Blog

Quick answer: cognitive shuffling has small, early evidence for helping people fall asleep, and cyclic sighing improved daily mood in a 2023 trial but did not change measured sleep. Both are free to try tonight, and I would treat them as low-risk experiments rather than proven sleep fixes. This is general information, not medical advice.

Method One: Cognitive Shuffling

Cognitive shuffling was developed by Dr. Luc Beaudoin, a cognitive scientist at Simon Fraser University (CNN Health via ABC17). You pick a random, emotionally neutral word, then think of other words starting with each of its letters for about 5 to 8 seconds each, or visualize them. The words should stay unrelated, and trying to control the randomness reduces the effect.

Evidence: in a 2016 study by Beaudoin’s team, 154 college students were assigned to standard pre-bedtime treatment, a formal version of cognitive shuffling, or both. The shuffling group was more likely to improve sleep quality, difficulty falling asleep and presleep arousal. The article notes the sample was small, that research is limited, and that Beaudoin owns companies selling products based on his research (CNN Health via ABC17).

Practical notes from the same article: most people fall asleep within 5 to 15 minutes; if you are still awake after about 20 minutes, get up and do something calming for 20 to 30 minutes, then try again. Try it for several nights before judging, and it does not replace CBT-I or medical advice for insomnia.

Method Two: Cyclic Sighing

In the Stanford-led randomized trial, participants practiced one daily 5-minute session for 28 days. The breathwork groups were cyclic sighing (30 people), box breathing (21) and cyclic hyperventilation with retention (33), compared with 24 people doing mindfulness meditation. The study was fully remote, ran from June to September 2020, and most participants were Stanford undergraduate psychology students (Cell Reports Medicine 2023).

The pattern: a slow inhale, then a second, shorter inhale to fill the lungs, then a slow, full exhale, repeated for 5 minutes. Ideally both inhales are through the nose and the exhale through the mouth. The paper gives no exact seconds for each phase, so there is no official count to follow.

What the Numbers Showed

• Positive affect. On a 10 to 50 scale, the mean daily pre-to-post change was +1.89 for cyclic sighing and +1.22 for mindfulness. In the mixed-effects model, breathwork as a whole and cyclic sighing specifically were both significantly higher than mindfulness (Cell Reports Medicine 2023).
• Negative affect and state anxiety. All groups improved, with no significant differences between groups.
• Respiratory rate. Cyclic sighing showed a significantly greater reduction than mindfulness.
• Sleep. Wearable-derived sleep hours, sleep efficiency and sleep score, and a questionnaire on sleep-related daytime disturbance, showed no significant changes in any group.

So the trial supports cyclic sighing as a mood practice, and it does not show a sleep effect. The authors list limits: the study was exploratory and registered retrospectively, delivery was remote so practice could not be controlled, follow-up was only four weeks, and group sizes were small. One author became an advisor to the wearable maker WHOOP in June 2022.

A Simple Way to Try Both

This routine is my own suggestion, not something the studies tested together:

• If your body feels wound up, try five minutes of cyclic sighing first, since that is the dose the trial used.
• If your mind is looping, try cognitive shuffling in bed.
• Write one line each morning (what you did, and how the night felt) for a couple of weeks to see which, if either, helps you.
• Try each for several nights before judging, as the CNN article suggests.

Let a chatbot tally the log. The averages are simple arithmetic, so a chatbot can do them, as long as it is told not to judge the methods. Remove your name before pasting.

Below is my two-week sleep log. Each row has the date, the method I used that night (SIGHING, SHUFFLING, or NONE), and my own estimate of how many minutes it took to fall asleep.

[paste the log]

Please:
1. For each method, count the nights and add up the minutes, showing each sum.
2. Calculate the average minutes per night for each method, showing the division.
3. Write the number of nights next to each average.

Rules:
- Use only the log.
- Do not say that any method works or does not work.
- Do not give medical advice or comment on whether my sleep is normal.
- If a method has fewer than 7 nights, write "FEW NIGHTS."

An invented example. Say over 14 nights you logged NONE on 6 nights (40, 30, 35, 45, 30 and 40 minutes), SHUFFLING on 4 nights (25, 30, 20 and 25) and SIGHING on 4 nights (35, 30, 40 and 35). NONE sums to 220, so the average is 220 divided by 6, or about 36.7 minutes. SHUFFLING sums to 100, so the average is 25 minutes.

SIGHING sums to 140, so the average is 35 minutes. The output also tags SHUFFLING and SIGHING as FEW NIGHTS. The 7-night cutoff is my rule of thumb, and these numbers are made up. A gap like this is a reason to keep logging for another week, not proof; it also matches what the studies suggest, since the sighing trial measured mood, not sleep.

When a Clinician Is the Better Next Step

The American College of Physicians’ 2016 guideline recommends cognitive behavioral therapy for insomnia (CBT-I) as first-line treatment for adults with chronic insomnia. According to ACP’s announcement, CBT-I includes cognitive therapy focused on sleep, behavioral interventions such as sleep restriction and stimulus control, and education such as sleep hygiene, and it can be delivered through individual or group sessions, telephone or web-based modules, or self-help books.

ACP describes chronic insomnia as symptoms that cause significant distress or impairment, occur at least three nights per week for at least three months, and are not linked to other sleep, medical or mental disorders. About 6 to 10 percent of adults meet the diagnostic criteria.

If your sleep has been difficult at that frequency, the structured, evidence-backed route is to talk with a clinician about CBT-I. Shuffling and sighing can sit alongside it as comfortable evening habits.

Where Relaxation Habits Fit in the Sleep Guidelines

Cognitive shuffling and cyclic sighing are both relaxation-style habits, so it helps to see how relaxation approaches are rated in general. The National Center for Complementary and Integrative Health (NCCIH) digest on mind and body practices for sleep disorders gives a balanced picture.

• Relaxation techniques: the digest describes “a small amount of low-quality evidence” that relaxation techniques alone can help chronic insomnia. It also notes that the American Academy of Sleep Medicine (2021) conditionally recommends relaxation therapy as a single-component treatment for chronic insomnia in adults, depending on patient preferences, provider qualifications and treatment availability.
• Yoga and tai chi: several studies suggest they may help sleep, but a 2019 VA/DoD guideline found insufficient evidence to recommend for or against either for insomnia.
• Meditation and mindfulness: both the 2019 VA/DoD guideline and the 2021 AASM guideline found not enough evidence to recommend mindfulness by itself.

Two honest points follow from this. First, the digest does not cover breathing exercises as a separate approach, so it does not rate cyclic sighing for sleep, which matches the 2023 trial’s finding of a mood effect and no measured sleep change. Second, the encouraging part is that a professional body does conditionally recommend relaxation therapy for chronic insomnia, so a calming evening routine is a reasonable thing to build, ideally next to the CBT-I route described in the clinician section.

A Night Plan Built From the Studies’ Own Timings

Both methods come with timings in the sources, which you can turn into a simple plan for the evening. The plan below is my suggestion built from those timings: five minutes is the daily dose in the cyclic sighing trial, and the CNN article says most people fall asleep within 5 to 15 minutes, with a break advised if you are still awake after about 20 minutes.

An invented night plan. Say you turn the lights out at 10:30 p.m. You spend 10:30 to 10:35 on cyclic sighing, which is five minutes. From 10:35 you start cognitive shuffling and let it run. By the 5-to-15-minute mark after lights out, which is 10:35 to 10:45, many people are asleep, according to the article.

If you are still awake at about 10:50, which is 20 minutes after lights out, the article’s advice is to get up and do something calming for 20 to 30 minutes, which means returning to bed between 11:10 and 11:20, then try again. The times are made up, and you can shift them to your own bedtime.

Using the plan for several nights before judging, as the CNN article suggests, and noting each morning how long it took you to fall asleep, gives you the log for the chatbot prompt above. After that you can decide which habit to keep from your own numbers instead of from a headline.

Where These Fit Among NIH-Listed Treatments

The NHLBI: Insomnia Treatment page lists three treatment groups for insomnia: cognitive behavioral therapy for insomnia (CBT-I), a 6- to 8-week plan usually recommended first for long-term insomnia and available in person, by phone or online; healthy sleep habits such as a cool, dark, quiet bedroom and a consistent schedule; and medicines. It also advises a consistent wind-down routine, with reading, soothing music or a warm bath as examples. Cognitive shuffling and cyclic sighing are relaxation-style tools that sit alongside those habits rather than replacing them, so if trouble sleeping lasts, CBT-I is the option NHLBI names first.

Setting Expectations for the Two Methods

Both methods cost nothing and ask for only a few minutes. The sources above do not show either one to be a proven sleep fix, so treat them as experiments and keep expectations modest until your own nights show a difference.

Cognitive Shuffling and Cyclic Sighing: Sources

• CNN Health (via ABC17): Cognitive shuffling may help quiet racing thoughts to fall asleep (April 2025)
• Yilmaz Balban et al., Brief structured respiration practices enhance mood and reduce physiological arousal, Cell Reports Medicine (2023)
• American College of Physicians: ACP recommends cognitive behavioral therapy as initial treatment for chronic insomnia (2016)
• NCCIH: Mind and body practices for sleep disorders
• NHLBI: Insomnia Treatment

Quick answer: a randomized trial by METR found that experienced developers using AI tools in early 2025 took 19 percent longer on real tasks, even though they believed AI had made them about 20 percent faster (METR).

METR has since marked that result as out of date, and its newer study points to likely speedups but is too unreliable to quantify (METR’s February 2026 update). The lasting lesson is the gap between feeling fast and being fast, and it gives you a habit to try: measure your own time.

The Study Behind the Headline

According to METR, the trial involved 16 experienced developers who had contributed for multiple years to large open-source repositories, averaging more than 22,000 stars and over a million lines of code. They worked on 246 real issues they had listed as valuable to their repositories: bug fixes, features and refactors, averaging about two hours each.

Each issue was randomly assigned to allow or disallow AI use. When AI was allowed, developers could use any tools they chose, primarily Cursor Pro with Claude 3.5 and 3.7 Sonnet. Developers recorded their screens and self-reported total implementation time.

With AI allowed, developers took 19% longer to complete issues. Before the study they forecast AI would speed them up by 24%, and after it they estimated AI had sped them up by 20%.

What the Authors Ruled Out, and What They Cautioned

METR reports that it ruled out several experimental artifacts: developers used frontier models, complied with their treatment assignments, did not differentially drop issues, and submitted pull requests of similar quality with and without AI. The slowdown persisted across different outcome measures and subsets of the data. The authors investigated 20 potential factors and found evidence that five likely contribute to the slowdown.

They also list limits. They do not claim AI slows most developers or fails in other domains, and they say existing AI could be used more effectively in this setting than the study’s tooling allowed. Developers had used Cursor for only a few dozen hours before and during the study, so learning effects cannot be ruled out. Self-reported time can be inaccurate and overoptimistic, which is part of why the study’s gap between perception and measurement is interesting.

What METR Reported Next

METR’s February 2026 update says its second study, using late-2025 tools with 57 developers across 143 repositories and more than 800 tasks, is unreliable because of selection effects. More developers said they would not work without AI, so the study is missing those with the most optimistic views of AI, and 30% to 50% of surveyed developers said they withheld some tasks because they did not want to do them without AI. Time tracking was also unreliable for developers running multiple agents at once.

The reported estimates were -18% for the 10 returning developers (95% interval from -38% to +9%) and -4% for newly recruited developers (-15% to +9%), on the same scale as the original study, where a positive number means longer task time and a negative one means shorter. Both intervals include zero. METR says it believes developers are likely more sped up in early 2026 than in early 2025, but that its data is “only very weak evidence” for that, and it is redesigning the experiment.

For you, the practical reading is: tools change quickly, any single study is a snapshot, and your own measurement is the evidence that applies to your work.

Turning the Gap Into a Habit You Can Test

The study’s most practical message is that felt speed and measured speed can point in opposite directions. These steps are my suggestions, not findings from the study.

• Log your own time for two weeks. Write one line per task: what it was, whether you used AI, and how long it actually took. Compare the averages at the end.
• Write the test or expected result first. That gives the AI’s output a concrete check instead of relying on how plausible the code looks.
• Read each suggestion before accepting it. Treat it the way you would treat a first draft from a colleague.
• Compare task types. If the log shows AI helps on small, well-defined tasks and costs time on complex ones, use it where it helps.
• Re-measure when your tools change. METR’s own results moved between 2025 and 2026, so a log from last year may not describe this year.

A Quick Calculation for Your Own Log

Here is the arithmetic with assumed numbers. Suppose a task takes 10 hours without AI. If AI made it 19 percent longer, it takes 10 × 1.19 = 11.9 hours. If you believed AI had made you 20 percent faster, you would have expected about 10 ÷ 1.2 = 8.3 hours. The gap between 8.3 and 11.9 hours is 3.6 hours, which is the kind of difference a personal time log can catch. The 10 hours is an assumption for illustration.

Below is my work log for two weeks. Each row has the task, whether I used AI (YES or NO), and the minutes it took.

[paste the log]

Please:
1. Count the tasks and add up the minutes for the YES group and the NO group, showing each sum.
2. Calculate the average minutes per task for each group, showing the division.
3. Calculate the percent difference as (YES average minus NO average) divided by the NO average, showing the steps. A positive number means tasks with AI took longer.
4. Recalculate step 3 after removing my single longest YES task, so I can see how much one task moves the result.

Rules:
- Use only the log.
- Do not say whether AI helps or hurts me.
- If either group has fewer than 10 tasks, write "SMALL SAMPLE."

An invented example. Say the NO tasks took 90, 110, 120, 130 and 150 minutes, which sum to 600, so the average is 600 divided by 5, or 120. The YES tasks took 200, 100, 110 and 130, which sum to 540, so the average is 540 divided by 4, or 135. The difference is 135 minus 120, or 15 minutes, and 15 divided by 120 is 12.5%, so AI tasks took 12.5% longer. Now remove the longest YES task, 200 minutes.

The remaining three sum to 340, the average is about 113.3, and (113.3 minus 120) divided by 120 is about -5.6%, so the sign flips. Both groups are far under 10 tasks, so the output should say SMALL SAMPLE. The 10-task cutoff is my rule of thumb, not a statistical standard, and the log is invented. The takeaway is the same as the trial’s: measure over enough tasks before trusting either feeling or a single number.

The Five Factors Behind the Slowdown, From the Paper

The blog post says METR examined 20 potential factors and found evidence that five likely contribute to the slowdown, but the details sit in a table. The full paper on arXiv (Section 3.3 and Appendix C) lists them:

• Over-optimism about AI usefulness. Developers forecast a 24% time reduction and afterward estimated 20%, while observed time rose 19%. The authors suggest this optimism may lead developers to keep using AI despite its cost.
• High familiarity with the repositories. Developers were slowed more on issues where they reported high prior exposure and low need for outside resources. They averaged 5 years of experience and 1,500 commits on their repositories, which the authors suggest makes AI less helpful than in earlier studies.
• Large, complex repositories. The repositories averaged about 10 years old and over 1,100,000 lines of code. Developers said AI performed worse there, sometimes making unwanted edits elsewhere in files.
• Low AI reliability. Developers accepted fewer than 44% of Cursor’s generations; 75% read every line of AI-generated code, and 56% often made major changes to clean it up. In a labeled subset of 44 issues, they spent about 9% of their time reviewing and cleaning AI outputs.
• Implicit repository context. This one rests on qualitative evidence only: developers said AI lacked tacit knowledge of the codebase, such as undocumented edge cases and backward-compatibility needs.

The paper reports the shift in how time was spent in a different section: with AI allowed, developers spent less time actively coding and searching, and more time prompting, waiting, reviewing and idling.

Turn Each Factor Into a Column in Your Log

The factors suggest columns worth adding to the log from the habit section above. These mappings are my suggestions, and the numbers they refer to come from the paper.

• Over-optimism: write your guess of the time before you start, then the real time. The paper’s 24% forecast versus a 19% increase shows why the guess alone is not enough.
• Reliability: note whether you kept the suggestion, edited it heavily or threw it away. The paper’s under-44% acceptance figure gives you something to compare against.
• Review cost: note the minutes spent reading and fixing AI output on each task. The paper put this at about 9% of time in its subset. For scale, 9% of a two-hour task is 0.09 × 120 = 10.8 minutes.
• Repository size and familiarity: tag each task as “know this code well” or “new to me.” The paper found larger slowdowns on well-known code, so your log may show a different pattern for unfamiliar code.

This prompt computes the review share from your log (my suggestion):

Below is my work log for two weeks. Each row has the task, whether I used AI (YES or NO), the total minutes, and for YES tasks the minutes I spent reviewing or fixing AI output.
[paste the log]
Please:
1. Add up the total minutes and the review minutes for the YES tasks, showing each sum.
2. Calculate review minutes divided by total minutes as a percentage, showing the division.
3. List the YES tasks where the review share is above 20%.
Rules:
- Use only the log.
- Do not say whether AI helps or hurts me.
- If I have fewer than 10 YES tasks, write "SMALL SAMPLE."

An invented example. Say the YES tasks in your log took 540 minutes in total, and 54 of those were spent reviewing and fixing AI output. The review share is 54 ÷ 540 = 10%, a bit above the paper’s about 9%.

One task at 120 minutes with 36 spent reviewing has a share of 36 ÷ 120 = 30%, so it appears on the list of tasks above 20%. Looking at that task tells you what kind of work makes the output expensive to check. The log is made up; the 9% figure is from the paper, and the 10-task cutoff is my rule of thumb.

The encouraging part is that the paper also says what could change the picture: more experience with the tool (the developers had only a few dozen hours of Cursor use), better setups, and different kinds of work. Your own log tells you which of those applies to you.

My View on AI Coding Assistants

This closing part is my opinion. I think using AI well is a skill, and the log is the quickest way I know to learn where it helps you and where it does not. Your results may differ from the trial’s, and that is the point of measuring them.

AI Coding Assistants: Sources

• METR: Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity
• METR: We are Changing our Developer Productivity Experiment Design (Feb. 24, 2026)
• arXiv:2507.09089: Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity

For another look at using AI without handing over the thinking, see learning to code with AI.

Quick answer: AI budget trip planning works best as a first draft, because AI tools can sound confident while getting transit times, locations, seasons and schedules wrong, according to Generali Travel Insurance. Use the risk types as a checklist: unrealistic transit times, incorrect locations, seasonal and cultural gaps, missing buffer time and accessibility gaps. This guide turns that checklist into a workflow.

What Generali Says Can Go Wrong

Generali Travel Insurance publishes a list of AI trip-planning risks. It notes that AI tools “sound confident, structured, and authoritative,” so travelers may trust them too much. The risks it lists:

• Unrealistic transit times. AI may compress travel between attractions unrealistically, assume ideal driving conditions, and leave out customs, immigration and terminal changes on flight connections.
• Incorrect locations. AI may place attractions in the wrong part of a city or give directions to places that do not exist.
• Seasonal and cultural gaps. It gives the example of recommending beaches during monsoon seasons, and may miss local holidays, seasonal closures and reduced transit hours.
• Missing buffer time. Packed schedules can produce rushed airport transfers with little margin for security lines or terminal changes.
• Accessibility gaps. A viewpoint reachable only by stairs, or a route with steep elevation changes.

Its practical advice is to check addresses, confirm public transportation schedules and ticket availability, confirm whether landmarks require timed-entry reservations, build in buffer time, and prepare backup plans for key activities. It recommends treating AI output as “a helpful starting point rather than a fully vetted solution.”

A Workflow That Turns the Errors Into Checks

• Generate a first draft with any AI tool. In my view, that is where the tool helps most.
• Check every closed-status and opening-hours claim before booking anything. Generali lists seasonal closures and reduced hours among the things AI can miss, so this check comes first.
• Restate your budget and ask the tool to justify every recommendation against it. This is my suggestion: an itinerary can include expensive picks even when the plan is meant to be economical.
• Use a second source to verify the first tool’s output, such as the venue’s own website.
• Check timed-entry rules and transit connections for each day, per Generali’s advice.
• Look at hotel changes. A plan that moves you to a new hotel every night deserves a second look, in my view.

Prompt Habits From Generali’s Tips

Generali’s AI travel planning tips suggest several ways to get better output from the start:

• Tell the AI what you would like to see, such as whether you prefer hiking or relaxing on beaches.
• Use loose dates, such as the season or how far away the trip is, so the AI can suggest timely events.
• Ask it how it got its information, since it may still produce inaccurate information.
• Compare suggestions from several AI tools rather than relying on one.
• Ask the AI why a recommended restaurant or museum is popular to learn what reviewers say.

Putting those together with the budget check, here is a prompt (my suggestion):

Plan [number] days in [city] in [season] for [group].

My total budget for lodging, food, transport and activities is [amount], split as [amount] for each category.
I like [interests] and want to avoid [dislikes].

For every item, give:
- the name
- the type of cost
- a cost estimate marked ESTIMATE

Rules:
- Label anything priced above [limit] as OVER BUDGET.
- Leave buffer time between items.
- Tell me which items need timed-entry tickets.
- End with the five facts I should confirm on official sites.

In my view, the ESTIMATE and OVER BUDGET labels address the expensive-picks problem, and the “five facts” line turns the tool into a to-do list for your verification.

An invented example. Say the plan is 4 days with a $1,200 budget split as lodging $480, food $300, transport $200 and activities $220. Those add up: 480 plus 300 is 780, plus 200 is 980, plus 220 is 1,200.

If the draft includes a $150 tasting-menu dinner and your per-meal limit is $60, the output should label it OVER BUDGET, and the subtraction 150 minus 60 shows it is $90 over. Swapping it for a $45 dinner leaves $15 under the limit and returns $105 to the food line. The city, amounts and dinner are made up for the example.

Check the Transit Line Against Official Fare Caps

Transit is the budget line where an official figure can quickly correct an AI estimate. The MTA publishes exact fares and caps for New York. According to the MTA’s announcement of its January 2026 fare changes, the subway and local bus base fare is $3, and no customer pays more than $35 for subway and local bus rides in a week, which covers 12 rides in a 7-day period. The reduced fare is $1.50, with a $17.50 weekly cap, and express buses have a separate $67 weekly cap.

The same announcement describes how the cap works:

• The cap is a rolling 7-day cap, so it follows your own seven days of riding rather than a calendar week.
• Customers don’t prepay. The cap applies as you tap and ride.
• It covers subway and local bus rides. Express buses have their own $67 weekly cap.
• Reduced-fare riders are capped at $17.50.

An invented check. Say an AI draft for a 7-day New York trip budgets 2 subway rides a day at $3, which is 14 rides and $42 in total. The official weekly cap is $35, so the week costs at most $35, and the draft overshoots by 42 − 35 = $7. That money can move to food or an attraction. The draft is made up, and the fare and cap are the MTA’s published January 2026 figures.

Another useful number from the same announcement: the cap starts to help on the 12th ride. Eleven rides cost 11 × 3 = $33, and a 12th at $3 would reach $36, which the cap holds at $35, a saving of $1. Every ride after that in the same 7 days is free, so rides 13 and 14 would otherwise have cost 2 × 3 = $6 more. That is my arithmetic from the published fare and cap.

Swap Each Estimate for the Official Figure

The same idea works for every line of a draft itinerary. Swap the chatbot’s guess for the figure or fact the organization itself publishes, which is the “second source” step in the workflow above made specific.

Budget line	Replace the estimate with
Local transit	The transport authority’s fare page
Opening hours	The venue’s own page
Timed entry	The venue’s ticket page

Generali’s advice points the same way: check addresses, confirm public transportation schedules and ticket availability, and confirm whether landmarks require timed-entry reservations. The positive side is that many of these gaps are the kind a visit to an official page fixes, and the money you save by correcting the transport line, as in the example above, pays for the time you spend on it. This framing is my reading.

Questions About AI Budget Trip Planning

Is it still worth using AI for budget trip planning given the risks Generali lists? In my view, yes as a first draft, as long as verification happens before booking, not after.

How do I quickly check if a recommended restaurant or attraction is still open? My suggestion: search the specific name plus “hours” or “closed” for each item before booking, and use the venue’s official page.

Booking Checks From the FTC

Once the AI draft is fact-checked, the FTC: Avoid Scams When You Travel page adds booking checks. Confirm whether you are buying from the app or from the airline or resort, since that affects refunds, points and change fees, and whether you are buying a ticket or only making a reservation. For vacation rentals, search the property address with the owner’s name, check whether the rental company’s own site lists the property, and be wary of premium properties priced far below comparable ones. Don’t pay only by wire transfer, gift card, payment app or cryptocurrency, get cancellation and refund policies in writing before paying, and walk away from anyone pressuring you to decide right away.

Where I’ll Add My Own View

Everything above is data. This closing part is mine.

I think of AI budget trip planning the same way I’d think of any first draft: useful for gathering ideas, pulling together a lot of raw material fast, and giving you a starting budget to react to, with a human double-check before anything gets booked.

The other thing I’d add: how much you get back depends on how much detail you put in. Giving the tool your actual budget number, your dates, what you care about and what you don’t, and asking it to justify each choice against those specifics, is what separates a plan worth using from one you’ll end up rewriting.

AI Budget Trip Planning: Sources

• Generali Travel Insurance: The Dangers of AI Trip Planning
• Generali Travel Insurance: AI for Travel, How to Plan a Trip and Enhance Your Experience
• MTA: MTA Board Adopts Fare and Toll Increases to Take Effect January 2026
• FTC: Avoid Scams When You Travel

Quick answer: in a July 2026 Kiplinger test, chatbots gave generally sound financial answers but often missed the human context that certified financial planners picked up (Kiplinger). The useful move is to use each for what it does well: let the chatbot organize your questions and options, and let a planner supply the follow-up questions, the trade-offs and the accountability. This is general information, not personal financial advice, and I am not a licensed advisor.

What the Kiplinger Test Did

Kiplinger writer Chris Taylor gave AI chatbots five financial scenarios: a 35-year-old building a budget, a 45-year-old couple asking about an investment mix, a 55-year-old weighing retirement against college costs, a 65-year-old couple planning Social Security and income, and a 75-year-old couple planning a legacy. The article names ChatGPT, Gemini and Claude, and quotes certified financial planners reacting to the answers (Kiplinger).

Scenario by Scenario: Where Chatbots and Planners Differed

• The budget. The 35-year-old earned $65,000, paid 30% of salary in rent, contributed 4% to a 401(k), owed $1,200 on a credit card and paid $300 a month on a student loan. ChatGPT made the credit card the top priority and raised the 401(k) to at least 8%. Claude recommended the avalanche method (highest-rate debt first) while building three months of expenses and opening a Roth IRA. Planner Valerie Rivera suggested the 50-30-20 rule as a starting point and contributing enough to get the full employer match. Planner Jeff Judge found it puzzling to recommend an IRA before the match.
• The investment mix. For a couple aged 45 with $125,000 income and $250,000 in a 401(k), Gemini noted they should have 3 to 4 times their salary saved and offered a target-date fund or a DIY index-fund mix. Claude named specific 2045 target-date funds and a DIY mix of 35% U.S. large-cap, 10% small and mid-cap, 20% international developed, 5% emerging markets, 15% U.S. bonds, 5% TIPS, 5% REITs and 5% cash. Planner Juan HernandezAriano said the couple should “push harder” and suggested about 80% stocks and 20% bonds. The article says the bots ignored risk tolerance, whether the funds were available in the 401(k), and the workload of rebalancing a complex mix.
• Retirement versus tuition. A 55-year-old with $400,000 saved, $140,000 household income and $40,000 a year in tuition for each of two children. Gemini called it a “tight spot” and, assuming 7% annual gains, projected $980,000 by 65 against the $1.375 million it estimated was needed. Claude called the plan “roughly on track, but with meaningful vulnerability.” HernandezAriano projected the client on track at 65 assuming returns a little over 6%, and said “Children can borrow for school, but parents can’t borrow for retirement.”
• The retirement playbook. For a 65-year-old with a $1 million portfolio, ChatGPT suggested a guardrails approach starting with a 4% withdrawal, and Gemini suggested the lower earner claim Social Security at retirement while the higher earner delays to 70. HernandezAriano said he is “not a fan” of fixed withdrawal rules and recommended four time-based buckets with a 3% to 3.5% withdrawal rate. The article notes both approaches use flexible withdrawals and delayed Social Security.
• The legacy. For a 75-year-old couple with a $1.5 million portfolio and a $750,000 home, ChatGPT wrongly said the $30 million married-couple federal exemption was set to decrease, when the article says it is designed to be permanent under the One Big Beautiful Bill Act. Claude warned that some states have estate tax exemptions as low as $1 million. Rivera focused on account type: brokerage accounts get a cost-basis step-up at death, Roth accounts pass tax-free, and inherited 401(k)s and traditional IRAs must generally be withdrawn within 10 years by nonspouse heirs.

The Pattern Across All Five

The article’s bottom line is that the chatbots’ advice was generally sound and followed common planning guidelines, but they sometimes stated facts incorrectly, their information could be outdated or ignore personal details, and they rarely asked follow-up questions. Judge said “The level of meeting people where they are is definitely lacking.” The article advises treating AI as a powerful tool and not a reliable authority, and concludes that combining AI with human planners may be the smartest approach (Kiplinger).

A Practical Way to Use Both

• Give the chatbot your whole situation. The test suggests the gaps came from missing context, so include your other debts, your timeline and what you could live with in a downturn. That is my suggestion.
• Have it ask you questions first. The article found the bots rarely asked follow-ups, so request them explicitly (prompt below).
• Feed it official figures for tax and legal specifics. The ChatGPT estate tax error is the example, so paste the current threshold from an official source into the prompt instead of asking the chatbot to recall it.
• Check plan details yourself. A suggested fund is only useful if your plan offers it.
• Keep a person accountable for big decisions. The article quotes K. Dane Snowden of the CFP Board: “There should always be a human accountable.”

A Prompt That Makes the Chatbot Ask First

Paste this before describing your situation (my suggestion):

I want to think through [decision].

Before you give any advice, ask me the questions a financial planner would ask, one group at a time:
- my goals and timeline
- my income and savings
- my debts
- the accounts available to me
- how I would react to a market drop
- my family situation

After I answer:
1. Give me options with the trade-offs of each, not one answer.
2. Mark any figure or rule that could be outdated.
3. List questions I should bring to a certified financial planner.

Rules:
- Use only what I tell you.
- Do not present any option as the one I must choose.

The last line turns the chatbot session into preparation for a planner meeting, which is the combination the article points toward.

Worked example from the article’s numbers. Take the 35-year-old in the test, who earned $65,000, paid 30% of salary in rent and put 4% into a 401(k). Rent is 0.30 times 65,000, or $19,500 a year. The 4% contribution is $2,600 a year. Raising it to the 8% that ChatGPT suggested would be $5,200, which is $2,600 more a year, or about $217 more a month (2,600 divided by 12).

A chatbot using the prompt above would ask about the employer match, the credit card rate and the emergency savings before ranking those choices, and the list of questions at the end is what you would bring to a planner. The arithmetic uses the figures reported in the Kiplinger article.

Questions to Bring to a Human Professional

The SEC’s investor education site publishes questions to ask when hiring an investment professional. A few that fit this topic:

• “What experience do you have, especially with people like me?”
• “How do you get paid?” and “What are the fees that I will pay for products and services?”
• “Are you registered with the SEC, a state securities regulator, or FINRA?”
• “How often should I expect to hear from you?”

The same bulletin also suggests asking whether a third party, such as a fund company, pays them or their firm, and roughly how much they would earn in a year from a $1,000 investment. These are the human-context questions that the test found chatbots did not raise on their own.

Settling the Claiming-Age Question With Official Numbers

Several of the test’s answers turned on when to claim Social Security, so the official rules are a good tiebreaker. In the 65-year-old couple scenario, Gemini suggested the lower earner claim at retirement while the higher earner delays to 70. The Social Security Administration’s page on early and late retirement explains the mechanics:

• Benefits are reduced by 5/9 of 1% for each month before normal retirement age, up to 36 months, and by 5/12 of 1% per month beyond that.
• For people born in 1943 or later, delayed retirement credits add 8% per year past normal retirement age.
• The page says no credit is given after age 69, so credits stop at age 70, the age of the largest benefit.

The numbers for full retirement age 67. At 62 the reduction is 36 months × 5/9% (20%) plus 24 months × 5/12% (10%), for a maximum 30% cut, so the benefit is 70% of the primary insurance amount. At 70, three years of 8% credits add 24%, so the benefit is 124%. The 124% figure is my arithmetic from the page’s rules, not a number the page states.

An invented example. Say a person’s full-retirement-age benefit is $2,000 a month. At 62 it would be 2,000 × 0.70 = $1,400. At 70 it would be 2,000 × 1.24 = $2,480. The difference is $1,080 a month, and $2,480 is about 77% more than $1,400 (124 ÷ 70 ≈ 1.77). The $2,000 is made up, but the percentages come from the SSA page.

The encouraging takeaway is that the claiming decision is a place where careful planning pays off, and it is also where the test’s planners and chatbots could differ. A chatbot can compute these figures if you give it the rules; a planner can tell you whether waiting fits your health, savings and spouse’s situation. The best use of the two together is to bring the calculation to the planner and ask about the trade-offs.

Verify the Planner Before You Book the Meeting

Choosing a human advisor deserves the same care as choosing a tool. CFP Board’s consumer site explains what it can and cannot confirm for you.

• Certification: CFP Board confirms that people listed in its Find a CFP® Professional search currently hold CFP® certification. It does not verify other information they provide, such as claimed areas of focus.
• Fiduciary duty: CFP® professionals agree to act as fiduciaries, putting your interests first at all times when giving financial advice and planning. CFP Board can sanction a professional who violates the standard, including by revoking certification.
• Discipline: the search tool shows whether CFP Board has publicly disciplined a professional and whether they have made a bankruptcy disclosure. For other records, CFP Board links to FINRA BrokerCheck, the SEC’s Investment Adviser Public Disclosure site and state regulators’ resources, and it does not guarantee the accuracy of those external sites.
• Interview several: because CFP Board does not verify the other details professionals provide, the site advises interviewing several advisors before choosing one.

A before-the-meeting routine (my suggestion). First, look the planner up in the CFP Board search and read the disciplinary information. Second, bring the questions from the SEC bulletin above, especially how they are paid. Third, bring the list of questions your chatbot session produced. Together these three steps turn the first meeting into a short, focused conversation instead of a sales pitch you have to decode.

My View on AI vs a Real Financial Advisor

This closing part is my opinion. I think chatbots are useful for research and first drafts, and that the more of your real situation you give them, the more useful the answer will be. For decisions with large consequences, I would pair that draft with a planner who asks you the follow-up questions.

AI vs a Real Financial Advisor: Sources

• Kiplinger: We Gave AI Chatbots 5 Financial Challenges. Here’s How They Did (July 20, 2026)
• Investor.gov: Investor Bulletin, Questions to Ask When Hiring an Investment Professional
• Social Security Administration: Effect of early or late retirement on benefits
• CFP Board: Let’s Make a Plan (verify and find a CFP professional)

Quick answer: Ladders’ 2018 eye-tracking study found recruiters spent about 7.4 seconds on the initial resume screen, up from about six seconds in 2012 (HR Dive). Simple layouts with clear sections did well; cluttered or multi-column ones did poorly. Two university career-center guides add a second checklist for resumes that software reads first. Together they give you a layout that works for both readers, and a resume checker can confirm it.

What the Ladders Study Reported

According to HR Dive’s November 2018 coverage, Ladders Inc. published an eye-tracking study that put average time on the initial resume screen at roughly 7.4 seconds, about 1.4 seconds more than the six seconds it reported in 2012. The article does not give the sample size or the method, so treat the exact figure as a rough indicator, and use the layout findings as the practical takeaway.

Layouts That Did Well and Poorly

• Did well: simple layouts with clear sections and heading titles, and resumes organized around E- or F-pattern reading with bold titles and bulleted accomplishments (HR Dive).
• Did poorly: cluttered layouts with too little white space, multiple columns, long sentences, missing section or job headers, and text that did not guide the eye down the page (HR Dive).

The Second Reader: Software That Parses Your File

Your resume may be read by applicant tracking system (ATS) software before a person sees it. Two university career-center documents give concrete formatting rules for that first read. The overlap with the Ladders findings is large, which makes the checklist easy to follow.

Utah Valley University’s guide to ATS-friendly resumes recommends:

• Clear, traditional section headings such as Education, Experience and Skills, instead of creative ones.
• Fonts such as Arial, Times New Roman or Calibri at 11–12 pt for body text, with no more than 3 font sizes and 2 font styles.
• Avoiding two-column layouts, photos, charts, tables, graphics, text boxes, and headers or footers.
• Tailoring the resume with the exact phrases for skills, qualifications and job titles from the job description, used naturally and not in excess.
• A Word file or a text-based PDF (PDF preferred), never an image-based or scanned PDF.
• Testing the PDF by copying its text into Word to see what the software would read.

Santa Clara University’s career center page on common ATS formatting mistakes adds:

• Consistent dates. Use one format throughout, such as “Jun 2020 – Present” or “06/2020 – Present.”
• Spelled-out acronyms. Write “Search Engine Optimization (SEO)” so both forms are present.
• Contact details in the body. Keep them out of headers, footers and margins.
• Conventional headings. “Work Experience,” “Education,” “Skills” and “Certifications” work; its examples of headings to avoid include “Professional Journey” and “Core Competencies.”
• Plain bullets. Use • or – and minimal color with high contrast.

UVU also notes that if you want a person to see a two-column resume, you can upload it as an “Additional Document” or email it directly to a recruiter or hiring manager. That lets you keep a design you like without costing the ATS read.

How to Turn That Into a Checklist

Combining the human-skim findings with the parsing rules gives this list. The combination is my suggestion, built on the sources above:

• One column, clear headers. Matches the layouts the Ladders study reported doing well and the UVU guidance.
• Standard heading names. Helps both a quick human scan and the software.
• Short bullets that start with results. The study reported long sentences doing poorly.
• White space. Cluttered pages did poorly.
• Consistent dates and spelled-out acronyms. Both are SCU recommendations.
• Copy-and-paste test. Copy your PDF text into a plain document and check that the order and wording survived.

Using a Resume Checker Well

In my view, a checker is useful for the mechanical items above: heading names, date formats, missing contact details, and long bullets. Run your resume through it, then use the checklist to decide which suggestions to accept. I would keep these rules of thumb (my suggestions):

• Accept layout and format fixes that match the checklist.
• Treat keyword suggestions as a prompt to check the job posting, and add a phrase only when it is true about your experience.
• Re-run the copy-and-paste test after each export.

Let a chatbot run the checklist. A chatbot can check pasted resume text against the list above and quote the lines that need work, as long as it is told not to score the resume. Remove your phone number, address, and email before pasting.

Below is the plain text of my resume, copied from my PDF, with my phone number, address, and email removed. Check it against the list below, using only the list.

[paste the text]

Checklist:
1. Section headings are standard (Education, Experience, Skills), not creative.
2. Dates use one format throughout.
3. Acronyms are spelled out at least once, with the short form in brackets.
4. Bullets are short and start with a result.
5. No contact details appear inside text that looks like a header or footer.

For each item write PASS, FIX, or CAN'T TELL, and quote the line it is based on. For every FIX, quote the line and describe the change in one sentence without rewriting my resume.

Rules:
- Use only the pasted text and the checklist.
- Do not score my resume or say whether it will pass any employer's software.
- If the text alone cannot show an item (for example columns), write CAN'T TELL.

An invented example. Say the pasted text has a heading called “Professional Journey”, the dates “Jun 2020 – Present” and “03/2018 – 05/2020”, the term “SEO” with no spelled-out form, and a bullet of 27 words: “Managed the scheduling of vendor deliveries across three warehouses while also coordinating weekly reports for the operations team and training two new coordinators on the ordering system.”

The output would mark item 1 FIX, because Santa Clara University’s list of headings to avoid includes “Professional Journey”; item 2 FIX, because the two date formats differ; item 3 FIX for “SEO”; and item 4 FIX, with my suggestion to split that bullet into two shorter ones. Item 5 would be CAN’T TELL, since the text alone cannot show where the contact details sat. The resume lines are made up.

The Six Data Points That Got Nearly 80% of the Time

The 7.4-second figure comes from the 2018 update, but the original 2012 report, which is available as a PDF, shows what happens inside those seconds. TheLadders’ 2012 eye-tracking report says 30 professional recruiters took part over a 10-week period, and eye tracking recorded where and how long they looked while reviewing resumes and candidate profiles. TheLadders, which ran the study, also sells resume rewriting services, so read its recommendation to hire a professional with that in mind. The measurements themselves are specific and useful:

• Recruiters spent about 6 seconds on the initial fit or no-fit decision, far less than the 4 to 5 minutes they reported in self-assessments.
• Nearly 80% of resume review time went to six data points: name, current and previous job titles and companies, current and previous position dates, and education.
• Beyond those six items, recruiters mostly scanned for keywords that matched the job.
• Poorly organized resumes produced erratic eye movement and higher cognitive load, which made decisions harder.

That is good news for anyone who can edit a resume, because it tells you exactly what to polish. If those six data points are easy to find, the page does its main job in the first glance.

The report also tested rewritten resumes on a 1-to-7 scale. Usability averaged 6.2 after the rewrite, up from 3.9 before it (6.2 ÷ 3.9 is about 1.59, which the report rounds to a 60% increase). Organization and clarity rose from 4.0 to 5.6 (5.6 ÷ 4.0 = 1.4, a 40% increase). The rewritten resumes contained less data and had even formatting, and recruiters described them as easier to read and clearer.

Check Where Your Six Data Points Sit

The prompt below checks those six data points against your own resume text. This prompt asks a chatbot only to find them and report where they sit, so it adds no opinions or invented details. Remove your phone number, address and email before pasting.

Below is the plain text of my resume, copied from my PDF, with my phone number, address, and email removed.
[paste the text]
Find these six data points exactly as written in the text:
1. My name
2. Current job title and company
3. Previous job title and company
4. Dates of the current position
5. Dates of the previous position
6. Education
For each one write FOUND with the quoted line and its position in the text (first third, middle third, or last third), or MISSING if the text does not show it.
Rules:
- Use only the pasted text.
- Do not rewrite my resume or score it.
- Do not guess a missing item.

An invented example. Say the pasted text starts with a name and a three-line summary, then lists “Operations Coordinator, Northline Supply, Jun 2021 – Present” and “Logistics Assistant, Harbor Freight Co.” with no dates, and ends with “B.S., Business Administration.”

The output would show items 1, 2, 4 and 6 as FOUND with their positions, item 3 as FOUND, and item 5 as MISSING, because the previous position has no dates. That tells you the one fix to make: add the dates so the second job reads as clearly as the first. The resume lines are made up.

Use the result alongside the checklist above, and keep the six data points in the top part of a single-column layout. That placement is my suggestion; the six data points and the share of time they received come from the report.

My View on Resume Checkers and the 7.4-Second Scan

This part is my opinion. A perfect checker score is not the goal. I would keep one line that sounds like you, such as a specific project or an unusual way you describe a result, so the resume reads as written by a person once it gets past the software.

Resume Checkers and the 7.4-Second Scan: Sources

• HR Dive: Eye-tracking study shows recruiters look at resumes for 7 seconds (Nov. 2018)
• Utah Valley University Career Services: Creating ATS-friendly resumes
• Santa Clara University Career Center: Common ATS resume formatting mistakes
• TheLadders: Eye-tracking study (2012 report, PDF)

Quick answer: exercises where the answer is already on screen, such as multiple choice and matching, test whether you can recognize a language. Speaking asks you to produce it. U.S. research on study techniques gives specific reasons to add production to an app routine: quizzing yourself from memory beat rereading, and spreading practice over time helped too. Trying to say something also shows you the words you are missing, which is the idea behind the drill below (my design). This guide turns those findings into a daily speaking habit you can add to whatever app you already use.

Recognition vs Production

In a multiple-choice exercise the right answer is one of the options in front of you. In conversation there are no options. You have to pull a word or a sentence out of memory and say it. These are two different tasks, which is why it makes sense to practice both on purpose.

Why Saying Words Beats Seeing Them

A 2013 review in Psychological Science in the Public Interest by John Dunlosky and Katherine Rawson of Kent State, with Elizabeth Marsh, Mitchell Nathan and Daniel Willingham, rated ten common study techniques. According to Kent State University’s account of the 2013 review, practice testing, such as quizzing yourself with flash cards, and distributed practice, meaning study sessions spread over time, earned the top ratings. Summarization, highlighting and underlining, and rereading received low ratings, and Dunlosky said students would benefit by replacing rereading with delayed retrieval practice.

For an app user, the connection is simple. A multiple-choice exercise shows you the answer, much like rereading a list. Saying a word out loud from nothing is a practice test. My suggestion is to treat every moment you cannot find a word while speaking as a quiz result: it tells you exactly what to study next.

A Gap-Finding Drill for Your Next Session

This drill is my own design, built on the practice-testing finding above. It is not a published protocol.

1. Pick one small scene: your breakfast, your commute, or the objects on your desk.
2. Describe it out loud in the language you are learning, with no notes and no app open.
3. Every time you stall, say the missing word in your own language and keep going. Write each one down.
4. When you finish, look the words up, or ask an AI voice assistant for them.
5. Describe the same scene again and try to use the words you just collected.

Steps 3 to 5 are my sequence: find the missing word first, then look it up, then use it. The second description is where the new words get used.

Pull Words From Memory Instead of Re-reading Them

Retrieval, bringing a word back from memory, matters for how long it lasts. In a study by Henry Roediger and Jeffrey Karpicke published in Science in 2008, participants learned foreign-word translations and were then split by whether they kept studying or kept being tested. As reported by Washington University in St. Louis, repeated retrieval after that first success was the critical factor for later retention, and additional study after recall made no measurable difference.

For your routine, that points to a simple rule: after you collect words in the gap-finding drill, test yourself on them the next day by covering the translation and saying the word, rather than rereading your list.

The Numbers Behind Testing Yourself

The details of that experiment show how large the effect was. Purdue’s account of Karpicke and Roediger’s 2008 study in Science says the materials were Swahili-English word pairs presented as computerized flashcards on day one, with a final test one week later. Forty college students took part, ten per group, in four conditions.

• Study all, test all. Words were studied and repeatedly tested even after they were recalled correctly. About 80% were recalled on the final test.
• Drop from study, keep testing. Once a word was recalled it was dropped from studying but still repeatedly tested. About 80% were recalled.
• Drop from testing. Once a word could be recalled it was dropped from repeated testing. Recall was 36% in one group and 33% in the other; the article does not say which condition produced which figure.

The pattern is clear. Repeated testing made what the article calls a “dramatic difference,” and further study of material already recalled easily produced no measurable additional learning. The gap is about 44 percentage points (80 minus 36) or 47 points (80 minus 33), and 80 ÷ 36 is about 2.2 times as much recalled. The researchers also found that students predicted their own recall poorly, and Karpicke’s advice was to keep practicing recall of flashcards even after you can recall an item, rather than dropping it.

What that could look like for you. Suppose you collect 10 new words this week. If your results matched the study’s proportions, which the study does not promise for other words or learners, 80% would be about 8 words remembered after a week, against roughly 3 to 4 words if you stopped testing a word as soon as you got it right. The 10-word list is invented; the percentages come from the experiment.

That is good news for an app user: an app is built around getting an item right and moving on, so a few minutes of saying older words from memory each day is a direct way to keep them.

What the Duolingo Research Does and Does Not Say

Independent research on Duolingo is limited. One study, Loewen and colleagues in ReCALL (2019), followed nine learners studying Turkish on Duolingo for a semester. Participants improved on second-language measures by the end, and time spent on the app had a positive, moderate correlation with learning gains. They also valued its flexibility and gamification, though motivation varied and some were frustrated with the instructional materials.

That is encouraging for using an app as a daily anchor, and the abstract does not break results down by skill, so it does not tell us how much the app helps with speaking. That gap is the reason to pair the app with the speaking habit below. The study is small, with nine learners in one language, so treat it as a good sign rather than a final answer.

Building the Daily Speaking Habit

• Attach it to your lesson. Speak right after your normal app session, using the vocabulary you just saw.
• Choose a length you can repeat every day. Consistency matters more than a long session you skip.
• Let mistakes happen. A stalled sentence is part of the exercise, because it shows you which word to study next.
• Ask for correction as you go. Tell an AI voice partner to correct you and explain briefly why.
• Add a human when you can. A tutor or conversation partner can add cultural context an AI may not.

These steps are my suggestions, not tested protocols.

A Prompt for an AI Voice Partner

Paste this at the start of a voice conversation:

I am learning [language] and my level is [beginner / intermediate / advanced]. Talk with me about [topic].

During the conversation:
- Speak slowly and keep your sentences short.
- When I make a mistake, finish my turn first, then give me the corrected sentence and one short reason.
- When I stall, wait, then offer the word I was missing.

At the end, please:
1. List every word I was missing, each with its translation.
2. List my corrected sentences.
3. Turn the word list into a self-test: the translations in one column, so I can say each word out loud before checking.

The last line gives you the word list for the next day’s retrieval check.

An invented session. Say a beginner describes their breakfast in Spanish and stalls twice, on the words for “toaster” and “spread.” The end-of-conversation list gives those two words with translations, plus the corrected sentences. The next day, the self-test shows only the English meanings, and the learner says each word aloud before checking, which is the retrieval step described above. The session and the words are made up to show the format.

A Ten-Minute Daily Session Built From the Research

You can combine the findings into one short daily session. The layout below is my suggestion, built from the ideas above: practice testing, practice spread over days, and the gap-finding drill. It adds up to 10 minutes (3 + 2 + 3 + 2).

Minutes	Step
3	Describe a scene aloud
2	Collect the words you missed
3	Say yesterday’s words from memory
2	Describe the scene again

Run it right after your normal app lesson, as the checklist above suggests. On day one, the retrieval step has nothing to review yet, so use that time to describe a second scene. From day two on, the words you collected the day before give the retrieval step its material. The Kent State review also gave distributed practice a high rating, so my suggestion is to add a few words from earlier in the week to the retrieval step, not only yesterday’s. Keep going even on days when you remember every word, because Karpicke’s advice, described above, is that success is the cue to keep testing, not to stop.

If you are ready for more, you can stretch each step by the same amount; that choice is up to you, and the research cited here does not set a session length.

My View on AI Language Apps

This closing part is my opinion. I think speaking imperfectly and often is the engine of fluency, and a voice conversation with an AI removes common excuses: no scheduling and no one to be embarrassed in front of. I would rather you try the gap-finding drill for a couple of weeks and see how it feels than trust any claim about timelines, including mine.

AI Language Apps: Sources

• Kent State University: All Study Strategies Not Created Equal, According to Kent State Researchers (Dunlosky et al., 2013)
• Washington University in St. Louis: Practicing information retrieval is key to memory retention
• Loewen et al. (2019): Mobile-assisted language learning: A Duolingo case study
• Purdue University: Karpicke & Roediger (2008) on the importance of retrieval

Another guide built on practice over recognition: learning to code with AI, using hints instead of answers.

Quick answer: the four apps below differ most on price, platform and how much they ask you to do by hand, and I read each one’s own pricing page for the numbers. Before choosing, it helps to know which budgeting step you need help with, because the CFPB breaks budgeting into four steps and each app supports them differently.

If you also want to ask your own questions about your own data, a spreadsheet is the option I keep coming back to. That last point is my opinion, and the rest is what the pages say. This is general information, not financial advice.

Why People Went Looking for a New Budgeting App

Mint shut down on March 23, 2024, and Intuit pointed users to Credit Karma (Monarch Money blog). That sent former Mint users looking for a replacement, and the four apps below are common places to look.

Start With the Four Steps, Then Pick the Tool

The CFPB describes budgeting as four steps, with free worksheets for each:

1. Income. Identify every source, including self-employment, multiple jobs, child support and government benefits.
2. Spending. Log what you spend in an average month in categories such as utilities, housing, eating out and entertainment. If that feels like too much, review one week at a time using receipts or your checking account.
3. Bills. Note each bill’s due date and compare it with when your income arrives, which flags the weeks when spending needs care.
4. Working budget. Combine the first three, and update it after a change in employment or spending habits.

The same CFPB page lists strategies for sticking with it: use a tracking method you can keep up with, such as a daily journal or a folder of receipts reviewed weekly, starting with one month at a time; analyze your spending to find areas to cut back; set a goal; and build a support system of friends or family. California’s Department of Financial Protection and Innovation adds that a simple budget does not require strong math skills (DFPI).

Mapping that to the apps is my suggestion: step 2 (spending) is where automatic transaction import saves the most effort, and step 3 (bills) is where reminders help. If you struggle with the first step or the last, the apps below may help less than the CFPB worksheets do.

What Each App Lists on Its Own Pricing Page

YNAB. The pricing page lists $109 a year, shown as $9.08 a month, or $14.99 a month, with a 34-day free trial and no credit card required when you sign up directly. It describes the YNAB Method as asking “What’s it for?” whenever you spend or make money (YNAB pricing).

Monarch. The pricing page lists $99.99 a year or $14.99 a month. It says its Plus tier adds forecasting tools for planning further ahead, tracking multiple income sources and deeper investment analysis, and that you can cancel anytime and there is a money-back guarantee (Monarch pricing). The page does not say which tier the two prices belong to.

Copilot Money. The pricing page lists $13 a month or $95 a year, and says it is available on iPhone, iPad, Mac and the web. It does not mention Android (Copilot Money pricing).

Quicken Simplifi. Its page shows $3.99 a month billed annually, with $6.99 a month struck through, and says the final price may differ from the monthly price times 12. It lists budgeting and spending tracking with a Spending Plan, investment and retirement portfolio management, and cash flow projections (Quicken Simplifi). The page offers a discounted first year for new memberships ordered directly from Quicken by November 22, 2026, so treat $3.99 as a promotion.

Free Trials and Discount Codes to Use First

Three of the four pages lead with a way to try the app before paying, and each has details that are easy to miss.

• YNAB: its pricing page lists a 34-day free trial with no credit card needed when you sign up directly, and says it sends a reminder email one week before the trial ends and does not charge you unless you subscribe. It adds that third-party app stores may require card details for the trial. College students can get a free 365-day trial by submitting proof of enrollment.
• Monarch: its pricing page lists a code, WELCOME, for 30% off the annual Core plan for one year. It applies to new users only, cannot be combined with other offers, and may expire at any time. The page does not state a trial length, although its headline mentions a first week.
• Copilot Money: its pricing page advertises a one-month free trial and says you can explore the app before connecting any accounts.

Some arithmetic on these offers. If you apply Monarch’s 30% to the listed $99.99 annual price, you pay 99.99 × 0.70 = about $69.99 for the first year, a saving of about $30. That assumes the $99.99 is the Core annual price, which the page does not state outright, so treat $69.99 as a calculation from the listed prices, not a quoted price.

For YNAB, an invented schedule makes the trial easy to plan: start on October 1, and 34 days later is November 4, with the reminder email arriving about a week earlier, around October 28. That gives you a clear date to decide.

In my view, a trial is a low-risk way to answer the most important question, whether you will actually open the app each week, and a trial where you log in a few times is worth more than any feature list.

What Each App Includes Beyond the Price

Beyond price, each page lists features that fit different households and habits. This is a summary of what the pages say, not a ranking.

• YNAB: direct bank import supports select US, Canadian, UK and EU banks, with file-based import elsewhere; real-time syncing across devices; debt tools including an integrated loan calculator; and the option to share one subscription with up to six people at no extra cost.
• Monarch: web plus apps on the Apple App Store and Google Play; the Plus tier adds forecasting, tracking of multiple income sources and deeper investment tools; no ads, and the page says it never sells financial data.
• Copilot Money: automatic transaction tagging that learns your habits, budget rollovers, subscription tracking, net worth tracking across bank accounts, stocks, crypto and real estate, and web, iPhone, iPad and Mac availability.

Sharing changes the math. YNAB’s sharing feature means one $109 annual subscription can cover up to six people. Split six ways, that is 109 ÷ 6 = about $18.17 per person per year, or about $1.51 a month (18.17 ÷ 12). The split is my arithmetic from the listed price and the six-person limit, and it applies only if everyone sharing is a partner or family member, which is how the page describes it.

Putting it together, a household that wants one shared budget can get value from YNAB’s sharing, someone who wants to watch investments alongside spending may like Monarch or Copilot, and a person who is not sure should start with a trial. These pairings are my suggestions from the listed features.

The Yearly Cost Behind Each Monthly Price

Multiplying the listed monthly prices by 12 shows what the annual plans save. YNAB at $14.99 a month is $179.88 a year, so its $109 annual plan saves $70.88. Monarch at $14.99 a month is $179.88 a year against $99.99, a $79.89 difference. Copilot at $13 a month is $156 a year against $95, a $61 difference, which matches the 39% its page advertises. Simplifi at $3.99 a month is $47.88 a year, before any change in the final price. These are arithmetic on the listed prices, and prices change.

Let a chatbot do the multiplication. Copy the price text from an app’s own page, and use this prompt:

Below is pricing text I copied from an app's own pricing page.

[paste the pricing text]

Please:
1. Write the monthly price and the annual price exactly as listed, and quote the line each came from.
2. Multiply the monthly price by 12 and show the multiplication.
3. Subtract the annual price from that total and show the subtraction.
4. If the page lists a promotion, a first-year price, or a trial, quote it separately.

Rules:
- Use only the pasted text.
- If a price is not stated, write "NOT STATED."
- Do not recommend an app or say which plan is best.

With YNAB’s listed $14.99 a month and $109 a year, the output should show 14.99 times 12 equals 179.88, and 179.88 minus 109 equals 70.88, matching the figure above. If your chatbot returns something different, the pasted text or the arithmetic is off, and the arithmetic is quick to redo on a calculator.

Questions Only Your Own Data Can Answer

This section is my opinion. A budgeting app shows the views its designers built, such as spending by category or bills coming up. If you want to know something specific, for example how much you spent on takeout on weekends over six months, a spreadsheet of your own exported transactions can answer it, at the cost of categorizing the transactions yourself. I have not tested whether each app above can answer a question like that, so I am not saying that any of them cannot.

The CFPB tools mentioned above are a free way to start: the Income Tracker, Spending Tracker, Bill Calendar and Budget Worksheet cover all four steps without a subscription (CFPB).

Before You Link Your Bank: The CFPB’s Questions

Every app in this comparison asks to connect to your accounts, so the CFPB: What to consider when sharing your financial data guidance applies to all of them. It says to confirm that only the data you want shared is shared, check whether the company shares data with third parties, find out how often it accesses your accounts and what it stores and for how long, and check whether the service can make payments or move money. To end access, cancel the authorization, ask the service to delete the data it collected where possible, and remember that deleting the app from your phone does not stop your data from being shared; some banks offer tools to see and switch off connected services. It also recommends reviewing bank and card statements regularly, using a separate strong password for each financial account, changing passwords if a company you share data with reports a breach, and submitting a complaint to the CFPB if a problem can’t be resolved with the company.

A Practical Way to Choose

• You want the YNAB Method and a long trial: YNAB’s page lists a 34-day free trial.
• You want forecasting and investment tools: Monarch’s page says its Plus tier adds them.
• You use Apple devices or the web: Copilot Money’s page lists iPhone, iPad, Mac and web.
• You want the lowest listed price: Simplifi’s promotional $3.99 a month billed annually is the lowest on this list.
• You want to ask your own questions, or start free: use the CFPB worksheets or a spreadsheet, as described above.

AI Budgeting Apps Compared: Sources

• CFPB: Budgeting strategies and free tools to help you reach your goals
• California DFPI: Consumer financial education, budgeting
• Monarch Money: When is Mint shutting down?
• YNAB: Pricing
• Monarch: Pricing
• Copilot Money: Pricing
• Quicken Simplifi
• CFPB: What to consider when sharing your financial data

Planning a trip budget too? Our guide to an AI travel budget check applies the same idea of fixed spending limits.

This article offers general information and my own opinions, not financial advice. Prices and features were read from each company’s page in October 2026 (re-read on October 24, 2026) and change often, so the current page is the final word.

No response after interview is one of the most stressful stretches of a job search, and the honest truth is that the silence itself can’t tell you much, so the useful move is to stop decoding it and follow a short plan instead.

You walked out of the room feeling decent, or at least not terrible. You sent a thank-you. Then nothing. A day passes, then three, then a week. You refresh your inbox, reread the last email, and start building theories. Maybe they chose someone else. Maybe the manager hated your answer about the gap. Maybe you said something wrong. Your brain wants a diagnosis, because a diagnosis would feel better than not knowing.

Here’s the part nobody likes to say: from the outside, you usually can’t get that diagnosis. What you can do is a handful of calm, practical things that give you a clearer picture and keep your search moving either way.

This guide covers what the silence can and can’t mean, how to set your own clock, how to write one short check-in note, when to stop, and how to prevent this situation at your next interview. A chatbot can draft the note from your facts if you stop it from guessing. All names, dates, and examples are invented. This is general information, not career advice, and every employer works differently.

What the Silence Can and Can’t Tell You

Think of all the things that could be happening behind the scenes while your inbox stays empty. A few possibilities, in no particular order:

• They’re still interviewing other people.
• Someone who needs to approve the hire is busy or away.
• The recruiter is handling many roles and hasn’t gotten to yours.
• You’re a strong second choice, and they’re waiting on the first.
• The role changed, paused, or was put on hold.
• You weren’t selected, and nobody has sent the message yet.

That last one is real, but it’s only one item on the list. There is no reliable number for how often each of these happens, so I won’t give one. The practical point is that the silence is consistent with good news, bad news, and no news. You can’t tell which from the quiet alone. The Yale SOM Career Development Office says delays often have little to do with the candidate and lists common causes: the team is still interviewing others, is waiting on hiring-team feedback, or the recruiter or hiring manager is out of the office.

So the question to ask isn’t “what does this silence mean?” It’s “what do I know, and what can I do about it?”

The One Fact That Changes Everything

Before you do anything else, look at your notes from the interview. Did anyone say when you’d hear back?

Maybe the interviewer said “we’ll be in touch by the end of next week,” or “we’re deciding in about ten days,” or maybe nothing at all. That one sentence is the most useful piece of information you have. If a date was given and hasn’t passed yet, the silence isn’t a signal. It’s just the process still running. If no date was given, you’re waiting blind, and the plan below is built for that.

If you can’t remember, check your calendar invitation and any emails from the recruiter. Sometimes the timeline sits in a scheduling message you didn’t reread.

Set Your Own Clock

Don’t let the wait float. Put dates on it. Here’s an invented example. You interviewed on Tuesday, March 10, and the interviewer said you’d hear “by the end of next week.” Next week runs Monday, March 16 through Friday, March 20. So the stated deadline is Friday, March 20.

Date	Action
Mar 20	Deadline passes
Mar 23	Send one note
Mar 30	Decide to move on

Why Monday, March 23 for the note? Because a Friday deadline can slip by a business day, and a note sent the day after the weekend feels reasonable rather than anxious. And March 30 is a week later, giving the employer time to answer before you stop spending energy.

If no timeline was given at all, pick your own. For example, count five business days after the interview, send one note, then wait another five before moving on. The exact numbers are your choice. There’s no official standard, and some employers move slowly while others move fast. What matters is that you decide in advance, so you aren’t deciding while anxious. For a sense of how others time it, the Yale SOM Career Development Office calls a check-in email appropriate once more than a week has passed since your interview or last conversation, and says a second follow-up after another week is fine. The Columbia Center for Career Education advises contacting the employer once if you have not heard back by the timeline they gave, restating your interest and asking about the new timeline. The five-business-day count above is my own example.

The One Note Worth Sending

Your note should be short, polite, and easy to answer. It has three jobs: remind them who you are, express continued interest, and ask a simple question about timing. Here’s an invented example.

Hi Priya, I hope you’re well. I wanted to check in on the Operations Coordinator role, since we spoke on March 10 and I understood a decision might come by the end of last week. I’m still very interested and enjoyed learning about the scheduling project. Could you let me know whether the timeline has changed? Thank you for your time.

A few things make it work. It names the date of your conversation so the reader doesn’t have to search. It mentions something specific from the interview, which signals that you remember the conversation. And it ends with a question that can be answered in one line, so replying is easy.

Leave out: apologies for bothering them, hints that you have other offers unless it’s true and relevant, guilt, and long recaps of your qualifications. Our guide to how to follow up after applying covers the earlier stage of the process, and the tone there, short and friendly, carries over here.

Let a Chatbot Draft It From Your Facts

If you’re too tense to write calmly, a chatbot can produce a first version. The danger is that it invents details or adds pressure. This prompt keeps it factual:

I interviewed for a job and haven't heard back. Please draft a short, polite check-in email using only the facts below.

Facts:
- My name: [name]
- Interviewer or recruiter name: [name]
- Role: [title]
- Interview date: [date]
- What they said about timing: [quote, or "nothing was said"]
- One specific topic we discussed that I enjoyed: [topic]

Write 3 to 5 sentences. Include: a reminder of the interview date, continued interest, the specific topic, and one simple question about the timeline.

Rules:
- Use only the facts I gave. Do not invent details, deadlines, or other offers.
- Do not apologize or sound anxious.
- Do not apply pressure or make demands.
- If the facts are missing something you need, list it as "NOT STATED" instead of guessing.

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

Read the draft out loud. If any sentence doesn’t sound like something you’d say, edit it. Short and plain beats polished and stiff.

If There’s Still Nothing

Sometimes the note gets no reply either. That’s frustrating, and it still doesn’t tell you much. Your options at this point are limited, and that’s okay.

You can send one more brief message after another week, or try a different channel if you have one, such as writing to the recruiter instead of the hiring manager. Keep it kind and short. After that, stop.

Then change the status in your tracker. If you use stages, our guide to job search pipeline stages explains how to close an application as “No reply” while leaving the door open in case they write back later. Closing a row isn’t a judgment about yourself. It’s a way to free your attention.

Keep the Rest of Your Search Alive

This may be the most useful piece of advice here. While you wait, don’t pause everything else. A single pending interview can swallow your attention for weeks if you let it. Keep applying, keep scheduling, and keep your other conversations warm. A job application spreadsheet with a next-action column makes this easier, because it shows you what needs attention today, whatever is happening with that one role.

An invented example: you have five open conversations. One has gone quiet. The other four each need a small action this week. If you spend all your energy on the quiet one, the other four stall. If you spend the week on the four, the quiet one either resolves or doesn’t, and you’ve lost nothing.

Keeping the Wait Manageable

Look After Yourself While You Wait

Waiting can wear you down, and that can affect the rest of your search. A few small habits help. Give yourself a time of day to check email, rather than checking all day. Plan something you enjoy that has nothing to do with work. Tell a friend you’re waiting, so you aren’t carrying it alone. If the stress feels heavy or goes on for long, talking to someone you trust, or a professional, is worth it. None of this changes the employer’s timeline, but it changes how you spend it.

Slip-Ups That Make the Wait Harder

• Reading the silence as a verdict. It doesn’t give you one.
• Writing before the stated date. Respect the timeline first.
• Sending long, anxious emails. Keep it to a few lines.
• Following up many times. One or two is plenty.
• Putting everything else on hold. Keep other threads going.
• Forgetting to log it. Note the dates in your tracker.

Never Walk Out Blind Again

The best fix for next time happens in the last two minutes of the interview. Before you leave, ask three simple questions (this list is my suggestion, not taken from a published guide):

• What are the next steps in the process?
• When should I expect to hear from you?
• Who will contact me, and how?

Write the answers down right away. With a date and a name, you’ll know whether silence is part of the plan or a reason to write. Our guide to questions to ask the interviewer goes deeper on what else to ask.

If you forget to ask, it’s not a disaster. You can ask in your thank-you email: “I’d love to know what the next steps look like and when you expect to decide.”

What Waiting Job Seekers Keep Asking

What does no response after interview mean? It can mean many things, including that the process is still running. The silence alone can’t tell you which.

How long should I wait before following up? If they gave you a timeline, wait until it has passed. If not, pick your own, such as five business days, and write one short note.

Should I follow up more than once? One note, and maybe one more after a week, is plenty. After that, move on and keep the door open.

Can AI write my follow-up? Yes, if you give it facts and forbid invention. Read and edit it yourself.

Does silence always mean I didn’t get the job? No. It’s consistent with several outcomes, so don’t treat it as a rejection until you hear one.

Set Your Dates Today

Open your notes, find what they said about timing, and put the deadline, your note date, and your move-on date on your calendar. Handling no response after interview is mostly about replacing guesswork with a plan you can follow even on a hard day.

This article offers general information, not career advice. Hiring timelines vary widely, and this guide can’t tell you what any particular employer’s silence means. The example people, dates, and messages are fictional.

Sources: Yale SOM Career Development Office; Columbia Center for Career Education.

A job application spreadsheet works when it does the remembering for you, which means it needs a status dropdown, two or three date formulas, and one column that tells you what to do today.

It’s easy to start a sheet during a burst of energy, type in the first ten applications, and watch it go stale. In my view the problem isn’t laziness. The sheet is a list, and a list can’t tell you that Thursday is the day to follow up on the Brightwell application. You have to remember to look, and remembering to look is the very thing you were trying to stop doing.

So this guide is a build guide. You’ll set up eight columns, add a dropdown, write three short formulas, add color, and test the whole thing with a fake row before you trust it. It works in Google Sheets and in Excel, though menu names differ a little. A chatbot can help you write the formulas if you give it your column layout and stop it from guessing. The company names, dates, and results in the examples are invented. This is general information, not career advice.

Why a Plain List Stops Working

A list records what happened. A working tracker also answers a question every morning: what needs my attention today?

That second job needs two things a plain list doesn’t have. One is a consistent status, so you can tell an application you just sent from one that has an interview attached. The other is a date that does arithmetic, so the sheet notices when an application has been quiet for a week without you noticing.

None of this needs a special app. A sheet with a handful of formulas does it.

The Eight Columns

Keep it to eight. Every extra column is one more thing you’ll skip filling in.

• A: Company. A short, consistent name.
• B: Role. The title as posted.
• C: Applied. The date you submitted, entered as a real date.
• D: Status. Picked from a dropdown.
• E: Next action. A formula, not typed.
• F: Flag. A formula that says “Follow up” or “Wait.”
• G: Days waiting. A formula.
• H: Link. The posting address.

You type in A, B, C, D, and H. The sheet fills in E, F, and G.

Seven Steps to Build the Sheet

Step One: Type Dates as Real Dates

This sounds trivial, and it’s where most broken sheets go wrong. A date has to be a real date for arithmetic to work. If you type 3/3/2026 and it lines up on the right side of the cell, it’s a real date. If it hugs the left side, the sheet is treating it as text and the formulas below will give you errors or blanks.

If your dates came out as text, retype one and watch how it aligns. Choose one date format and keep it. Don’t mix “3/3” in one row with “March 3” in another.

Step Two: Add the Status Dropdown

A dropdown keeps your wording identical every time, and formulas depend on identical wording. “Applied” and “applied ” with a trailing space are two different things to a formula.

In Google Sheets, select the Status column cells, open the Data menu, choose Data validation, and add a rule. In Excel, look for Data Validation under the Data tab and choose a list. Type your options separated by commas.

A short list is enough to start:

Status
Applied
Talking
Closed

You can split “Talking” into more steps later. Our guide to job search pipeline stages explains how to decide what each one means. For now, three is plenty.

Step Three: The Next Action Formula

In E2, enter this:

=IF(D2="Applied",C2+7,"")

Read it out loud: if the status is Applied, take the applied date and add seven days. Otherwise, leave the cell blank.

The 7 is your choice, not a rule. There’s no official right number of days. Pick a gap you’re comfortable with, and change the number if you want a different rhythm. Our guide on how to follow up after applying goes through the wording and the timing choices.

An invented check: if C2 says March 3, E2 should show March 10. If it shows a strange number like 46091, the cell is formatted as a number. Change the format to a date and it will display correctly.

Copy the formula down the column so every row has it. In most sheets you can drag the corner of the cell, or double-click it.

Step Four: The Flag Column

In F2, enter:

=IF(E2="","",IF(TODAY()>=E2,"Follow up","Wait"))

This says: if there’s no next action date, show nothing. Otherwise, if today has reached that date, say “Follow up.” If not, say “Wait.”

TODAY() updates itself whenever you open the sheet. That’s what makes it work like an alarm. You don’t need to do anything.

Step Five: Days Waiting

In G2:

=IF(D2="Applied",TODAY()-C2,"")

If the status is Applied, subtract the applied date from today. The result is the number of days you’ve been waiting. An invented example: applied March 3, today is March 12, so G2 shows 9. This number is useful for spotting the ones that have gone quiet for a long time, and it feeds directly into the archive rule in the stages guide.

Some regional settings use a semicolon instead of a comma inside formulas. If a formula gives an error and looks right, try swapping the commas for semicolons.

Step Six: Add Color So You See It Instantly

You want the Follow up rows to jump out. Use conditional formatting. Select the rows you want to color, open the conditional formatting menu, and choose a custom formula rule. Enter:

=$F2="Follow up"

Pick a light yellow fill. Now any row whose flag says Follow up turns yellow on its own. You can add a second rule for Closed rows, such as gray text, so finished applications fade into the background.

Step Seven: Test With a Fake Row

Before you trust anything, enter a row you know the answer to. An invented test:

Field	Value
Applied	8 days ago
Status	Applied
Expected flag	Follow up

If the flag says Follow up and the row turns yellow, your formulas work. Then change the status to Closed. The next action date and the flag should go blank. If they don’t, check that your dropdown text matches the formula exactly.

Delete the fake row afterward. Skipping this test is the most common reason a person trusts a sheet that has been quietly broken for weeks.

Let a Chatbot Write the Formulas

If your layout differs from mine, a chatbot can adapt the formulas. The risk is that it guesses column letters or assumes a tool you aren’t using. This prompt prevents that:

I'm building a job application spreadsheet in [Google Sheets / Excel]. My columns are:
A: Company, B: Role, C: Applied (date), D: Status (dropdown: Applied, Talking, Closed), E: Next action, F: Flag, G: Days waiting, H: Link.

Please write:
1. A formula for E2 that adds [7] days to C2 only when D2 is "Applied", and is blank otherwise.
2. A formula for F2 that shows "Follow up" when today has reached E2, "Wait" when it hasn't, and blank when E2 is blank.
3. A formula for G2 that shows days since C2 only when D2 is "Applied".
4. A conditional formatting rule that colors a row when F is "Follow up".

For each formula, write one sentence explaining it, then give a test row with the expected result.

Rules:
- Use only the columns I listed.
- If any menu name or function differs between my tool and others, say so instead of guessing.
- Do not add columns or features I didn't ask for.
- If anything is unclear, ask before writing the formula.

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

Always run the test row it gives you. If the result doesn’t match the expected answer, the formula is wrong, even if it looks plausible.

Keeping the Sheet Useful After Week One

Sort by the Date, Not by When You Applied

Once the formulas work, the most useful habit is a single sort. Sort by the Next action column, earliest first. The applications that need you are at the top, and the finished ones sink to the bottom. Some sheets also let you set up a filter so you can hide Closed rows altogether.

If you keep one view only, keep this one. It turns the sheet from a record into a to-do list.

When to Add a Column

Resist the urge for a while. After a few weeks, if you notice a real gap, add one. A “Response date” column can help if you want to compare later. A “Resume version” column helps when you try different versions, which our guide on the resume A/B test explains. Both are worth adding only when you’ll actually use them.

If you want to measure effort and replies, the job application response rate guide shows how to add a minutes column and calculate from it.

If You Are Collecting Unemployment, This Sheet Can Double as Your Log

The U.S. Department of Labor, Training and Employment Notice 17-19 says federal law requires states to have a work-search requirement for unemployment claimants, and many states require you to keep a log of your search activity. Your Company, Role, Applied and Link columns already hold most of what a log asks for, so add the contact method if your state asks for it. Requirements differ by state, so use the log format your state’s unemployment agency specifies when it gives one.

Why a Quiet Application Is Not Always a No

The Yale SOM Career Development Office says delays are often unrelated to the candidate. That is why the sheet marks a row “Follow up” instead of “Rejected” when the waiting days pass.

Where These Sheets Usually Break

• Dates typed as text. Check the alignment.
• Dropdown wording that doesn’t match the formula. Copy it exactly.
• Formulas typed only in the first row. Fill them down.
• Skipping the fake-row test. Always run one.
• Too many columns. Eight is enough to start.
• Never opening it. Put a weekly time in your calendar.

Questions About Job Application Spreadsheet

What should a job application spreadsheet include? Company, role, applied date, status, a next action date, a flag, days waiting, and a link. The dates and flags can be formulas.

Does this work in both Google Sheets and Excel? The formulas are the same in both. Menu names for dropdowns and conditional formatting differ slightly.

How many days should I wait before following up? There’s no official number. Choose a gap you’re comfortable with and change the 7 in the formula to match.

Can AI write the formulas? Yes, if you give it your exact columns and ask for a test row. Run the test yourself before you trust it.

Do I need a paid tool instead? Not to get started. A simple sheet is enough to find out what you actually need.

Build It Tonight, Test It Once

Set up the eight columns, add the dropdown and three formulas, and run one fake row. A job application spreadsheet that flags its own follow-ups is worth far more than a longer list you have to remember to read.

This article offers general information, not career advice. Menu names and features can differ between spreadsheet tools and versions. The example companies, dates, and results are fictional.

Sources:

• Google Docs Editors Help: Create an in-cell dropdown list
• U.S. Department of Labor, Training and Employment Notice 17-19
• Yale SOM Career Development Office