Can You Trust an AI-Written CV? The Hidden Risk of AI Hallucinations

Ask most job seekers why they turned to ChatGPT to help with their CV and the answer is usually the same: it's fast, it's free and it sounds impressive. What most people don't ask is whether any of it is actually true.

That's the problem with AI hallucinations. They rarely show up as obvious nonsense. Instead, they slip in as small, plausible-sounding details - a tool you never used, a budget you never managed, a percentage you never calculated. On a CV, where every line is meant to be a verifiable fact about your working life, that's not a minor glitch. It's a landmine.

What Is an AI Hallucination, Exactly?

In simple terms, a hallucination is when an AI tool generates content that sounds confident and correct but isn't grounded in anything you actually told it. It's not lying in the human sense, the model has no intent to deceive you. The problem is that it's built to keep the sentence flowing, and a gap in your input is just another gap to smooth over. Convincingly.

This happens across all kinds of AI-generated content, but CV writing is one of the riskiest places for it to happen, because a CV isn't a piece of creative writing. It's a document that stands behind background / reference checks and interview questions. Get caught with a hallucinated line on it and the consequences are a lot more serious than a typo.

The Four Ways Hallucinations Actually Show Up

I've reviewed a lot of CVs over the years, AI-assisted and otherwise, and the fabrications rarely look like fabrications. Here's what they actually look like.

Why Job Descriptions Drive Hallucinations

If you want to know when hallucinations are most likely to happen, look at the prompt, not the model. The single biggest trigger is pasting a detailed job description straight into the AI alongside your CV and asking it to "tailor this."

A job description listing fifteen or twenty specific tools and methodologies puts the model in an impossible position: match the job description closely or stay faithful to what you actually gave it. Cheaper or smaller models tend to default straight to keyword-matching and fabrication. The more advanced frontier models are better at avoiding outright invention, but they still collapse into conflation surprisingly often when they're straining to hit every keyword in the ad.

The Conflated Metric Trap

Here, I'll walk you through how this actually plays out, because it's easier to spot in theory than in your own CV. Meet Hayley, a mid-level data analyst going for a Senior Analytics Lead role.

Hayley pasted her raw bullet points next to a job description asking for "6+ years' experience in Python, SQL and managing enterprise analytics budgets over £1M," and told the AI to tailor her CV to match. Here's what came back:

Every part of that sentence is wrong, and every part of it looks completely reasonable on paper. What actually happened: the AI pulled her Python skills from Company B and attached them to Company A. It multiplied her real quarterly budget several times over to hit "enterprise scale." And it invented an 18% reduction in reporting variance out of thin air, because that's the kind of thing a punchy bullet point says.

The Four-Step Audit Every AI-Assisted CV Needs

Before you submit anything an AI has touched, run it through this quick checklist:

Step 1: The Metric Traceability Test

• Circle every number. Every percentage, dollar or pound figure in the draft.
• Trace the source. Did you explicitly give the AI this exact figure? If not, delete it or work out the true number yourself.

Step 2: Tool-to-Employer Alignment

• Map technical skills to roles. Cross-reference every piece of software, language or platform against the specific job it's listed under.
• Spot migration errors. A skill you picked up in 2024 has no business appearing under a role from 2021.

Step 3: Background Check Verification

• Match official records. Job titles and dates should align exactly with your contract.
• Strip out "title creep". If the AI upgraded "Marketing Assistant" to "Growth Marketing Lead," put it back, or use something transparent like "Marketing Assistant / Growth Focus" if that's genuinely the work you did.

Step 4: The Interrogative Pressure Test

• Read every bullet out loud. Imagine an interviewer asking, point blank: "how exactly did you do this?"
• The 30-second rule. If you can't explain the how, why and tools involved within 30 seconds without checking your notes, rewrite it or cut it.

How to Use AI Without Setting a Trap for Yourself

The safest approach I've seen, and the one I'd recommend to anyone using AI to help with a CV, is "draft first, audit second."

• 1
  Write the facts yourself. Your work history, dates, employers and the specific tools you used should come from you, not the model. Don't hand AI a blank page and hope it knows the truth.
• 2
  Use AI for style, not substance. Ask it to sharpen sentence structure and tighten your language, not to decide what you achieved.
• 3
  Trace every bullet back to its source. If you can't point to where a claim came from, it doesn't go on the CV.
• 4
  Delete anything you can't defend. If the AI slipped in a tool you'd be uncomfortable discussing in a technical interview, remove it. No exceptions.

Recruiters Are Fighting Back With AI of Their Own

Here's what a lot of job seekers don't realise: this has turned into AI checking AI, and the recruiting side is winning more often than you'd think.

The Applicant Risk Matrix

Not every hallucination carries the same weight. Here's roughly how they stack up.

Trust Is the Real Casualty

Beyond any single rejected application, there's a bigger cost building up. As AI-written CVs become more common, some employers are already treating suspiciously "perfect" CVs with more scepticism, not less. The polish that AI provides so easily is starting to work against genuine candidates too, simply because recruiters can no longer assume a well-written CV reflects a well-qualified one.