We Showed One CV With a 14 Month Gap to 3 AI Screeners, 45 Times. The Gap Was Never the Problem.

If you have a gap in your work history, you have already read the advice. Hide it. Use years instead of months. Switch to a skills based layout so nobody can follow the dates. Whatever you do, do not draw attention to it, because the software will throw you out before a person ever sees your name.

It is repeated everywhere, it is stated with total confidence, and as far as we can tell nobody has actually tested it. So we did.

We built one candidate with one career history. Then we produced five versions of her CV that were identical in every respect except how they handled a 14 month gap, and we put each version in front of three different tiers of AI screener, three times each, with the same neutral screening brief. Forty five scored runs. We never mentioned the word gap in the instructions. We simply asked each model to screen the candidate and tell us what concerned it.

The short version: not one screener rejected her. The strong models all spotted the gap and wrote it down, the cheapest model missed it completely, and the single most effective thing we did was stop hiding it. Labelling the gap in one dated line raised her average score from 78.3 to 85.0 and made the phrase unexplained employment gap disappear from the concerns list entirely. The gap was never what hurt her. The silence was.

What we actually did, and why the boring bit matters

Most comparisons of this kind have a flaw that quietly ruins them. Someone writes a CV with a gap, then writes another CV that handles the gap differently, and compares the outcomes. But the two documents also have different wording, different achievements and different spacing, so when the results differ you cannot tell whether the gap treatment caused it or the writing did.

We fixed that by starting from one fixed career history and never touching it again. Maria Okonkwo is an operations manager in Manchester with nine years in logistics. Same employers in every version. Same three roles. Same bullets, down to the character. Same figures, including a 23% cut in fulfilment cost per order, GBP 412k saved over 18 months, and voluntary turnover pulled from 31% down to 12%. Same education. Same skills list.

The only thing that moved between versions was the 14 month stretch between June 2024 and September 2025. Five treatments:

Each version went to three model tiers, a frontier model, a mid tier and a budget tier, three runs each at temperature zero. We asked every one of them to act as a screener for an operations manager role, extract the employment history, report any gap it found, list anything that gave it pause, score the candidate from 0 to 100, and decide advance or reject.

Finding one: nobody was rejected. Not once.

Forty five runs. Five versions. Three tiers of model. Every single result came back advance.

That includes version A, where a 14 month hole sits in the middle of the timeline with no explanation anywhere on the page. Both the frontier and the mid tier models found it, named the exact dates, and wrote it into their concerns. Then they advanced her anyway, because her achievements were strong and the rest of the history was solid.

This is worth sitting with, because it contradicts the most repeated claim in this entire subject area. The screeners in our test behaved less like a filter and more like a reader taking notes. A gap became a question mark next to her name, not a trapdoor under it.

Finding two: the good models see it, and they say so

Detection split cleanly along price. Here is how often each version was flagged as containing an employment gap, out of nine runs.

When the dates carry months and nothing explains the space between them, the two capable models caught it every time and quoted the period back to us. The frontier model wrote unexplained 14 month employment gap. The mid tier wrote employment gap between June 2024 and September 2025. Neither needed prompting, neither was told to look for one, and neither missed it.

Assume the gap is visible. Plan from there.

Finding three: one line moved the score more than any layout trick

This is the result that should change what you do this afternoon.

Version B is version A plus one dated entry, sitting in the timeline where the gap was, naming the reason and noting a certificate completed during the period. Nothing else changed. Not a word of the roles, not a single figure.

The average score went from 78.3 to 85.0. The mid tier model went from 75 to 90. And the concerns list, which is the part that follows you into the interview, changed completely.

The mid tier model went further. Screening the labelled version, it returned an empty concerns list. Nothing at all. Same candidate it had flagged a moment earlier.

Read that again, because the implication is genuinely useful. The models were not judging whether caregiving is a good enough reason to stop working. They were checking whether the period was accounted for. An explained gap is a closed question. An unexplained gap is an open one, and open questions are what get a CV set aside in favour of the one without any.

Finding four: the years only trick works, on the readers who matter least

Version C is what most advice tells you to do. Strip the months, show only years, and the gap between 2024 and 2025 stops looking like anything at all.

It worked more often than not. Seven of nine readers saw no gap, and it produced the joint highest average score in the test at 88.3.

But look at where it failed. The frontier model still caught it in two of three runs and wrote down an unexplained gap between 2024 and 2025. So the trick is not invisibility. It is invisibility to weaker readers, and you do not get to choose your reader. Employers pick their own tooling, almost none of them publish what it is, and cost pressure on high volume screening pushes plenty of them toward exactly the cheap tier that missed everything.

There is a second cost that our scoring could not capture. Years only dates hide more than the gap. They also hide that she held her last role for five years, which is a genuine strength in an operations career. Blur the timeline and you blur the good parts along with the awkward one. And any recruiter who has been doing this for more than a month reads years only dates as a signal to go looking, which puts you back where you started with less credibility.

Finding five: the cheapest reader is the worst reader, and that cuts both ways

The budget model never once detected a gap on its own. Not in version A, where 14 months of nothing sits in plain sight with month level dates on both sides. Zero for three.

It did notice the gap in version B, but only because it was written down as an entry it could read. In other words the cheap model does not analyse your timeline, it reads your list. If you spell something out it knows about it, and if you do not, it does not go looking.

People sometimes take this as good news. It is not, particularly. It means the weakest system is also the one least able to recognise anything clever you have done, and it flattens the difference between a strong candidate and an adequate one. The budget tier scored her 75 in four of the five versions regardless of what we did. Careful presentation earns you the most from the readers that are actually paying attention.

The part you came for: what to write, for nine real situations

Every version we tested that worked had the same ingredient. A dated line, in the timeline, saying what the period was. Here is the wording for the situations people actually have. Keep them short. One line does the job, two if you genuinely used the time.

1. Redundancy or layoff

Write: “Role made redundant in a company wide restructure, June 2024.” Put it as the final bullet under the role that ended, not as a separate entry.

This is the easiest one to say plainly, and the one people most often bury. Redundancy describes a decision a company made about headcount. It carries no implication about your work at all, and every recruiter reading in 2026 has seen hundreds of them. Vagueness here costs you far more than the fact ever could.

2. You were dismissed

Write: “Left [Company] in June 2024. Happy to talk through the circumstances.”

Do not write the word fired on your CV, and do not invent a redundancy that did not happen, because references exist. A neutral dated line closes the timeline without arguing your case on paper. You cannot win that argument in one line anyway, and the CV only has to earn you the conversation where you can.

3. Caring for family

Write: “Career break, full time caregiving for a family member, June 2024 to September 2025.”

This is the exact line we tested, and it took the score from 78.3 to 85.0. It needs no apology and no detail. If you kept a hand in with a course or a certificate, add it as a second line, because it shows the timeline never really went quiet.

4. Health

Write: “Career break for personal health reasons, now fully resolved and available for full time work, 2024 to 2025.”

You are not obliged to disclose a medical reason and you should not feel pushed into it. The screener is checking whether the period is accounted for and closed, not auditing your diagnosis. The phrase now resolved is the part doing the work, because it answers the only question a hiring manager is actually allowed to care about.

5. Parental leave

Write: “Parental leave, March 2024 to September 2025.”

One line, dated, no framing needed. If you returned to the same employer afterwards, you may not need it at all, because there is no gap in your history to explain. It matters when the break sits between two different jobs.

6. Study or retraining

Write: “Full time study, [qualification], [institution], 2024 to 2025.”

Put this in the experience section rather than only under education. A qualification listed at the bottom of the page does not visibly cover a hole in the middle of it, and the point of the line is to close the timeline where the timeline is.

7. Sabbatical or travel

Write: “Planned career break for travel, June 2024 to September 2025. Returned as planned.”

Those last three words matter more than they look. They turn a stretch of absence into something you decided and finished, which is a very different thing from something that happened to you.

8. A business that did not work out

Write: “Founder, [Business name], 2024 to 2025. Built and ran [what it did]. Closed the business in 2025.”

Treat it as the job it was, because it was one. Give it a dated entry and one real outcome if you have one, revenue, customers, whatever is true. Founders often hide a venture that closed, which is a shame, because running something end to end is exactly the experience an operations or management role is looking for.

9. You have been looking for a while

Write: “Actively seeking a new operations role since June 2024. Completed [certificate or course] during this period.”

This is the hardest one, and the honest advice is to give the period a shape. A long search with nothing attached to it reads as time passing. A long search with a course, some freelance work or a volunteer commitment attached reads as someone who kept moving. If you have been out a while and have nothing to point at, that is a reason to go and get one small thing, not a reason to leave the space blank.

Where the line goes on the page

Position matters more than people expect, and it follows directly from how the models read. The strong screeners built a chronology and then looked for holes in it. So the explanation has to live inside the chronology, not in a footnote somewhere else.

Put it in your experience section, in date order, exactly where the gap falls. Give it real dates in the same format as everything around it. Keep it to one line, or two if you genuinely have something to add. Do not put it in your cover letter and nowhere else, because plenty of screening never sees the cover letter. Do not put it in a personal statement at the top, where it becomes the first thing anyone reads about you. And do not hide it in a notes section at the bottom, where it does not close the hole it needs to close.

If you are working in our CV builder, the simplest approach is to add it as an experience entry with the dates of the break and the reason as the role title. It renders as part of the timeline, which is exactly where you want a screener to meet it.

Where the humans and the machines disagree

A screening model and a recruiter are not looking for the same thing, and the difference is worth understanding before you act on any of this.

The models in our test were consistent, literal and completely incurious. They checked whether the timeline was accounted for. Once a dated line existed, the question closed and they moved on. None of them evaluated whether caregiving was a reasonable thing to take 14 months for, because that is not a judgement they were making.

A person is different. Recruiters we have spoken to over the years describe gaps the same way almost word for word. A gap is not a problem, an unexplained gap is a problem, and a defensive explanation is a bigger problem than either. That is the one place where a human will punish something a machine ignored. Write three sentences justifying your break and you have signalled that you think it needs three sentences of defence. Write one factual line and you have signalled that it is a normal part of a working life, which it is.

The second difference is memory. A model reads your CV in isolation, scores it, and forgets you. A recruiter is comparing you to the four people they read before lunch. If two candidates are close and one has an open question on the page, the open question is an easy way to break the tie, and nobody ever tells you that is what happened. This is the practical reason the concerns list from our test matters more than the score. The score got her through. The concern is what would have followed her into the room.

Saying it out loud

The line on your CV buys you the conversation. It does not have the conversation for you, so it is worth knowing what you are going to say when someone asks, because they will.

The structure that works is short and it has three parts. What happened, in one sentence and without apology. That it is finished. What you are ready for now. Something like: my father needed full time care from the middle of 2024, I took the time to do that properly, and that came to an end last September. I used part of it to finish my CILT Level 3. I have been back at full capacity since then and I am looking for the operations role I would have been going for anyway.

Notice what is missing from that. No apology, no over explaining, no long story about the circumstances, and no hint that the person is expected to feel awkward about having asked. The reason it works is the same reason the labelled CV outscored the raw one. It closes the question rather than opening a discussion.

One more thing that people get wrong here. If your gap ended a while ago and you have been back at work since, the gap is no longer the headline of your career and you should stop treating it as one. Maria has been back in a role since September 2025. Her most recent achievement is a 23% cut in fulfilment cost, not a career break. Lead with the work.

What this test does not tell you

We would rather flag the limits than let you over read the numbers.

All three models were from one family, spanning a frontier tier down to a budget tier. We did not test the parsers inside Workday, Greenhouse or Lever, because no outside party has access to those. Treat this as evidence about how language models read a CV, not as a measurement of any particular employer system.

We used one candidate, one industry and one gap length. A 14 month gap in UK logistics is not a three year gap in investment banking, and a strong track record either side of the break is doing work here that a thinner history would not do for you.

We measured a screening decision, not a hiring one. No model in this test had to sit across a table and ask why. And scores are directional. A move from 78.3 to 85.0 is a real and repeatable shift in how a reader described the same person, not a claim that anyone becomes 6.7% more employable.

Finally, every version advanced, which means our scenario never found the boundary where a gap actually costs someone the shortlist. That boundary exists. We just did not reach it with a strong candidate and 14 months.

So what should you actually do

Assume the gap is visible, because every capable reader in our test found it without being asked.

Then write the line. One dated entry in your experience section naming the period for what it was. It was the highest value change we measured, it takes a minute, it needs no redesign, and unlike the hiding tricks it also survives the human who reads you afterwards.

Keep your month and year dates. Blurring them to years only bought a few points from weak readers, failed against the strong one, and quietly hid a five year tenure that was worth showing.

Fill the period honestly if you can, with freelance work, a certificate, a volunteer commitment, something real that had dates. And leave the functional layout alone, whatever it does to a machine, because the person who reads next is the one who decides.

The finding underneath all 45 runs is simpler than the advice industry built around it. The machines were not looking for a reason to bin her. They were looking for an account of her time. Give them one, in a single line, and the question closes.

If you want to see how your own CV reads to a screener before you send it, our ATS simulator shows you what gets extracted, and our file format experiment covers what survives the trip through different document types. For the wider picture of how these systems make decisions, how to beat AI resume screeners pulls the whole process apart.

Frequently asked questions