Can you trust an AI-written resume?

Can you trust an AI-written resume?

Trust isn't a property of the document. It's a property of each claim inside it.

That distinction sounds academic until you try to act on it. Ask "is this resume real?" and you get stuck, because there's no answer that helps you decide anything. Ask "which parts of this can I check?" and the resume becomes useful again, immediately, for the things it was always good for.

Some claims are load-bearing because the world can confirm them. Someone either worked at that company between those dates or they didn't. They either hold that license or they don't. AI writing changes how those facts are phrased. It doesn't change whether they're true.

Everything qualitative is a different story. Skill level, scope of ownership, the size of the thing someone claims to have built, how senior they really were. Those claims were always self-reported and always hard to check. What changed is that they're now expressed fluently and consistently by everyone, which removes the rough signal you used to get from watching someone struggle to describe their own work.

What is an AI-written resume?

An AI-written resume is any application document where a language model shaped the wording, from a light polish through to a fully generated draft.

That range matters more than the label does, because it contains four different things:

1. Light editing. A real resume, tightened. Grammar, phrasing, length.

2. Heavy rewriting. Real experience, restructured and re-worded to sound stronger.

3. Full generation. A resume drafted from a job description and a few prompts, describing real work the person did.

4. Fabrication. Invented experience.

Only the fourth is dishonest, and it existed long before language models did. The first three are people writing well with the tools available to them. Lumping all four together as "AI resumes" is where most of the current confusion comes from, and it's why so much advice on the topic ends up aimed at the wrong problem.

How many candidates use AI on resumes?

Fewer than most hiring teams assume, and the gap between belief and behavior is the interesting part.

TestGorilla's research found that 37% of candidates use AI in their applications. In the same research, 73% of employers reported seeing more AI-generated resumes than before. Those two numbers describe different things. One measures what candidates do. The other measures what hiring teams notice, and noticing is heavily influenced by the resumes that stand out as suspicious.

Volume is climbing at the same time. In March 2026, Robert Half found that 67% of US HR leaders say AI-generated applications are slowing their hiring down.

Put together, that's a real operational problem sitting on top of an inflated perception problem. Both are worth solving. They need different solutions, and treating the second as though it were the first is how teams end up buying detection software.

Our full breakdown of how many job seekers actually use AI has the rest of the numbers.

How do you spot an AI-written resume?

There are reliable tells, and they're worth knowing before we get to why they don't solve your problem.

1. Uniform sentence rhythm. Every bullet is roughly the same length and built the same way.

2. Achievements with no numbers. "Drove significant improvements in team performance" with nothing attached to it.

3. Vocabulary lifted from the job ad. Not similar language, the same language, including unusual phrasings.

4. Polished prose, vague specifics. The writing is clean and the substance stays general no matter how far you read.

5. Tonal flatness across sections. A career summary and a hobbies line written in the identical register.

Here's the catch, and it's the reason this section is short. Every one of those tells you something about the writing. None of them tells you anything about the person. A brilliant engineer who used a tool to fix her phrasing trips all five. A weak candidate who writes beautifully trips none.

Spotting an AI-written resume is a real skill. It's just not the skill you need.

Do AI resume detectors work?

Not well enough to make decisions on, and the deeper problem isn't accuracy.

Start with accuracy anyway, because it matters. Detection tools are unreliable on short, formulaic, heavily formatted text, which describes every resume ever written. Their false positives land hardest on people writing in a second language and on anyone whose prose is simply clean and plain. When researchers at Stanford tested seven widely used detectors on essays by non-native English speakers, the average false-positive rate was 61.3%. More than half of real, human-written work was flagged as AI-generated. Building a filter that disadvantages competent non-native English speakers is a serious thing to do by accident. Now assume the technology gets solved. You have a perfect detector. What does it give you?

It tells you how a document was produced. It tells you nothing about whether the person can do the work. You'd have spent real budget answering a question you never actually had, and built a hiring process that treats writing style as a proxy for ability.

Which, followed back far enough, is the same mistake that got everyone into this.

What can a resume still tell you?

More than the current mood suggests. Resumes still carry every claim that can be checked against something outside the document.

Employment dates. Company names. Job titles held. Formal credentials, licenses, and certifications. Publications, patents, and public filings. Educational qualifications.

AI writing doesn't touch the truth value of any of these. It rewrites the sentence around them. "Senior Engineer at Acme, 2021 to 2024" is either accurate or it isn't, and no amount of polish changes which. These are verifiable claims, and they're still doing useful work: they tell you career shape, industry exposure, progression, and whether someone has the formal qualification a role legally requires.

Read a resume for those and it remains a genuinely useful document. Read it for anything else and you're guessing.

What can't a resume tell you anymore?

The qualitative half. Skill level, ownership, scope, seniority, and impact.

These are unverifiable claims: assertions with nothing outside the document to check them against. "Led the migration" could mean architected and executed it, or attended the standups. "Expert in Python" is a self-assessment with no scale behind it. "Grew revenue 40%" has no denominator and no attribution.

Two things are worth being precise about here.

First, this gap is not new. Unverifiable claims were unverifiable in 2015. Recruiters compensated with pattern recognition, reading hesitation and vagueness and awkward phrasing as weak signals about how well someone really knew their own work.

Second, that compensation is what broke. Fluent, confident, specific-sounding description is now available to everyone, so the weak signals have gone quiet. The gap didn't widen. It became invisible, which is worse, because an obvious gap gets handled and an invisible one gets trusted.

This is also where genuine fabrication hides, and always did. It hides in the unverifiable column, because that's the only column where lying is safe.

Should you reject a candidate for using AI?

No. And a policy of rejecting suspected AI use fails on three counts.

It's unenforceable. You can't reliably detect it, so the policy becomes "reject people whose writing pattern-matches to a tool," which isn't a policy so much as a bias with paperwork. It filters for the wrong thing. You'd be selecting on writing style. Writing style correlates with ability in roughly the way handwriting used to: a little, unreliably, and mostly for jobs where writing is the job.

It falls unevenly. Second-language writers use these tools more, and get flagged more. The people your policy would quietly exclude are disproportionately the ones who'd benefit most from a fair read.

There's also a simpler point. The market spent 20 years telling candidates that keyword-matched, achievement-heavy, ATS-friendly resumes were the price of being considered. Candidates responded rationally. Treating that response as a character failing is both unfair and unhelpful, because it points your attention at the applicant instead of at the measurement problem you actually have.

Fabrication is different, and it's worth naming clearly: inventing experience is dishonest, it's a minority behavior, and it's caught by checking claims rather than by analyzing prose.

How do you check skills before outreach?

This is where most advice on this topic quietly stops being useful to anyone doing candidate sourcing.

The standard recommendations are structured interviews, work samples, reference checks, and asking candidates directly about their AI use. Every one of those requires the person to already be in your funnel. If you sourced someone and haven't made contact, you have none of them. You have a profile, a decision to make, and a finite number of outreach messages.

So work with what exists before contact. The verifiable column above, which you can check without speaking to anyone. Public work, where the profession produces any, though be honest that this covers engineering and design well and covers operations, finance, nursing, and skilled trades barely at all. And prior scored assessment results, where they exist, which is the only pre-contact signal that spans every role and can be compared candidate to candidate.

Worth addressing the obvious objection, because a sophisticated recruiter will raise it. An AI sourcing tool that reads resumes more cleverly is still reading claims that were never checkable. Better comprehension of an unverifiable assertion produces a more confident guess, not a better one. The difference that matters isn't how well the tool reads. It's which class of claim you decide to act on.

That shift is what our argument about why talent sourcing isn't working covers at the process level, and our complete guide to sourcing candidates is where to go for building it into a workflow.

Check what candidates can do, not what their resumes say

TestGorilla's 350+ skills assessments give you scored, comparable results before you make contact. Stop reading resumes and start verifying skills. Try TestGorilla for free.

Frequently asked questions

Can you trust an AI-written resume?

Not as a single document. Trust the claims inside it that can be checked against the world: dates, employers, titles, credentials, licenses. Don't trust the qualitative claims: skill level, ownership, scope, impact. That split held before AI existed. AI made the unreliable half sound as convincing as the reliable half.

How can I tell whether a candidate's resume is genuine or just AI-generated keywords?

You often can't, and it's the wrong question. Look instead at which claims are checkable. Verify the employment history and credentials directly, and treat every skill and achievement claim as unconfirmed until you have evidence from outside the document, such as public work or a scored assessment.

Do AI resume detectors actually work?

Not reliably on text as short and formulaic as a resume, and false positives fall hardest on second-language writers. More importantly, even a perfect detector would only tell you how a document was written. It would tell you nothing about whether the candidate can do the job.

Should you reject a candidate for using AI on their resume?

No. The policy is unenforceable, it selects on writing style rather than ability, and it disadvantages people writing in a second language. Most candidates using AI have real experience and are describing it in the register the market rewarded. Fabricated experience is a separate issue, caught by checking claims.

What parts of a resume are still reliable?

Anything checkable outside the document: employment dates, company names, job titles, formal credentials, licenses, publications, patents, and qualifications. These tell you career shape, industry exposure, and whether someone meets a formal requirement. AI writing changes their phrasing, not their truth.

How do you check a candidate's skills before you contact them?

Use the pre-contact signals: verifiable history, public work where the profession produces it, and prior scored assessment results. Interviews, work samples, and reference calls all require the candidate to be in your funnel already, which makes them useless for deciding who to approach in the first place.

The takeaway

Stop asking whether the resume is real. Start asking which of its claims you can check.

That one substitution turns an unanswerable question into a short list of tasks, and it moves your attention off the applicant and onto the evidence. The recruiters who make the switch will spend their outreach on people whose claims hold up. Everyone else will keep reading beautifully written documents and hoping.