A Five-Year Career Break Costs Tool Exposure, Not Judgment

What did five years away actually take?

Five years away took one of two things, and it was not the one worth pricing. Tool exposure went: which assistant a team runs, where it fails, what a working prompt looks like this quarter. The judgment the job actually runs on stayed. Framing a problem before generating, demanding a source for the claim that carries the decision, refusing an output nobody can check: all of it was built in the years before the break.

On the resume it is one line: 2021 to 2026, with no reason attached.

The exposure half is younger than it looks, including on the team doing the hiring.

In nationally representative US surveys fielded in August and November 2024, 23% of employed respondents had used generative AI for work at least once in the previous week, and 9% used it every workday ³. The two surveys drew 5,014 and 5,329 responses.

Both numbers are self-reported, and both are already dated. A level like that moves within months. Whatever the current team knows about these tools, most of it arrived after the candidate's break had begun.

The screen never makes the split.

Harvard Business School and Accenture surveyed 2,275 senior leaders across Germany, the UK and the US in early 2020. Among the employers whose recruiting software filters or ranks candidates, 48% said they filtered middle-skills candidates on employment gaps of more than six months. The report is plain about the mechanism: such a filter cannot infer what caused the gap ¹.

The penalty does not keep tracking length either.

Kory Kroft, Fabian Lange and Matthew Notowidigdo sent roughly 12,000 fictitious resumes to about 3,000 real online postings across the 100 largest US metropolitan areas. They randomized the current unemployment spell between one and 36 months. Callbacks fell from roughly 7% at one month to roughly 4% at eight, and after eight months additional time made almost no further difference ².

The experiment ran on a single job board around 2011 and 2012, in clerical, customer service and sales roles, so the rates belong to that market rather than to yours. The shape is the part worth keeping, within the one to 36 months it tested: a discount that arrives early and then stops learning anything about the person.

Test the judgment the break could not touch

Hand over work and watch what gets refused. A returning candidate cannot tell you which assistant the team standardized on last spring, and asking is a trivia question. Give them a short model-written output with a real defect in it, a confident claim resting on nothing, and ask which line they would not send. The answer is what the job is buying.

Nothing about the exercise is special to a returner, which is the point. The material is a document with something wrong in it, and the question is what they found and what they did next. The identical brief goes to the candidate who never left a desk.

Hiring for someone who can tell when AI is wrong is a general problem, and a five-year gap does not change the instrument.

Absence of exposure is also the cheaper half to fix.

A conversational assistant went out in a staggered rollout to 5,179 customer support agents at one software firm. Access raised issues resolved per hour by 14% on average: 34% for novice and low-skilled agents, and close to nothing for experienced ones ⁴.

The measure was issues resolved per hour on problems that have correct answers, at one firm, in one occupation, with an assistant trained on that firm's own transcripts.

Read narrowly, the result still says this much: the lift ran largest for novice and low-skilled agents, so the gain from access did not wait for experience.

A break is not a refusal, and the two get scored alike by accident. A candidate who says they do not use AI at all has taken a position. Whether the refusal is a dealbreaker if your best candidate doesn't use AI turns on what they say when pressed.

Somebody who had no access for five years took no position at all. Pressing them for one produces an answer about the last twenty minutes of thinking.

Should tool familiarity be a shortlist bar, or a first-week line?

A bar, but only where the job cannot wait. Tool familiarity carries real weight in a few roles: a support desk running one fixed internal assistant, an engineering team whose code review assumes everyone is on the same coding assistant. Everywhere else it is an orientation line, and the honest place for one is the first week rather than the shortlist.

If it is a bar, run the orientation for everybody. Fifteen minutes before the exercise, the same fifteen minutes: here is the assistant, here is the login, here is what it does badly. A candidate who needs it never has to ask, or explain why.

A walkthrough everybody gets marks nobody.

The posting carries the same problem one stage earlier. A requirement written as two years of hands-on AI experience excludes people who did the work somewhere the tools were banned, and it excludes a returner by construction. Writing "or equivalent" into an AI requirement, and saying what counts as equivalent, reaches the problem before anyone reads a resume.

Where does an unwritten gap discount land?

Close to characteristics the law already names, and in a place no record reaches. A multi-year absence has a reason, and three reasons sit close to protected ground: caregiving, disability or illness, and military service. A discount applied in a screener's head at the resume stage, with nothing written down, produces a selection rate on those characteristics and no reason attached to any decision.

An informal screen is a selection procedure. The Uniform Guidelines, a 1978 US federal regulation, define one as any measure, combination of measures, or procedure used as a basis for any employment decision. The definition reaches the full range of assessment techniques, through "informal or casual interviews and unscored application forms" ⁵.

Deciding in your head does not move a decision outside it.

Caregiving is the one easiest to get backwards. The Equal Employment Opportunity Commission's 2007 enforcement guidance on workers with caregiving responsibilities says the federal EEO laws "do not prohibit discrimination against caregivers per se" ⁶. The guidance then sets out circumstances where the treatment might be unlawful anyway, because the decision ran on a protected characteristic rather than on the caregiving: a sex-based stereotype, race or national origin, or the disability of someone the worker cares for.

Military service works the other way.

USERRA, a 1994 federal law, says a person who has performed service in a uniformed service "shall not be denied initial employment, reemployment, retention in employment, promotion, or any benefit of employment by an employer" on the basis of that service ⁷.

Where the gap was that service, it is the protected thing itself. Confirm current requirements with employment counsel; none of this is legal advice.

Counting is the only way to see a discount nobody wrote down. Compute the advance rate for applications carrying a gap and the rate for those without, stage by stage, and keep the individual records rather than only the rate. Before anyone quotes 80% as a pass mark, read why the four-fifths rule is a trigger for scrutiny, not the legal threshold.

The same five years that dated the tools may have dated the judgment. What counts as a claim worth checking has moved. So have the cheap mistakes. A person can hold excellent verification instincts and still not know where the new failures hide.

Write the criterion beside the date range

Next to the years on the resume, write the criterion the round applied and what met or missed it. Two lines is enough: what the role needs a person to be able to do with a model, and what this candidate did when handed one. A gap that carries a reason somebody can argue with stops being a gap and starts being a note.

The note takes a sentence. "Asked to review a model-drafted brief; found the unsourced figure, missed the stale date, named what they would not send without checking."

A rejection built on that sentence can be read back to the candidate, a panel, or a lawyer, and it still says why.

Stage-level rates can be rebuilt from notes later. Notes cannot be rebuilt from a rate.

Criteria written beside decisions are all anyone will hold when a rejected applicant asks what happened, and they cost one field in the tracking system already running.

The date range stays 2021 to 2026. What the record says beside it is the part still open.

Read the evidence >