list, the differences add up to 600 plus 500 plus 400, minus 300, plus 200, minus 300, plus 300, which is $1,400. Add the $600 permit line only Quote B includes, and you get the $2,000 gap. So of the $2,000 difference, $600 is a line Quote A may simply have left off, not a real price gap. That changes the conversation.

Quote C’s $28,500 is $2,400 below Quote A, but it tells you nothing yet. Without a breakdown, you can’t tell whether it includes permits, debris removal, or installation. All of these contractors and numbers are invented.

What Each Written Estimate Should Contain

The FTC Consumer Advice: How To Avoid a Home Improvement Scam advises getting multiple written estimates that each describe the work, the materials, the completion date and the price, and not automatically picking the lowest one: if estimates differ significantly, ask each contractor to explain why. That is exactly what the comparison table makes visible, because a gap in one row points to the question to ask.

Turning the Table Into Next Steps

Turn Gaps Into Questions

The most useful output of the comparison isn’t a winner. It’s a list of questions, such as:

• To Contractor A: “Does your price include permits? Debris removal? If not, what would they cost?”
• To Contractor B: “Is the cabinet price for a specific product line? What happens if I choose a different one?”
• To Contractor C: “Can you provide an itemized breakdown of the total, including what’s included for permits and cleanup?”

Send the same questions to each contractor in writing, so you have the answers on paper. Then update the table. A comparison improves with each answer.

What the Table Can’t Tell You

A table shows prices and gaps. It doesn’t show quality, reliability, schedule, or whether a contractor is properly licensed and insured for your area. For those, check your state’s or city’s licensing resources, ask for references, and read each contract carefully. Our guide to a renovation contract checklist covers what the contract itself should contain. The AI helps you organize information. It can’t vouch for a person.

Check Your Own Inputs

Quotes may include allowances, which are placeholder amounts for items you’ll select later. If tile is listed at an allowance of $600 and the tile you like costs more, the final price rises. In the table, add a column labeled “allowance?” and mark any line where the quote says so. Then ask the contractor what happens if your choice costs more or less than the allowance.

Keep the Comparison Alive After You Pick

Don’t throw the table away once you’ve chosen a contractor. The line items you matched become the baseline for the rest of the project. If the final contract has a line that wasn’t in the quote you compared, or a number that moved, you’ll see it right away. Save the table next to the signed contract, and note the date of each quote. A renovation quote comparison done carefully at the start also makes later conversations easier, because you can point to the exact line and the exact amount you both agreed on.

Where Comparisons Go Wrong

• Entering a blank as zero. Write “NOT LISTED” instead.
• Comparing totals before normalizing. Group lines the same way first.
• Asking the AI which quote to pick. Ask it to organize, and make the decision yourself.
• Ignoring allowances. They can move the final price.
• Skipping the written follow-up. Verbal answers disappear.
• Pasting personal details into a chatbot. Remove names, addresses, and phone numbers first.

Questions About Renovation Quote Comparison

What is a renovation quote comparison? It’s a side-by-side table of contractor quotes, matched line by line, that shows what each one includes and what it leaves out.

Can AI compare contractor quotes? It can match line items and calculate differences if you tell it to use only the pasted text and mark missing items. It shouldn’t choose a winner.

Why is one quote so much lower than the others? It may leave out items the others include, or it may be a lump sum without detail. Ask for an itemized breakdown before reading anything into the price.

How many quotes should I get? There’s no single right number; you need at least two to compare.

What if a quote is a lump sum? Ask the contractor for an itemized version. Without one, you can’t compare it to the others.

Should I pick the cheapest? Price is one factor. Scope, schedule, references, licensing, and the contract matter as well.

Is this legal advice? No. Contractor rules vary by state and city, so check your local requirements.

Build the Table Before You Choose

Gather your quotes, list every line item, and mark every blank “NOT LISTED.” Run the prompt, send your questions, and update the table. A renovation quote comparison turns a vague sense that one bid is cheaper into a specific list of what each price covers.

This article offers general information, not legal, financial, or construction advice. Licensing, contract, and permit rules vary by state and locality. The example contractors, quotes, and figures are fictional. Check requirements with your local authorities and consider professional advice for your own project.

Sources:

• California Contractors State License Board: Contracting for Success (California-specific guide)
• FTC Consumer Advice: How To Avoid a Home Improvement Scam

AI business plan numbers are the part of a plan you can’t hand off, because a chatbot will happily produce a revenue figure that looks professional, sounds confident, and has nothing underneath it.

That’s the quiet risk with AI plan writers. In my view, they’re good at structure and language. Ask for a financial section with no inputs, though, and a chatbot can fill the gap with plausible-looking assumptions: a growth rate here, a price there, a customer count that appeared from nowhere. The page looks complete. The first person who asks “where does this number come from?” is the one who finds out it doesn’t come from anywhere.

This guide shows a way to use AI for what it’s good at while keeping every number traceable to you. You’ll build a one-page assumptions sheet, hand it to the AI with strict rules, check the output against your own arithmetic, and test how fragile your plan is. The business in the examples is invented. This is general information, not financial, tax, or legal advice, and an accountant or advisor can help with your real situation.

Why Invented Numbers Do the Most Damage

A weak paragraph is easy to fix. A wrong number can get copied into other places: the summary, the funding request, the break-even section. By the time anyone notices, several pages may depend on it.

The U.S. Small Business Administration describes what a financial section is supposed to contain. For a traditional plan, it recommends a prospective outlook for the next five years, including forecasted income statements, balance sheets, cash flow statements, and capital expenditure budgets.

For a new business it suggests more detail in the first year, using quarterly or even monthly projections. It also says the funding request should explain how much you need over the next five years and what you’ll use it for. Every one of those items is a set of numbers you’re expected to understand and defend.

That’s a good test for any figure in your plan: could you explain it out loud, without looking at the tool that produced it?

Three Ways to Handle the Financial Section

• Do you know where numbers come from: Let the AI generate it – No; Do it all by hand – Yes; Your inputs, AI’s wording – Yes.
• Good at explaining the numbers in words: Let the AI generate it – Yes; Do it all by hand – Depends on you; Your inputs, AI’s wording – Yes.
• Risk of invented assumptions: Let the AI generate it – High; Do it all by hand – Low; Your inputs, AI’s wording – Low, if you set rules.
• Easy to update when something changes: Let the AI generate it – No; Do it all by hand – Yes; Your inputs, AI’s wording – Yes.

The third column is what this guide covers. You keep the inputs and the arithmetic. The AI turns them into readable sentences and flags what’s missing.

Build the Assumptions Sheet First

Before opening any chatbot, make a plain table with every number your plan depends on. Each row gets a value, a unit, and where it came from.

Assumption	Value	Source or reason
Selling price per unit	$24	My own planned price
Variable cost per unit	$9	Supplier quote I have in hand
Fixed costs per month	$1,200	Rent, insurance, software: list below
Units sold per month	100	My estimate, based on [explain]
One-time startup costs	$6,000	Equipment and first inventory

This invented example is a small candle business. Notice the last column. “My estimate” is honest. It tells you which numbers are guesses, and the guesses are the ones to test hardest later. If a row has no source, that’s a gap to fill before you write any prose.

The SBA’s startup-cost guidance adds useful structure. It separates one-time expenses from ongoing monthly ones, recommends counting at least a year of monthly expenses (five years is ideal), and suggests looking at what similar companies pay and talking with vendors and mentors. Use that to fill in the source column with something better than a hunch.

Give the AI the Sheet and a Short Leash

Once the sheet is ready, paste it into the chatbot with rules that prevent invention:

I'm writing the financial section of a business plan. Below is my assumptions table. These are the only numbers you may use.

[paste the table]

Please:
1. Calculate monthly revenue, monthly variable costs, gross profit, and monthly profit or loss from these inputs. Show every step of the arithmetic.
2. Calculate the annual figures by multiplying the monthly results by 12, and show that step.
3. Write one plain-English paragraph a lender could read that describes these results without adding any claims.
4. List every assumption you would need that is NOT in my table, marked "MISSING." For example, growth rates, seasonal changes, or taxes.

Rules:
- Use only the numbers in my table.
- Do not invent growth rates, market sizes, or customer counts.
- Do not say the numbers are realistic or good. Just calculate and describe.
- If something is unclear, ask me instead of guessing.

In my view, the most valuable part is item 4. It makes the AI list the holes instead of silently filling them.

A Worked Example With an Invented Business

Using the candle table above, the numbers work out like this:

• Monthly revenue: 100 units times $24 equals $2,400
• Monthly variable costs: 100 times $9 equals $900
• Gross profit: $2,400 minus $900 equals $1,500
• Monthly profit: $1,500 minus $1,200 fixed costs equals $300
• Annual profit: $300 times 12 equals $3,600

The AI should return exactly these. If it returns anything else, either it made an error or it used a number you didn’t give it. Check it against your own arithmetic, with a calculator, every time. It should also list items such as “no growth rate stated” and “no taxes stated” under MISSING, which tells you what to decide before the plan is done.

Now ask the question the plan has to survive: what if you sell fewer units?

Units per month	Revenue	Variable costs	Monthly profit
100	$2,400	$900	$300
80	$1,920	$720	$0
70	$1,680	$630	−$150

A drop from 100 to 80 units, a 20% shortfall, erases the profit completely. At 70, the business loses $150 a month. That’s what a sensitivity table reveals, and it’s the type of thing a reviewer may want to see you’ve thought about. All of these figures belong to an invented business, and yours will differ.

Let the AI Argue With You

Once your numbers are down, ask the AI to challenge the weak points rather than praise the plan:

Here are my assumptions and results. Without changing any numbers, list the three assumptions that, if wrong, would change the outcome the most, and explain why using only the numbers I gave. Do not say whether the plan is good or likely to succeed.

In the candle example, the answer should point at units per month and variable cost, since small changes there swing profit the most. If the AI names something else, ask it to show the arithmetic.

Sort Costs the Way the SBA Does

The SBA: Calculate your startup costs page says to sort expenses into one-time costs, such as major equipment, logo design and permits or licenses, and monthly costs such as salaries, rent and utilities, and to budget for at least one year of monthly expenses, ideally five. It lists common categories (office space, equipment and supplies, communications, utilities, licenses and permits, insurance, lawyer and accountant fees, inventory, salaries, advertising, market research, printed materials and a website), notes that costs with published fees like permits are straightforward while less certain ones like salaries need research, and suggests asking mentors, vendors and service providers what similar companies pay. Those categories make a good first column for the assumptions sheet, with a source in the next column.

Keeping Every Number Traceable

When you ask the AI to write the financial narrative, keep the connection between sentence and number. For each claim, you should be able to point to a row in your sheet. A simple method: after the AI writes a paragraph, ask it to put the row name in brackets after every figure. Then you can check them in a minute, and so can anyone who reads your working file.

If the draft contains a number that isn’t in your table, delete it or add it to the table with a source. Don’t leave an orphan figure in the plan.

Keep a Change Log

Numbers move. A supplier raises a price, a market stall costs more than you expected, a friend tells you what they’d really pay. Each time an input changes, update the assumptions sheet first, write the date next to the new value, and rerun the prompt. A simple change log, even three columns of date, what changed, and why, turns your AI business plan numbers into something you can defend months later. It can also stop two versions of the plan from circulating with different figures, an inconsistency a careful reader may notice.

Slip-Ups That Sink the Financial Section

• Asking for projections without inputs. The AI will invent them.
• Treating the AI’s arithmetic as verified. Recalculate everything by hand or in a spreadsheet.
• Leaving the source column blank. Unsourced numbers are the first thing to question.
• Showing only the best case. Include at least one lower-sales scenario.
• Forgetting the one-time costs. Keep them separate from monthly expenses, as the SBA’s guidance does.
• Pasting private financial records into a chatbot. Use rounded or illustrative inputs unless you’ve checked the tool’s privacy terms.

Questions About AI Business Plan Numbers

What are AI business plan numbers? They’re the financial figures in a plan, such as prices, costs, sales volumes, and profit, and the point of the term here is that you should supply and understand them, not let a chatbot invent them.

Can AI write the financial section of a business plan? It can describe numbers you provide and calculate from them if you show the steps, but it shouldn’t create the assumptions.

How do I stop AI from making up numbers? Give it a table of your inputs, tell it to use only those numbers, and require it to list anything missing.

Should I trust the AI’s arithmetic? No. Recalculate yourself. Chatbots can make arithmetic mistakes, especially in longer chains.

How far ahead should projections go? The SBA’s guidance for a traditional plan recommends a five-year outlook, with more detail in the first year. Check what your lender or reader expects.

Do I need an accountant? For anything involving taxes, loans, or investors, a qualified accountant or advisor can review your numbers. This article is general information only.

Start With the Sheet, Not the Chatbot

Open a blank table, list five assumptions, and write where each one came from. Then paste it in and run the prompt. Your AI business plan numbers may take longer to build than a generated draft, and you’ll be able to explain every one of them to anyone who asks.

This article offers general information, not financial, tax, legal, or investment advice. Requirements for plans vary by lender and situation. The example business, numbers, and results are fictional. Talk to a qualified accountant or advisor about your own plan.

Sources:

• U.S. Small Business Administration: Write your business plan
• U.S. Small Business Administration: Calculate your startup costs

AI journaling encryption is harder to judge than the word “encrypted” on an app’s landing page suggests, because the moment an AI reads your entry, the protection that covered it while it sat in storage may not cover it anymore.

That’s not a scandal, and it isn’t a reason to avoid every journaling app. It’s just a gap between how privacy gets advertised and how it works. A journal can be a place where you write things you’d never post, never text, and maybe never say out loud. So it’s worth knowing what the labels on these apps actually promise, and where an AI feature changes the picture.

This guide explains the common encryption terms in plain language, shows what one real app says about how its AI features interact with its encryption, and gives you a short method for reading any app’s own words without taking marketing at face value. The app in the worked example is invented. This is general information, not security or legal advice, and terms and features change, so always read the current documents yourself.

Why One Word Hides Four Different Promises

“Encrypted” can mean very different things. Here are the levels you’re most likely to run into, from weakest to strongest:

• Encrypted in transit. What it generally means: Data is scrambled while traveling between your phone and the server. Who could still read your entries: The company, once it arrives.
• Encrypted at rest (server-side). What it generally means: Data is scrambled while stored on the company’s servers. Who could still read your entries: The company, because it holds the keys.
• End-to-end encrypted (E2EE). What it generally means: Only your own devices hold the keys needed to read it. Who could still read your entries: In principle, only you.
• “Zero-knowledge” What it generally means: A term some companies use, often meant to say the company can’t read your content. Who could still read your entries: Depends on the app’s own definition. Check it.

The first two protect you against outsiders: someone snooping on the connection, or someone who steals a hard drive. They don’t protect you from the company itself, since the company can unlock the data when it needs to. Only end-to-end encryption is designed so that the company can’t read your content, and even then, the details matter.

“Zero-knowledge” isn’t a standard with one meaning. Some companies use it carefully, and some use it as a slogan. When you see it, look for the sentence that says what the company can and can’t see.

The Step Where Protection Can Pause

Here’s a part that is easy to miss. For an AI feature to respond to what you wrote, something has to read your words. If your journal is end-to-end encrypted, that means your words have to be unlocked somewhere before the AI can use them.

Day One, a well-known journaling app, spells this out in its own help pages. Its AI features guide says that for AI features to work with encrypted content, the content must first be temporarily decrypted on your device before being sent securely to the AI service for processing, and that content from end-to-end encrypted journals is temporarily not end-to-end encrypted while being processed with AI.

It also says some features run on your device using Apple Intelligence on supported devices, while others run through Automattic’s infrastructure or a third-party provider, currently OpenAI, and that it doesn’t keep the request on its servers once it’s processed and the AI services it uses don’t use your content for training unless you give permission.

Notice how useful that is. The company didn’t claim the AI features are “just as private.” It told you exactly what changes, who may process it, and what it says it does afterward. That’s what a clear answer looks like, and it gives you something to compare other apps against.

The guide is on Day One’s own site and may be updated, so read the current version before you rely on any of this. The point isn’t that this app is good or bad. It’s that the honest version of the answer has several moving parts, and a single label can’t carry all of them.

Where the AI Runs Matters More Than the Label

Think of each AI feature as having one of three homes:

• On your device. What happens to your text: Processed locally and not sent anywhere for that feature. What to check: Which features, which devices, which settings.
• The app company’s servers. What happens to your text: Sent to the company, processed, possibly logged. What to check: Retention, who can access, whether it’s used to improve products.
• A third-party AI provider. What happens to your text: Sent onward to another company. What to check: Which provider, what the provider may keep, and what the app promises about training.

An app can use more than one of these. In the Day One example above, different features use different homes. That’s why a blanket statement like “your journal is private” can be an incomplete story. A more useful question is “which of these features, and which of these homes?”

What Encryption Usually Doesn’t Cover

Even a strongly encrypted journal has edges. Two worth knowing:

• Metadata. Day One’s encryption document says entry text and life-record data are encrypted, but notes that some details such as timestamps, image dimensions, and device information are excluded. Other apps may have something similar. Metadata may be less sensitive than your words, but you should know it exists.
• Lost keys. The same document explains that if you lose the key to your encrypted data, the company can’t decrypt it for you. That’s the flip side of true end-to-end encryption. If an app promises it can always restore your journal after you forget your password, ask how, because the answer tells you who holds the keys.

Reading an App’s Own Words, Step by Step

You don’t need technical knowledge to judge AI journaling encryption. You need the app’s privacy policy, its security page, and any help article about AI features. Then look for the answers to five questions:

1. Can the company read my entries in normal use? Look for sentences about who holds the keys.
2. What happens to my entries when an AI feature is used? Look for “decrypted,” “sent to,” “processed by.”
3. Which companies are involved in processing? Look for named providers or the phrase “third party.”
4. What’s kept afterward? Look for retention and training language.
5. What isn’t encrypted? Look for metadata or exceptions.

If a page doesn’t answer a question, that’s information too. An unanswered question is a reason to ask the company or choose another app, not a reason to assume the best.

Let AI Do the Reading, With Strict Rules

Privacy policies are long and dull, and a chatbot can help you pull out the relevant lines, as long as it can’t fill gaps with guesses. Paste the app’s own text and use this:

I'm trying to understand how a journaling app handles privacy. Below is text copied from the app's privacy policy, security page, and AI features page. Please read only this text.

[paste the text]

For each question, quote the exact sentence that answers it. If the text doesn't answer it, write "NOT STATED."
1. Who holds the encryption keys for my entries?
2. Can the company read my entries during normal use?
3. What happens to my entry text when an AI feature is used?
4. Which third-party companies process my content?
5. How long is content kept after processing, and is it used for training?
6. What data is not encrypted (for example, metadata)?
7. What happens if I lose my password or key?

Rules:
- Use only the pasted text.
- Do not assume anything about how the app works.
- Do not say whether the app is safe or private. Just quote.
- If two sentences seem to conflict, quote both and write "CONFLICT."

I wrote this prompt myself as a suggestion; no source publishes it as a standard.

The “do not say it’s private” rule is the important one. You want evidence, not a verdict.

Worked Example for AI Journaling Encryption

Imagine a journaling app called Notebird, which doesn’t exist. Its marketing page says “Your journal is encrypted and private.” You paste its security page and AI help page into the prompt, and the AI returns:

• Q1 (keys): “NOT STATED.”
• Q2 (company can read): “NOT STATED.”
• Q3 (AI feature): “Entries are sent to our AI partner to generate reflections.”
• Q4 (third parties): “NOT STATED,” since the partner isn’t named.
• Q5 (retention): “NOT STATED.”
• Q6 (not encrypted): “NOT STATED.”
• Q7 (lost password): “You can reset your password by email.”

That’s a lot of blanks for an app promising privacy. The reset-by-email line doesn’t prove anything by itself, but it’s worth asking how a reset works if only you hold the keys. And the only AI sentence confirms your text leaves your phone. Now you have a short, specific list of questions to send to the company. If it can’t answer clearly, that’s your answer. The app, its text, and the responses here are all invented.

Two Settings to Check Outside the App

The FTC: How Websites and Apps Collect and Use Your Information article points to your phone’s own privacy settings as a second layer. It says to check which apps can access your location, contacts or photos, to consider turning off permissions an app doesn’t need to work, and to consider deleting apps that request a lot of permissions they don’t need. It also notes that advertisers can use a unique advertising identifier to track you across apps, and that you can opt out of personalized ads and reset that identifier in the advertising section of your phone’s privacy settings. For a journaling app, those settings sit outside the app’s encryption claims, so they are worth a minute.

Matching Your Choice to What You Write

Good AI journaling encryption isn’t the same need for everyone. Not every journal needs maximum protection, and not every person writes the same things. A gratitude list is different from a diary about a family conflict. Here’s a way to think about it:

• Everyday notes and reflections: A cloud AI feature may be a fair trade for the convenience, as long as you’ve read what it does.
• Very private material: Consider keeping those entries out of AI features entirely. Some apps let you exclude entries, so check whether yours does and how the exclusion works.
• Material you’d never want read by anyone: Consider a journal with end-to-end encryption and no AI processing, or paper.

You can also mix. Nothing says all your writing has to go in one place.

Slip-Ups That Undo the Whole Check

• Reading only the marketing page. The honest details live in the help and security pages.
• Assuming E2EE covers the AI features. Check, because an AI step may need to unlock your text.
• Treating “zero-knowledge” as a standard. Look for what the company says it can and can’t see.
• Skipping the lost-key question. It reveals who really holds the keys.
• Assuming settings don’t change. Apps add AI features over time. Recheck after major updates.

Questions About AI Journaling Encryption

What does AI journaling encryption actually protect? It depends on the type. Server-side encryption protects stored data from outsiders, end-to-end encryption is designed so only your devices can read it, and either may behave differently when an AI feature processes your text.

Is an end-to-end encrypted journal always private when I use AI features? Not necessarily. One major app’s own documentation says content is temporarily not end-to-end encrypted while being processed with AI.

Does on-device AI avoid the problem? For the features that run on your device, your text isn’t sent elsewhere for that step. Check which features run locally and on which devices.

What does zero-knowledge mean? There’s no single definition. Look for the app’s own statement of what it can and can’t see.

Can I use AI to check an app’s privacy policy? Yes, if you paste the real text and require it to quote and label missing answers. Don’t let it judge safety.

Should I stop using cloud-based AI journaling? That’s your call. The goal is to know what you’re choosing, and to keep your most sensitive writing in a place that matches your comfort level.

Pick One App and Run the Five Questions

Choose the journaling app you use or are considering, open its security and AI pages, and answer the five questions with quotes. Understanding AI journaling encryption doesn’t require technical knowledge, and it replaces a vague feeling about “private” with specific sentences you can point to.

If tracking apps make you anxious about your own data, our sleep tracker anxiety reset offers a calmer routine.

This article offers general information, not legal, security, or mental-health advice. Features, terms, and policies change, so read each app’s current documents. The example app, text, and results are fictional. A journal can help with everyday reflection, but if you’re struggling with persistent or serious emotional difficulty, consider talking with a licensed professional.

Sources:

• Day One: AI Features
• Day One: End-to-End Encryption (PDF)
• FTC: How Websites and Apps Collect and Use Your Information

A weekly team progress report built from your task list tells you what actually got finished and what has gone quiet, using the dates and statuses already in your project tool instead of a manager’s impression of who “seems busy.”

On a remote team, you can’t see anyone’s desk. That makes it tempting to judge progress by how full the calendar looks, how fast people reply in chat, or how many meetings someone attends. None of those directly measure what got done. Your task list may hold better evidence: what moved to done, what hasn’t been touched in days, and what’s blocked. The trouble is that nobody has time to read it all every Friday.

This guide shows how to turn a plain export of your tasks into a short weekly report, how to define “shipped” and “stalled” in your own terms, an AI prompt that summarizes and counts without inventing anything, and how to share the report without turning it into a ranking of people. The team and tasks in the examples are invented. Nothing here is legal or HR advice.

Why Outcomes Beat Impressions

In my view, an impression-based review has three problems. It rewards visibility over results, it depends on who speaks up in meetings, and it can’t be checked. A report based on task records can be checked by anyone on the team. When someone says “that was shipped Tuesday,” the record either shows it or it doesn’t.

That doesn’t mean task counts tell the whole story. Tasks differ in size, some work never becomes a task, and a quiet week can be a good week if someone was solving a hard problem. The report is a conversation starter that points you to places to ask questions, not a verdict.

Three Ways to Check on Progress

• Based on records: Go by impressions – No; Read the whole board – Yes; Weekly report from an export – Yes.
• Spots stalled work: Go by impressions – By luck; Read the whole board – If you notice; Weekly report from an export – By date.
• Can be checked by the team: Go by impressions – No; Read the whole board – Yes; Weekly report from an export – Yes.
• Risk of misreading: Go by impressions – High; Read the whole board – Medium; Weekly report from an export – Medium, if you add context.

Building the Report

Step One: Export the Fields That Matter

Most project tools can export tasks as a spreadsheet. You need only a few columns:

• Task name: What the work is
• Owner: Who’s responsible
• Status: Not started, in progress, blocked, done
• Created date: How old the task is
• Completed date: When it finished, if it did
• Last updated date: When anyone last touched it
• Blocker note: Why it’s stuck, if recorded

If your tool doesn’t track “last updated,” you can still report on shipped work, and you’ll need another way to spot stalled items, such as a standing weekly check-in.

Step Two: Define Your Terms in Writing

Words like “shipped” and “stalled” mean different things on different teams. Decide yours before you run anything:

• Shipped: a task whose status changed to done during the reporting week
• Stalled: a task in progress or blocked, with no update in more than a set number of days
• Aging: a task not started that was created more than a set number of days ago

Pick the number of days yourself, based on how long your team’s tasks normally go without an update, and write the numbers on the page so everyone shares the definition. The example below uses five.

Step Three: Give AI the Date and the Rules

AI can’t know today’s date unless you tell it, and it will guess if you don’t. Supply it. Here’s a prompt:

I manage a remote team. Below is an export of our tasks and today's date. Please produce a weekly progress report.

Today's date: [date]
Reporting week: [start date] to [end date]

My definitions:
- "Shipped" = status is Done and the completed date falls in the reporting week.
- "Stalled" = status is In progress or Blocked and the last updated date is more than [N] days before today.
- "Aging" = status is Not started and the created date is more than [M] days before today.

Task export:
[paste the table]

Please:
1. List shipped tasks with owner and completed date.
2. List stalled tasks with owner, last updated date, days since update (show the date subtraction), and the blocker note if there is one. If no blocker is recorded, write "NOT STATED."
3. List aging tasks with owner and days since creation (show the subtraction).
4. Give counts for each status as plain numbers.
5. List any task where the data seems inconsistent (for example, Done with no completed date), marked "CHECK."

Rules:
- Use only the data I pasted and my definitions.
- Do not guess reasons for delays, and do not judge any person's performance.
- Do not rank people or compare their output.
- Do not add tasks or details that are not in the export.

The rule against ranking people is worth keeping. In my view, task counts are a poor basis for comparing individuals, and a report that invites that comparison may lead people to game the numbers.

Worked Example for Weekly Team Progress Reports

Suppose today is March 14 (a non-leap year) and your five-day threshold applies. The invented export has eight tasks:

Task	Owner	Status	Date detail
Update pricing page	Ana	Done	Completed Mar 10
Fix signup bug	Ben	Done	Completed Mar 12
Draft Q2 plan	Cleo	In progress	Last updated Mar 6
Vendor contract review	Dev	In progress	Last updated Mar 13
Onboarding doc	Ana	Not started	Created Feb 20
Customer survey results	Ben	Blocked	Last updated Mar 4, “waiting on data export”
Newsletter draft	Cleo	Done	Completed Mar 11
Help-center update	Dev	In progress	Last updated Mar 3

The report would show:

• Shipped (3): pricing page, signup bug, newsletter draft.
• Stalled (3): Draft Q2 plan (8 days since update), Customer survey results (10 days, blocker: waiting on data export), Help-center update (11 days, blocker NOT STATED).
• Not stalled: Vendor contract review (1 day).
• Aging (1): Onboarding doc, 22 days since creation.
• Counts: 3 done, 3 in progress, 1 blocked, 1 not started.

The date math works out: March 6 to March 14 is 8 days, and February 20 to March 14 is 22 days in a non-leap year. Read the stalled list as questions: Is Cleo’s Q2 plan actually stuck, or deep in thinking? Who can unblock the data export for Ben? Why does the help-center update have no blocker note? All of the tasks and people here are invented.

Add the Context AI Can’t See

The report is a starting point. Before sharing, add notes only you can supply:

• Work not in the tool. A long conversation with a customer or a hard design problem may not appear as a task.
• Task size. A big task done and three small ones done aren’t equal.
• Reasons for delay that you already know.
• What you’re asking for. For each stalled item, is there a decision, a person, or a resource needed?

In my view, a good practice is to ask each owner about their stalled items before the report goes out, so nobody is surprised by seeing their name in a list.

Sharing the Report Well

Share It as a Team Document

How you share the report matters as much as what’s in it.

• Frame it around work, not people. Group by project or goal when possible.
• Share the definitions so the numbers aren’t mysterious.
• Invite corrections. If a task is wrongly marked stalled, update the record.
• Don’t use it to compare individuals. Different roles produce different kinds of tasks.
• Keep it short. One page that people actually read beats five that they don’t.

If you’re thinking about adding any tracking beyond task records, explain what it measures and why before you start, and check the rules that apply where you and your team work. This article doesn’t cover legal requirements.

Keep Sensitive Data Out

Task names can include customer names or internal details. Before pasting an export into an AI tool, remove anything confidential, check your company’s rules for using AI tools, and strip personal data that doesn’t belong. If you can’t remove it, summarize the tasks by hand.

A Weekly Rhythm That Works

1. Friday morning: export tasks.
2. Run the prompt and check the arithmetic.
3. Ask owners about stalled items (a few short messages).
4. Add context and send a one-page report.
5. Review last week’s stalled items to see what moved.

After a few weeks, you can see patterns, such as a type of task that always stalls or an approval step that causes delays. Those patterns point to fixes in the process, not in people.

Weekly Team Progress Reports: Mistakes to Avoid

• Treating counts as performance. Task size and type vary.
• Skipping the definitions. “Stalled” means nothing without a number of days.
• Letting AI guess why work is late. Ask people.
• Forgetting today’s date. Without it, any “days since” figure is invalid.
• Sharing a report that names people without asking them first. Talk before you publish.

Questions About Weekly Team Progress Reports

What should a weekly team progress report include? What shipped, what’s stalled and for how long, what’s blocked and why, and the questions that need answers. Keep it short and based on task records.

Can AI write a weekly team progress report? It can summarize an export and calculate dates if you give it today’s date and your definitions, and tell it not to guess reasons or judge people.

Is counting completed tasks a fair way to measure performance? Not by itself. Tasks differ in size and some work isn’t tracked. Use counts to start conversations, not to rank people.

How do I decide what counts as stalled? Choose a number of days that fits how your team works, and write it down so everyone shares the definition.

What if my tool doesn’t export last-updated dates? Report on shipped work from completion dates, and use a weekly check-in to find stalled items.

Should I share the report with the whole team? Possibly yes, framed around work and with definitions included. Ask owners about their items first.

How do I keep confidential information out of the AI tool? Remove names and details you shouldn’t share, follow your company’s AI policy, or summarize by hand.

Run Last Week’s Tasks Through It

Export last week’s task list, write your definitions for shipped and stalled, add today’s date, and run the prompt. A weekly team progress report built this way is a short routine and replaces guesses about who seems busy with a short list of what finished, what stopped, and who might need help.

If your report starts from meetings, our guide to meeting follow-up without recording shows a voice-memo method.

This article offers general management tips, not legal or HR advice. Workplace monitoring and data-handling rules vary by location. The example team, tasks, and dates are fictional.

To test an AI workout app, build the same profile twice, change only one detail, such as a limitation or the equipment you have, and compare the two plans exercise by exercise to see whether the app actually responds.

A plan labeled “personalized” can still be one template with a few numbers swapped. If you have a sore shoulder, a bad knee, or only a pair of dumbbells at home, it matters whether the app really changes what it gives you or just acknowledges what you typed.

This guide shows a simple before-and-after method, a way to score the results, an AI prompt that compares two plans without judging safety, and what the test can and can’t tell you. The apps, plans, and numbers in the examples are invented. This is general information, not medical advice. If you have an injury or health condition, talk to a qualified professional about what exercise is right for you.

What the Test Measures, and What It Doesn’t

The test measures one thing: whether changing an input changes the output. If you flag a limitation and the plan stays the same, the app isn’t using that input. That’s useful to know.

It doesn’t measure whether any plan is safe. A plan that changes when you flag a shoulder issue hasn’t been shown to be good for your shoulder. It has only shown that the app responds. Safety is a separate question, covered in our guide to an AI workout safety check.

Public health guidance supports the cautious attitude. The U.S. Physical Activity Guidelines say people should choose activities appropriate for their current fitness level and health goals, and should be under the care of a health care provider if they have chronic conditions or symptoms. An app test is not a substitute for that.

Three Ways to Judge an App

• Based on: Trust the description – Marketing; Read reviews – Other people’s accounts; Run the before-and-after test – Your own inputs.
• Shows if your input matters: Trust the description – No; Read reviews – Sometimes; Run the before-and-after test – Yes.
• Free to try: Trust the description – Yes; Read reviews – Yes; Run the before-and-after test – Often during a trial.
• Risk of being misled: Trust the description – High; Read reviews – Medium; Run the before-and-after test – Lower.

Reviews describe someone else’s account and version. The test shows what the app does for yours, today.

Before You Start: Privacy and Safety Notes

• You don’t need to share a real medical history. If you’d rather not enter personal health details into an app, use a different test variable, like equipment, as covered below.
• Don’t enter more than the app needs. Check what it asks for and what it does with the data.
• Use a free trial or free tier if one is available, so you can test before paying.
• If you have a real injury, the plan the app gives you is not medical guidance. Ask a qualified professional before following any exercise program.

Running the Before-and-After Test

Step One: Build Profile A

Create a profile as you normally would, with a real limitation or a clear equipment constraint. For example, an invented tester might enter:

• Age 32
• Goal: general strength
• Equipment: dumbbells only
• Limitation: sore shoulder (flagged in the app’s injury field)
• Three workout days per week

Generate the plan. Save a screenshot or copy the exercise list with sets and reps.

Step Two: Build Profile B, Changing One Thing

Go back and change only one detail. Two good options:

• Remove the limitation, leaving everything else identical
• Change the equipment, such as dumbbells only to a full gym with barbells and cables

Generate a second plan and save it. Changing more than one variable muddles the result, because you won’t know which change caused which difference.

Step Three: Compare Exercise by Exercise

Line the two plans up in a table: the exercises, then sets, reps, and notes.

Exercise	Plan A (shoulder flagged)	Plan B (no limitation)
Goblet squat	Yes	Yes
Overhead press	No	Yes
Dumbbell row	Yes	Yes
Hip hinge	Yes	Yes
Bench press	No	Yes
Band pull-apart	Yes	No
Glute bridge	Yes	No
Plank	Yes	Yes

This table is invented. Here, four exercises appear in both plans, and the pushing and overhead movements swapped out when the shoulder was flagged, with pulling and glute work added. That’s an app responding to the input.

Score the Result

A simple rubric:

• Plans nearly identical, only a generic line like “adjust as needed” added: The app acknowledged your input but didn’t act on it
• A few exercises changed, volume the same: Some response, limited
• Clear substitutions in the areas you flagged: The app responds to this input
• Big changes everywhere: Check whether the changes make sense, or whether it’s random

If the plans are identical, you’ve learned something valuable: don’t rely on that app to handle the limitation you entered. The U.S. guidelines say people with chronic conditions or symptoms should be under a health care provider’s care, and an unchanged plan is another reason to ask a professional rather than trust the app.

Let AI Compare the Plans, Without Judging Them

A chatbot can do the tedious comparison, but it must not decide whether either plan is safe. Use this prompt:

I tested a workout app by generating two plans with one change in my profile. Below are both plans, copied exactly. Please compare them.

What I changed between Plan A and Plan B: [for example, "removed the shoulder limitation" or "changed equipment from dumbbells only to full gym"]

Plan A:
[paste]

Plan B:
[paste]

Please:
1. Make a table listing every exercise and whether it appears in Plan A, Plan B, or both.
2. List the exercises that were added, removed, or changed, and any changes in sets, reps, or rest.
3. Count how many exercises are the same in both plans and show the math as a fraction.
4. State, using only the plans, whether the difference between the two plans matches the one thing I changed. Quote the lines that support your answer.
5. List anything in either plan that the app did not explain, marked "NOT STATED."

Rules:
- Use only the text I pasted.
- Do not say whether any exercise is safe or appropriate for me or for an injury.
- Do not give medical advice or recommend what I should do.
- Do not assume anything about the app's methods.

This prompt is my suggestion, not an official template.

The “do not say whether any exercise is safe” rule matters. A chatbot can’t evaluate your body, and a confident-sounding opinion could mislead you.

A Worked Example With Invented Plans

Suppose the AI returns, for the plans above: eight distinct exercises in total, four shared by both plans. The shared fraction is 4 of 6 per plan, or about 67%. The differences: Plan B includes overhead press and bench press, while Plan A includes band pull-apart and glute bridge instead. It reports that this matches the single change, flagging the shoulder, and quotes the lines. It also notes the app gave no reason for the substitutions (NOT STATED).

That’s an app that responds. Now imagine the other outcome: six of six exercises identical, with one added sentence at the bottom saying “consult a professional.” The AI would report the plans are identical except for that sentence. That result tells you the input didn’t change the plan, and the added line isn’t the same as an adjustment. Both scenarios are made up, and your results will vary.

Run the Test a Second Way

If you’d rather not enter an injury, use equipment as your variable. Generate one plan with dumbbells only and another with a full gym. If the exercise list barely changes, the app is accepting information without acting on it. Equipment is also easier to judge, since the difference is clear in the exercises themselves.

After the Test

Repeat Occasionally

Apps update. A result from months ago may not hold today, so repeat the test when you notice a big update, change your goals, or switch devices or accounts. Note the date and app version, if shown, with your saved plans.

What to Do With the Results

• The app responds. You can use it for general programming, while keeping your own judgment and professional advice for anything involving a real limitation.
• The app doesn’t respond. Don’t rely on it for that limitation. Consider another app, or ask a qualified professional to build or review your plan.
• You’re unsure. Ask a professional to look at both plans.

No test result replaces advice tailored to your body.

If you only remember one thing, remember that you can test an AI workout app with a single changed input, and the result tells you more than any rating.

AI Workout App Testing: Mistakes to Avoid

• Changing several inputs at once. You won’t know what mattered.
• Reading a changed plan as a safe plan. Responding isn’t the same as being appropriate.
• Skipping the screenshot. Save both plans, since apps may regenerate differently.
• Entering private health details you’re not comfortable sharing. Use equipment instead.
• Letting AI judge safety. Use it for comparison only.

Questions About AI Workout App Testing

How do I test an AI workout app? Create two profiles that differ in one detail, such as a limitation or equipment, generate a plan for each, and compare the exercises side by side.

How long does it take to test an AI workout app this way? The work is generating two plans, saving them, and comparing them line by line; a chatbot can build the comparison table for you.

Can AI help me compare two workout plans? Yes, if you tell it to use only the text you paste and not to judge safety or give medical advice.

What if both plans come out identical? It suggests the app doesn’t act on that input. Don’t rely on it for that limitation, and ask a qualified professional for guidance.

Does a plan that changes mean the app is safe? No. It shows the app responds to the input, not that the plan suits your body.

Can I run the test during a free trial? Often yes, if the app offers one. Check the app’s terms and refund policy.

Do I have to enter a real injury? No. Equipment makes a good test variable and avoids sharing health details.

How often should I repeat the test? After major updates or when your situation changes.

Run Your First Test Tonight

Pick one app, save one plan, change one detail, and compare. When you test an AI workout app this way, you replace a marketing promise with something you observed yourself, which is a far better basis for deciding how much to rely on it.

This article offers general information, not medical or legal advice. Talk to a qualified health professional before starting or changing an exercise program, especially if you have an injury, symptoms, or a chronic condition. Example apps, plans, and results are fictional.

Sources:

• U.S. Department of Health and Human Services: Physical Activity Guidelines for Americans, 2nd edition
• CDC: Physical activity for adults

Primary vs secondary market research comes down to one choice: do you look up what others have already collected, or do you ask your own customers directly, and the right answer depends on the question you’re trying to settle.

Small teams often waste effort by picking the wrong kind. They interview twenty people to learn something a public statistics page would have told them. Or they lean on a general report to decide something only their own customers can answer, like whether they’d pay for a specific product at a specific price. AI can help with both, but only if you know which kind of research you’re doing.

This guide gives you a simple way to sort your questions, a table for deciding between the two, an AI prompt that classifies questions without making up facts, and a plan for combining both cheaply. The sample business and questions are invented. The information is general and not financial advice.

What the SBA Says

The U.S. Small Business Administration’s guide to market research and competitive analysis draws the line clearly. Of existing sources, it says they “can save you a lot of time and energy, but the information might not be as specific to your audience as you’d like.”

Of asking consumers yourself, it says it “can give you a nuanced understanding of your specific target audience,” but “direct research can be time consuming and expensive.” It lists surveys, questionnaires, focus groups, and interviews as direct methods, and points to federal sources for existing data, including the Census Bureau, the Bureau of Labor Statistics, the Federal Reserve, and the Bureau of Economic Analysis.

That’s the whole trade-off in two sentences: existing sources are cheap and broad, direct research is specific and costly. The skill is matching each question to the right side.

Three Approaches Compared

• Cost: Only look things up – Low; Only ask customers – Higher; Sort questions, then choose – Moderate.
• Specific to your audience: Only look things up – Rarely; Only ask customers – Yes; Sort questions, then choose – Where it matters.
• Risk of wasted effort: Only look things up – Medium; Only ask customers – Medium to high; Sort questions, then choose – Lower.
• Answers “how big” or “who” questions: Only look things up – Often; Only ask customers – Slowly; Sort questions, then choose – Yes.
• Answers “would they buy” questions: Only look things up – No; Only ask customers – Yes; Sort questions, then choose – Yes.

The third path takes a bit of planning and saves a lot of effort.

Which Questions Belong Where

Here is my own rough sorting rule, not a published standard: broad questions about a population or an industry usually have existing answers. Narrow questions about your own offer usually need direct research.

• Who lives in my area, and how many – Existing sources: How many households are in our town?
• What do people in this industry earn or spend – Existing sources: What do local bakeries typically employ in staff?
• Who else sells something similar – Existing sources (public info): Which bakeries sell gluten-free bread nearby?
• Would my specific customers want this – Direct research: Would regulars buy a weekly bread subscription?
• What would they pay – Direct research: Would $X per week feel fair?
• Why do they choose or leave – Direct research: Why do some customers stop coming?

The line isn’t always sharp. A direct question can be sharpened by existing data, and existing data can raise new direct questions.

Sorting Your Questions in Two Steps

Step One: Write Your Real Questions

Start by writing the decisions you’re facing and the questions behind them. For example: “Should I add a bread subscription?” breaks into:

1. How many households could realistically be reached?
2. Are other bakeries offering subscriptions?
3. Would our regulars sign up?
4. What price would they accept?
5. What would make them cancel?

Keep each question specific enough that an answer would change what you do.

Step Two: Let AI Sort the Questions

AI can help you classify questions and plan the work. It must not supply numbers, because it can’t know your local market and it can invent plausible-sounding figures. A prompt that keeps it in bounds:

I'm deciding: [your decision]. Below are my research questions.

[paste your questions, numbered]

For each question, tell me:
1. Whether it is best answered by existing sources (secondary research) or by asking my own customers (primary research), and why in one sentence.
2. For existing-source questions: what TYPE of source I should look for (for example, "Census Bureau," "industry association," "competitor websites"). Do not name statistics or give numbers.
3. For direct-research questions: what method fits (survey, short interview, etc.) and what I would ask.
4. If I'm not sure, mark it "NOT SURE" and explain what extra information would settle it.

Rules:
- Do not provide any statistics, market sizes, or percentages.
- Do not claim that a specific report or dataset exists unless I mention it.
- Do not tell me what the answers will be.

The prompt above is my own wording, not an official or published template.

The rule against statistics matters most. A model asked for “the market size of local bakeries” may produce a confident number from nowhere. Asking only for the type of source keeps you responsible for finding and checking real data.

A Worked Example With an Invented Bakery

The invented bakery asks five questions. The AI’s classification might look like this:

#	Question	Type	Source or method
1	Households in our town	Existing	Census Bureau or local data
2	Other bakeries with subscriptions	Existing	Competitor websites and public pages
3	Would regulars sign up	Primary	Short survey at the counter
4	Acceptable price	Primary	Survey with a few price options
5	Why customers cancel	Primary	Three or four short conversations

Now the plan is concrete. Two questions can be answered from your desk this afternoon. Three need you to talk to real customers. And you didn’t pay for a report that answers none of them specifically. The bakery and the questions are made up, but the sorting works for any small business.

Free Federal Sources the SBA Lists

The SBA: Market research and competitive analysis page points to free government data by category: business and industry (NAICS codes, Census Business Builder, Statistics of U.S. Businesses), demographics (Census Bureau and the Bureau of Labor Statistics), economic indicators and income (CPI, Bureau of Economic Analysis, BLS earnings data), and consumer data (Federal Reserve consumer credit data). It says secondary sources work best for general, quantifiable data such as industry trends, demographics and household incomes, while asking consumers yourself gives a more nuanced picture of your specific audience but can be time-consuming and expensive, using surveys, questionnaires, focus groups or in-depth interviews. For customers, it lists demand, market size, income and employment, location, market saturation and what customers pay for alternatives as the questions to answer.

Doing the Research: Existing Sources First, Then Customers

Do the Existing-Source Work First

Start with what’s already public, since it’s cheap and can sharpen your direct questions. Free sources in the U.S. include those the SBA names: the Census Bureau, the Bureau of Labor Statistics, the Federal Reserve, and the Bureau of Economic Analysis. Industry associations and local government pages can add detail.

When you find a number, record:

• The figure and what it measures in the source’s own words
• The source and link
• The date of the data
• Any limits, such as geography or definitions that don’t match yours

If you’re using AI to summarize a long report, ask it to quote the exact sentences with page numbers, and check them yourself. Our guide to reading Google Trends data covers another existing source with its own quirks.

Then Ask Your Own Customers

Direct research doesn’t need to be fancy. Options the SBA lists include surveys, questionnaires, focus groups, and interviews. For a small business, the cheapest are:

• A three-question survey on paper or a free online form
• Short conversations with five to ten customers
• A simple test, such as offering a trial and seeing who signs up

AI can help you write clear, neutral questions. Ask it to avoid leading wording, such as “Wouldn’t you love a subscription?” Replace it with “How would you feel about a weekly bread subscription?” with answer choices. Then read the results yourself and don’t ask AI to tell you what customers “really” want from a handful of answers.

Be Honest About Small Samples

Five conversations tell you what five people think. That’s useful for finding ideas and problems, and it isn’t a measurement of a whole market. Write down how many people you asked, how you found them, and who’s missing. If the decision is expensive, you may want a larger or more carefully chosen sample, or advice from a research professional.

When the Decision Is Big

The weight of the decision should set how much of each kind of research you do. A small, reversible choice, like trying a new flavor for a month, can rest on a quick look at existing sources and a handful of conversations.

A costly or hard-to-undo choice, like signing a lease or buying equipment, deserves more: better sources, a larger and more carefully chosen group of customers, and possibly advice from someone who does research for a living. Writing down how much you could lose if you’re wrong helps you decide how much effort the research deserves. In every case, the primary vs secondary market research split stays useful as a checklist of what you’ve covered and what you haven’t.

Primary vs Secondary Market Research: Mistakes to Avoid

• Surveying for facts that are already public. Look up first.
• Using broad reports to answer narrow questions. A national average won’t tell you what your regulars would pay.
• Asking leading questions. They produce answers you wanted to hear.
• Letting AI supply numbers. Find them in real sources.
• Overgeneralizing from a few answers. Describe the sample honestly.

Questions About Primary vs Secondary Market Research

What’s the difference between primary and secondary market research? Secondary research uses existing information collected by others. Primary research is information you collect yourself, directly from customers or the market.

How do I decide between primary vs secondary market research? Sort each question: broad questions about a population or industry usually have existing answers, and narrow questions about your own offer usually need direct research.

Which is better for a small business? Neither alone. Use existing sources for broad questions and direct research for questions about your specific customers and offer.

Can AI do primary vs secondary market research for me? It can help you sort your questions, plan a survey, and draft neutral wording. It can’t collect real customer answers or supply trustworthy statistics.

Is secondary research always cheaper? Usually, but it may not match your audience, as the SBA notes. Direct research can cost more in time and money.

How many customers do I need to ask? It depends on the decision and the precision you need. For early ideas, a few conversations can reveal problems. For measuring something, you need a larger, carefully chosen group.

Where can I find free secondary data? The SBA names U.S. federal sources such as the Census Bureau, the Bureau of Labor Statistics, the Federal Reserve, and the Bureau of Economic Analysis.

Should I trust a number an AI gives me? Not on its own. Find the original source, check the date and definitions, and confirm it applies to your situation.

Sort Your Questions This Week

Write down the decision you’re facing, list the five questions behind it, and mark each one “look up” or “ask.” Run the prompt if it helps, and start with the look-ups. Understanding primary vs secondary market research saves you from paying, in time or money, for answers that were free, and from guessing about answers only your customers can give.

This article offers general business information, not financial or professional advice. The example bakery and questions are fictional.

Sources:

• U.S. Small Business Administration: Market research and competitive analysis

A customer email answer bank is a short document of approved answers to your most common questions, and when AI drafts replies only from that document, you answer faster without letting a chatbot invent your policies.

A small business can answer the same questions again and again: Where’s my order? Can I return this? Do you ship abroad? Typing each reply from scratch wastes time. Asking AI to “reply to this customer” is faster, but it has a catch. The tool doesn’t know your return window, your shipping rules, or your hours. It will write something that sounds confident, and sometimes it will be wrong in a way that costs you a refund or a customer.

This guide shows how to build a customer email answer bank in a few steps, how to write entries that are safe to reuse, how to prompt AI so it uses only those entries and says so when it can’t, and how to keep the bank current. The sample business and emails are invented. Nothing here is legal advice, and your actual policies are yours to set.

Why “Only Use My Answers” Matters

The U.S. Consumer Financial Protection Bureau’s report on chatbots in consumer finance was written about banks, not small shops, but one of its warnings applies broadly. It notes that complex chatbots built on large language models “may rely on datasets that include instances of misinformation or disinformation that are then repeated.” In plain words: a general-purpose AI draws on a huge pile of text, not on your business, and it can repeat things that are false.

You can’t fix that by asking nicely. You can reduce it by giving the AI a short, approved source and telling it to stay inside that source. That’s what the answer bank is for.

Three Ways to Answer Repeat Questions

• Accuracy of policies: Type each reply – Depends on memory; Ask AI with no source – Unreliable; Answer bank + AI drafts – Matches your approved text.
• Consistency across replies: Type each reply – Varies; Ask AI with no source – Varies; Answer bank + AI drafts – High.
• Risk of invented promises: Type each reply – Low; Ask AI with no source – High; Answer bank + AI drafts – Low.

In my view, the third option takes setup, and it’s the one that keeps working as your inbox grows.

Building the Bank, Step by Step

Step One: Find the Questions You Really Get

Don’t guess at your common questions. Look at your inbox. Take your last 30 to 50 customer emails, remove names, addresses, and order numbers, and sort them into piles by topic. A small candle shop (invented, called “Maple Street Candles”) might find:

Topic	Count in 40 emails
Where is my order	12
Return or exchange	9
Scent or product questions	7
Wholesale inquiries	4
Damaged item	4
Other	4

The numbers are invented. Yours will differ. The point is that three or four topics may cover most of your mail, and those are the entries to write first.

Step Two: Write Each Answer Yourself

This step belongs to you. AI can help polish wording later, but the facts must come from you, because you know your actual rules. For each entry, fill in:

• Topic: A short label, like “Return window”
• Typical customer question: The question in the customer’s own words
• Approved answer: Two to five sentences stating your rule exactly
• Details that must be exact: Numbers and conditions, like days, fees, and exclusions
• Do not promise: Things AI or a helper must never offer on their own
• Hand off to a person when: Situations that need your judgment
• Last checked: The date you last confirmed the answer is current

An example entry, for the invented shop:

• Topic: Returns
• Approved answer: “You can return unused items within 30 days of delivery for a refund to your original payment method. Please email us your order number, and we’ll send return instructions.”
• Must be exact: 30 days, unused, original payment method
• Do not promise: Refunds for items past 30 days, free return shipping
• Hand off when: The item arrived damaged, the customer mentions a dispute with their card company, or the request is outside the window
• Last checked: (today’s date)

Those rules are invented for the example. Write down what’s true for your business.

Step Three: Prompt AI to Use Only the Bank

Now the part that saves time. When a customer email comes in, paste it with your bank. Use a prompt like this:

You are helping me draft a reply to a customer email for my small business.

Here is my approved answer bank:
[paste the entries]

Here is the customer email (personal details removed):
[paste email]

Please:
1. Identify which entry in my answer bank applies. If more than one applies, list them.
2. Draft a short, friendly reply using ONLY the facts in that entry.
3. If the email asks about something that is not in my answer bank, do not answer it. Instead write "NO APPROVED ANSWER" and list what the customer is asking, so I can reply myself.
4. If the email matches a "Hand off to a person when" situation, write "HAND OFF" and explain which situation it matches. Do not draft a customer reply in that case.

Rules:
- Do not add policies, timeframes, discounts, or promises that are not in the bank.
- Do not apologize in a way that admits fault or promises compensation.
- Do not guess order status, tracking details, or stock levels.
- Keep the reply under 120 words.

Treat this prompt as my suggestion: it is not taken from any official guide.

The “NO APPROVED ANSWER” and “HAND OFF” labels are the heart of the system. They turn the AI from a confident guesser into a sorting assistant that knows its limits.

Worked Example for Customer Email Answer Banks

Take three invented emails sent to the candle shop.

Email one: “Hi, I bought a candle two weeks ago and it’s the wrong scent. Can I send it back?” The AI matches the Returns entry. The customer is within 30 days, but the email says nothing about whether the item is unused. The draft reply uses your approved wording, asks for the order number, and asks the customer to confirm the candle is unused, because that condition is in your entry.

Email two: “Do you do custom labels for my wedding?” No entry covers custom orders, so the output is “NO APPROVED ANSWER,” with a note that the customer is asking about custom labels for a wedding. That’s your signal to answer personally, and maybe add an entry later.

Email three: “My candle arrived cracked, and I want a refund today or I’ll dispute the charge with my bank.” This matches “damaged item” and “dispute,” both listed under hand-off. The output is “HAND OFF,” with a short explanation. You reply yourself, calmly and quickly.

None of these emails are real, but they show how the bank works: one routine reply drafted, one gap spotted, one sensitive case sent to a human.

Add a Tone Note

Your answers should sound like you. Add a few lines at the top of your bank:

• How you greet people (“Hi [first name],” or “Hello,”)
• How you sign off
• Words you like and words you avoid
• Whether you use exclamation points

Paste a sample of two or three real replies you’re proud of, with personal details removed. The AI will match the style without changing the facts.

Keep Customer Information Out of the Prompt

Remove names, addresses, phone numbers, payment details, and order numbers before pasting an email into any AI tool, and check your tool’s privacy settings and your own privacy policy. Customers share things in emails that they’d expect you to protect. If a reply requires a name, add it yourself after the draft is done.

Three Entries the FTC Shipping Rule Helps You Write

If you sell online, the FTC: Business Guide to the Mail, Internet, or Telephone Order Merchandise Rule gives the rules behind the “Where’s my order?” entry. It applies to internet orders, and if you state no shipping time you must be able to ship within 30 days of receiving a properly completed order (50 days if the buyer applied for credit to pay). If you can’t ship on time, you must offer the buyer, without being asked, a choice between agreeing to the delay and canceling for a full, prompt refund, and the notice must include a revised ship date, the right to cancel and a free way to cancel. If the revised date is 30 days or less away, silence counts as agreeing to the delay; if it is longer, or unknown, the order is automatically canceled unless the buyer agrees otherwise. A refund is generally due within seven working days of cancellation, and the rule does not let you substitute store credit for a required refund. Write the delay entry with those four items in it, and keep it as a template the AI fills only with your actual new date.

Keeping the Bank Current and Safe

In my view, an outdated answer is worse than none. Build in a routine:

• Review dates. Check each entry’s “last checked” date every few months, and whenever a policy changes.
• A change log. One line each time you edit: date, entry, what changed.
• New entries from gaps. Every “NO APPROVED ANSWER” that shows up three times may deserve an entry.
• Retire old entries. Remove seasonal rules once the season is over.

It’s also worth checking the bank against what your website, receipts, and policy pages say, so customers don’t see two different rules.

What AI Should Not Decide for Customer Email Answer Banks

• Refunds, exceptions, and compensation. Those are business decisions.
• Legal or contractual statements. If something sounds legal, handle it yourself or ask a professional.
• Facts about a specific order. AI can’t see your shipping system.
• Whether a customer is telling the truth. That’s your judgment.

Mistakes That Break the System

• Letting AI write the policies. Write them yourself, then use AI for wording.
• Sending drafts without reading them. Every reply is still your reply.
• Making entries too long. Short entries leave less room for misreading.
• Skipping the hand-off list. It’s what protects you in the hard cases.
• Never updating. A stale bank keeps repeating old rules.

Questions About Customer Email Answer Banks

What is a customer email answer bank? It’s a document of approved answers to your most common customer questions, used as the only source when drafting replies, whether by you, a helper, or AI.

Can AI write customer emails without making things up? It can, if you give it your approved answers and tell it to use only those, to flag anything not covered, and to hand off sensitive cases.

How many entries do I need? Start with the three or four topics that make up most of your email. Add more as gaps show up.

Should I paste customer emails into AI tools? Remove personal and payment details first, and check the tool’s privacy terms. When in doubt, summarize the question instead of pasting the email.

How often should I update the answer bank? Whenever a policy changes, and at regular check-ins, such as every few months.

What if the AI ignores my rules? Restate the rules at the top of your next message. If it keeps adding details, trim your bank to the essentials and say “use only the text above.”

Does this replace human customer service? No. It speeds up routine replies, and the hand-off rule keeps people in charge of anything sensitive.

Build Three Entries This Week

Look at your last 30 customer emails, pick the three most common topics, and write an approved answer for each, including what to hand off. A customer email answer bank isn’t complicated, and once it’s in place, your replies may get faster, your answers more consistent, and the AI stays inside what you’ve actually approved.

This article offers general business tips, not legal advice. Policies, consumer-protection rules, and privacy obligations vary by location and business. The example shop, entries, and emails are fictional.

Sources:

• CFPB: Chatbots in consumer finance
• FTC: Business Guide to the Mail, Internet, or Telephone Order Merchandise Rule