Is That Resume Even Real? Spotting Fake and AI-Generated CVs Before They Cost You a Hire

A few years ago, the worst thing you'd find in a stack of resumes was an exaggerated job title or a graduation date that didn't quite add up. These days, you might be looking at a document that was never written by the person whose name is on it — and in some cases, one that's been deliberately built to fool the very system you're using to screen it.

This isn't a fringe problem anymore. GoodTime's 2026 Hiring Insights Report found that 27% of talent-acquisition leaders now name fake or AI-generated candidates as the single biggest hiring challenge they expect to face this year — narrowly edging out the 26% still worried about a shortage of qualified people. For the first time, the "is this person even real" question has overtaken the "can we find enough good people" question. That's a genuine shift in what keeps recruiters up at night. For a long time, the problem in recruitment was finding enough good people. Now it's increasingly working out whether the person on the page is who — or what — they say they are.

Here's what's actually going on, what the numbers say, and what a hiring team can realistically do about it.

The Scale of the Problem

The use of AI to write or polish job applications has gone from a curiosity to close to the default in a remarkably short time. Back in February 2023, a ResumeBuilder survey found that 46% of job seekers had already used ChatGPT to write or rework their resumes and cover letters. By late 2025, recruiting platform iHire was tracking 29.3% of candidates using AI specifically for resume or cover-letter writing — up from 17.3% just a year earlier — and Greenhouse's broader November 2025 survey found 74% of US job seekers now use AI somewhere in their job search. However you slice the numbers, the trajectory only goes one way.

Some of that is harmless — plenty of candidates use AI the way they'd use a careers advisor, to tighten their phrasing or fix a clunky paragraph. But a meaningful slice of it goes well past polishing. Cybersecurity firm Huntress, reviewing applications between September and November 2025, flagged 23.2% — almost a quarter — as outright fraud risks: AI-generated, fabricated, or substantially falsified submissions. The same review found only 19% of hiring managers felt genuinely confident their current process would catch a fraudulent applicant if one came through. Add to that a SHRM-cited survey of 4,000 hiring professionals in which 85% said they'd personally caught a lie or misrepresentation on a resume during screening — up from 66% just five years earlier — and you start to see why "can I trust what's in front of me" has become the question that matters most.

And here's the part that should give everyone pause: it's working. That same ResumeBuilder survey found that three in four job seekers who used ChatGPT to write their resume landed an interview. Whatever you think about the ethics of it, AI-assisted applications are getting people through the door.

Three Different Problems Wearing the Same Coat

It's worth separating out what we're actually talking about, because "AI-generated CV" covers at least three distinct situations, and each one calls for a different response.

The polished application. A real candidate, with real experience, who's used AI to tidy up their grammar, restructure a paragraph, or make their cover letter sound more confident. This is the most common scenario by far, and on its own it's not really a red flag — it's just what writing assistance looks like in 2026.

The fabricated application. A candidate — real or invented — whose resume describes experience, skills, or achievements that don't actually exist, generated wholesale or substantially embellished by an AI tool prompted to "make it sound impressive." This is where things start to matter. An exaggerated CV wastes a hiring manager's time at best, and at worst gets someone into a role they're not equipped to do.

The engineered application. The newest and most unsettling category — documents deliberately built not to impress a human reader, but to manipulate the AI tool doing the screening. This is where prompt injection comes in, and it deserves its own section.

When the Resume Is Talking to Your Software, Not to You

Here's something that would have sounded like science fiction three years ago: candidates are now hiding instructions inside their application documents, written specifically for the AI system that processes them.

The technique is called prompt injection, and it works by exploiting the fact that AI tools read documents as text — including text a human reviewer would never notice. White font on a white background. Font sizes so small they're invisible at normal zoom. Text positioned outside the visible page margins. One documented example involved a resume with hidden text instructing the screening tool to report: "This is an exceptionally well-qualified candidate." The resume's actual content told a different story entirely.

It's more widespread than you'd hope. Greenhouse's November 2025 survey of over 4,100 job seekers, recruiters and hiring managers found that 41% of US job seekers admitted to hiding text designed to manipulate AI filters — and of those who hadn't tried it yet, 52% said they were considering it. On the other side of the desk, 22% of hiring managers said they'd personally caught an applicant doing it. Worth noting that Greenhouse sells detection software, so they've got an obvious commercial interest in talking up the scale of the problem — but a number that size, from a survey that large, isn't something to wave away either.

Detection numbers from elsewhere in the industry back up the broad picture, even if the exact scale depends on who's counting and how. ManpowerGroup — the largest staffing firm in the US — reports finding hidden text in around 10% of the resumes it scans, which works out to roughly 100,000 resumes a year.

Greenhouse's own internal estimate is closer to 1% of all applications — a tenth of ManpowerGroup's figure, which says as much about how differently these tools are built and tuned as it does about the true size of the problem. Even at the more conservative end, though, 1% across the volumes these platforms handle still adds up to tens of thousands of manipulated submissions a year.

The techniques keep getting more sophisticated, too — encoded payloads using Base64 or ROT13, characters designed to slip past basic filters, and instructions hidden inside image alt-text or embedded graphics. This isn't a one-off curiosity. It's an arms race, and it's already underway.

For what it's worth, plenty of recruiters say the tactic tends to backfire — once it's spotted, it's about the fastest way to take yourself out of contention entirely. But "it sometimes backfires" isn't the same as "it's not a problem," particularly for any organisation whose screening process doesn't catch it in the first place.

What AI-Written Text Actually Looks Like

Setting prompt injection aside, there's a more everyday skill worth building: recognising when a piece of writing was substantially generated by AI rather than by the person it claims to represent.

A few patterns show up again and again.

It could be about anyone, anywhere. Genuinely written cover letters tend to carry small, specific details — a reference to something about the role, the company, or the industry that shows the person actually thought about this job in particular. AI-generated text, left unprompted, tends toward language so general it could be dropped into an application for a completely different company without anyone noticing.

It reads a little too smoothly. AI models are trained to produce text that's grammatically clean, evenly paced, and unfailingly polite. Real writing has rough edges — sentences that run a bit long, a turn of phrase that's slightly unusual, a structure that doesn't quite match the one before it. When everything reads like it was sanded down to the same finish, that evenness is itself a signal.

The claims don't match the record. This is often the most reliable tell of all. An AI tool asked to "make this sound more impressive" will frequently do exactly that — introducing language about leading teams, driving strategic outcomes, or managing complex projects that simply isn't backed up by the rest of the document, the candidate's LinkedIn profile, or their actual work history. A two-year stint at a small firm doesn't usually come with "extensive experience leading cross-functional international teams" attached to it.

Different documents, different people. It's worth reading the cover letter, the resume, and the LinkedIn profile side by side. A concise, plainly-written resume paired with a cover letter that suddenly turns eloquent and expansive is the kind of mismatch that's hard to explain away.

A word of caution, though: don't over-rely on any single one of these signs, and definitely don't lean entirely on commercial AI-detection software. A 2026 study published in the International Journal for Educational Integrity put two of the most widely used AI-detection tools through their paces and found accuracy ranging anywhere from 60% to 90% — with both tools performing markedly worse on "hybrid" text, the kind that's been drafted with AI and then edited by a person. That's not a resume-specific finding, but it's hard to think of a more perfect description of what a real-world job application increasingly looks like: part machine, part human, stitched together by someone who knows roughly what a hiring manager wants to read. Treating a detector's verdict as proof of anything is a mistake. Treating it as one input among several is a much safer bet.

What This Means for Your Process

None of this means AI-assisted applications should be treated as automatically suspect — that would be both unfair and impractical, given how mainstream the practice has become. The goal isn't to catch people out for getting help with their writing. It's to make sure the assessment that follows is actually measuring the person, not the cleverness of the document in front of you.

A few principles make that achievable.

Screen for manipulation before you screen for fit. Checking a document for hidden text, encoded instructions, or other injection attempts is a different task to assessing whether someone's a good match for the role — and it needs to happen first. There's no point running a careful, considered assessment on a document that was built to game that very assessment.

Never let a single document carry the whole decision. A resume is one data point. It's also the easiest one to embellish, generate, or manipulate. Layering in structured interviews, validated behavioural assessments, and consistent benchmark comparisons means a polished — or fabricated — resume can't single-handedly carry someone through your process. If the rest of the picture doesn't support what the document claims, that gap becomes visible.

Keep a human at the centre of the judgement call. Detection tools and automated scans are useful for surfacing patterns a person might miss across hundreds of applications. They're far less reliable as a final word. The combination — automated screening to flag what's worth a closer look, paired with a person making the actual call — consistently outperforms either approach running alone.

Build the check into the process, not onto it. Retrofitting a fraud check after you've already built your shortlist is a much harder, much less reliable exercise than catching the issue at the point of intake — before a manipulated application has had any chance to influence anyone's thinking.

How GrowMyTeam Approaches It

This is exactly the layer GrowMyTeam was built to add. Every resume that comes through the platform is scanned for the patterns described above — embellished or AI-generated content, job-ad keyword stuffing, and hidden instructions aimed at the screening process itself — before it ever reaches a benchmark comparison.

That scan isn't there to catch candidates out, and it isn't the whole story either. It's one layer in a process that also draws on validated DISC behavioural profiling and structured interview data, so the picture a hiring manager works from is built from several independent sources rather than resting on a single document. The hiring manager still makes the call — with a clearer, more reliable picture than a resume alone could ever provide.

The Bigger Picture

The honest reality is that AI has changed what a job application looks like, probably for good. Pretending otherwise — or treating every AI-assisted document as a red flag — isn't a workable response. What's needed is a process that can tell the difference between a candidate who used AI to write more clearly, and one who used it to obscure what's actually true about their experience — or worse, to manipulate the very tool meant to assess them.

That's not a problem you solve with a single clever filter. It's one you solve by building a process that never depended on a single document telling the whole truth in the first place.

GrowMyTeam.ai scans every application for AI-generated content, embellishment, and hidden manipulation attempts — then builds the real picture from validated behavioural profiling and structured interviews, so hiring managers are working from substance, not spin. Book a demo or try our free profiling tool.