Screening
What to Say When You're Wrongly Accused of Using AI
If an employer says AI wrote your application or take-home, ask what the finding was and what tool produced it, in writing, before you defend anything. Accusations like this often rest on nothing more specific than an impression, and independent testing has found no detector reliable enough to support one. Offer process instead of protest: version history, timestamped drafts, a live walkthrough of how you built the work. Skip anything that looks like retrofitting evidence after the fact. Some employers won't reopen a decision, and that's worth knowing going in.
The takeThe instinct to defend your writing style is understandable and mostly beside the point. No detector that independent researchers have tested is accurate enough to carry an accusation on its own, which means the claim itself is often resting on something the employer can't actually produce. Asking them to name it does more work than any explanation of your process, because it puts the weight back where the evidence should live. If they can't name a tool and a score, say so plainly and ask what happens next.
Where Olive fits
Open a role and see what the work shows
No screen can tell which resume or take-home a model helped write, and Olive doesn't try to. If an employer sends you an Olive assessment, it's a work sample completed openly with an AI assistant, and the report describing what happened is a person's account you receive free, the same copy the employer reads.
Rank your shortlistAsk What the Finding Actually Was
Before you explain anything, find out what produced the accusation: a named detector and a specific score, or an impression with neither behind it. Put the question in writing and ask for both, and ask when the finding was made and by whom, since a hiring manager's own read and a tool's output deserve different responses.
The arithmetic is worth knowing before you ask. One university ran the numbers on its own detector's vendor-claimed 1 percent false-positive rate against 75,000 papers submitted in a single year and got roughly 750 papers that could have been wrongly flagged, then turned the tool off rather than defend that math 1. A hiring process running any volume of applications through a similar tool is doing the same arithmetic, usually without publishing it.
Two outcomes follow from asking. If the employer names a tool and a number, you now have something concrete to respond to: what the tool measures, and what its own publisher says about relying on it. If they can't name either, say so plainly, in writing, and ask what happens next. An accusation that can't be specified isn't one you can be expected to disprove.
What Counts as Process Evidence
Offer process, not protest. Version history in your document editor, timestamped drafts, an outline that predates the final version, or a short walkthrough call where you talk through how you built the piece all do more than any statement defending your writing style. They show work existing over time, which nothing about the accusation actually contradicts.
Keeping a running record of your AI-assisted work as you go is the cheapest version of this insurance, done before you ever need it: a folder of drafts, a note of which sections you wrote first and which you refined with a tool, dated as you go.
A live re-do of a comparable task is the strongest version of this and worth proposing yourself if the employer seems open to it. Fifteen minutes on a related problem, talked through out loud, settles more than a paragraph of explanation ever will. None of this requires proving where every sentence came from. It only requires showing the work has a history a fabricated submission wouldn't have.
Bring whatever you already happen to have on hand before you build anything new for the occasion. A messy folder of half-finished drafts you wrote before applying is more convincing than a clean one assembled the night the accusation arrived, precisely because the mess is the kind of thing nobody fakes on purpose.
Don't Retrofit Evidence, and Don't Reach for a Humanizer
Two moves turn a recoverable misunderstanding into something worse. The first is manufacturing evidence after the fact: backdating a document, or writing a new draft and claiming it's the original. The second is running the finished piece through an AI-humanizer tool once you're accused, hoping a rerun would come back clean. Both spend effort chasing an unreliable target rather than answering the actual question.
The target really is unreliable, not just inconvenient. Independent testing of the tools an employer might be running concluded they were neither accurate nor reliable, so there's no stable score worth chasing by rewriting anything 2. Vendors selling these tools say the same about their own limits. One states plainly that its results shouldn't be used to punish students, and that cases where human writing gets classified as AI always exist 3. Another, a detection vendor used across academic settings, has said on its own blog that a score is not a determination of misconduct, and that a person still has to apply judgment to it 4. Both statements exist because both companies already know their tools get this wrong on their own account, which means you're allowed to hold an employer's decision to the same standard the vendor already stated in public.
Employers face the mirror version of this question. Whether a detector that flags half a batch of intern applications justifies rejecting them is answered no on the employer side, for exactly this reason, and the same standard should hold when the flag is pointed at you specifically rather than at a pile of resumes nobody looked at closely.
Know the Honest Ceiling
A pile of documents can carry a real signal even when no single one in it can be judged. One analysis of conference peer reviews estimated substantial AI-modified text across thousands of submissions while noting that the same pattern may be too subtle to read off any individual document 5. That's the honest shape of what a detector, or a hunch, can tell an employer: something about a pile, nothing reliable about you specifically.
Say that plainly if it helps, but don't expect it to always win. Some employers won't reopen a decision once it's made, whatever you send them, and no amount of process evidence changes that in every case. A rejection you can't get reopened isn't proof they were right; it's just a door that stayed shut, and treating it as a verdict on your integrity gives the moment more authority than the evidence behind it ever earned.
What you can control is leaving a clean, honest account of your work on the record, in writing, in case anyone looks again later, and moving on to the next application without spending more of your week defending one that's already closed. Keep the correspondence. Keep the drafts, even the messy ones. Then let it go and put the same energy into an application where the answer isn't already decided before you've had a chance to make your case.
Common questions
Should I ask the employer directly what tool flagged me?
Yes, and ask in writing. Naming the tool and the score is a low-cost request for the employer to fulfill if the finding is real, and their inability or refusal to answer tells you almost as much as an answer would.
Is it worth hiring someone to write a rebuttal letter?
Usually not before you've tried the direct route. A short, specific written request for the evidence, followed by concrete process documentation, does the same work at no cost and doesn't read as adversarial before you know what you're responding to.
What if the accusation came up in a live interview instead of after a submission?
The same principle applies: ask what specifically prompted the question rather than launching into a defense. A vague 'you sound rehearsed' is different from a specific claim about a document, and each deserves a different answer.
Can I sue if I was rejected over a false AI accusation?
That depends heavily on where you live and what was actually said or written, and it's a question for an employment attorney rather than a general answer. Keeping your process evidence and the employer's correspondence preserves your options either way.
Does it help to explain my writing process unprompted, before anyone asks?
Only if you already suspect scrutiny is coming, and even then briefly. Volunteering a long defense before any accusation exists can read as anxious rather than reassuring; the useful moment for process evidence is once a specific concern has been named.
References
- 1. Guidance on AI Detection and Why We're Disabling Turnitin's AI Detector vanderbilt.edu Supports the arithmetic: a 1% vendor-claimed false-positive rate against 75,000 papers implies roughly 750 wrongly flagged.
- 2. Testing of Detection Tools for AI-Generated Text arxiv.org Supports that no independently tested detector was accurate or reliable, so there's no stable score to chase by rewriting.
- 3. Frequently asked questions - What are the limitations of the classifier? gptzero.me Supports that the vendor itself says results should not be used to punish anyone and human writing is misclassified.
- 4. Understanding false positives within our AI writing detection capabilities turnitin.com Supports that the vendor states a score is not a determination of misconduct and needs human judgment applied.
- 5. Monitoring AI-Modified Content at Scale: A Case Study on the Impact of ChatGPT on AI Conference Peer Reviews arxiv.org Supports that a corpus-level estimate of AI use does not translate into a verdict on any individual document.
5 sources, numbered by first appearance. How Olive sources claims
General guidance for hiring teams. What works at one company and one volume may not transfer to yours.
Olive assesses how a person works with AI. It does not detect AI-written documents, and it never produces a score, a ranking, or a match percentage for a person. Candidates read the same report the employer reads.