The Hours-Saved Number Is the One They Test First

Where a Number Actually Comes From

Résumé builders push a number into every bullet, and AI time savings are the easiest number to produce, because nobody but you saw the before. That is also why it is the least checkable figure on the page, and the one an interview question is likeliest to test first. Use a number only when something outside your own estimate produced it: a system that logged throughput, a process step that visibly no longer exists, an error rate someone else tracked.

When a generative assistant was rolled out to 5,179 customer support agents at one firm, the company's own logs showed issues resolved per hour rising 14% on average, and 34% for novice and low-skilled agents ¹. That is a real, sourced number, but it belongs to that company's own dataset for that specific job. Copying its shape onto a different role at a different employer, without your own system behind it, is exactly the move a follow-up question is built to catch.

The distinction is not about honesty in the sense of lying versus telling the truth. Most people writing an inflated AI bullet believe it. The distinction is between a number a system produced and a number your own impression produced, and only the first kind survives someone asking a second question about it.

This matters more for an AI claim specifically than for most other résumé metrics, because the tooling is new enough that almost nobody has a settled sense of what a normal gain even looks like yet. A sales number or a revenue figure gets compared against industry norms an interviewer already knows. An AI productivity claim gets compared against nothing but your own explanation, which is exactly why that explanation has to hold up.

When the Gain Was Real But Only You Saw It

Plenty of AI use genuinely helped and left no outside record at all. Describe the change to the work instead of inventing a percentage: the specific task that stopped being done by hand, the report that used to take a day and now gets a usable first draft in an hour, the step you cut out of a process entirely. A sentence like that survives a follow-up question a manufactured percentage cannot.

The honest baseline for self-reported gains is smaller than most résumé bullets claim. In a nationally representative 2024 U.S. survey, people who used generative AI at work reported saving a mean of 5.4% of their work hours, which works out to 1.4% across all workers once non-users are included ². A bullet promising a round, memorable figure sits well outside that measured range, and a reader who has seen the research behind it will notice the gap.

That survey measured self-reported time, not output someone can verify, and the researchers themselves call it a rough estimate rather than a hard number. It is not a ceiling on what any individual can achieve. It is a reminder that the plausible range for an honest, unmeasured claim is a lot narrower than the round figures career-advice pages tend to suggest, and a bullet that overshoots that range invites exactly the question it cannot answer.

Test the Number Before You Write It

Ask who besides you would confirm the figure, and whether it describes a measured outcome or a private guess. Self-reported speed estimates are not reliable even from people whose job depends on precision, and the direction of the error is what makes it worth checking before you write anything down.

In a randomized trial, sixteen experienced developers forecast a 24% speedup from AI tools before starting, still believed afterward that it had saved them roughly 20%, and were in fact measured at 19% slower on the real tasks ³. It is a small, specific study, sixteen developers on codebases they already knew, and it does not prove every self-estimate is wrong. What it establishes is the shape of the risk: a person's felt sense of how much faster AI made them can diverge sharply, and even run backward, from what actually happened.

Verifying a claim like this takes a recruiter about ten minutes, and what those ten minutes ask for is the trail the work left behind: which tool on which task, an artifact daily use produced, an output you rejected and why. A candidate with that trail spends the conversation talking about the work. A candidate without it spends it explaining the gap between the résumé and the truth.

A Believable Bullet Beats an Impressive One

Several famous productivity figures in career advice come from one narrow setting, not from ordinary daily work. The widely quoted '55% faster with AI' number traces to a single trial where developers wrote one small, self-contained web server with a known right answer, not to years of maintaining a real codebase ⁴.

Even inside that one trial, the study's own 95% confidence interval ran from 21% to 89% faster, so quoting 55% as a precise figure overstates what was actually shown. Borrowing that number, or one that sounds like it, for a résumé attaches someone else's narrow benchmark to your name, and it does not describe the maintenance work most jobs actually involve.

The gap between that number and everyday work matters because it is the kind of thing an interviewer who has read even one article on this topic already knows. A candidate who quotes a headline percentage as if it were their own measured result reads as someone repeating something they saw, not someone reporting on their own work, and the two land very differently once a follow-up question arrives.

The same discipline applies to naming a tool at all: a specific, defensible line beats an impressive one nobody, including you, can actually source under a question.

See a sample report >