6 Ways AI Ruins a CV — and How to Use It Without Getting Caught
AI rewriting fails in specific, recognisable ways. Here are the six, why each one is dangerous at interview, and how to keep the benefit without the risk.
Using AI on your CV is now normal, and so is being caught doing it badly. The failures are not random — they fall into six recognisable categories, and five of them are only discovered at interview, which is the worst possible moment. This is what goes wrong, why each one is more dangerous than it looks, and what a sensible use of the tools actually looks like.
1. It invents a number, because numbers sound like evidence
Ask a language model to make a bullet point more impressive and it will often reach for a figure. "Improved process efficiency" becomes "improved process efficiency by 30%". Nobody decided to lie. The model was asked to strengthen the sentence, and a percentage is the strongest-sounding thing available.
Why it is dangerous: this is the single most likely thing to be asked about. An interviewer who sees 30% will ask how it was measured, and there is no recovery from not knowing the answer to your own claim.
2. It moves a real figure to the wrong job
Subtler and more common. You genuinely did manage a £1.2m budget — three roles ago. The rewrite tidies your history and the figure surfaces under your current employer.
Every individual fact is true. The document as a whole is now false, and it is false in a way that survives a careless proofread, because you recognise every number on the page as one of yours.
3. It upgrades your job title
"Analyst" becomes "Senior Analyst". "Acting Team Lead" loses the "Acting". "Assistant Manager" becomes "Manager". These read as harmless tidying and are the easiest claims on the entire document to verify — a reference check confirms titles and dates before anything else.
4. It flattens what made you distinctive
Rewriting for polish converges on the same professional register everybody else is converging on, because the models were trained on the same corpus. The specific, slightly awkward detail that made you memorable gets smoothed into a phrase a recruiter has read four hundred times that week.
This one costs you nothing at interview and quite a lot at the shortlist stage, which is why it goes unnoticed.
5. It writes prose you cannot defend
"Spearheaded a cross-functional transformation initiative delivering step-change improvements in operational resilience." What did you actually do? If you cannot answer that in plain words in an interview, the sentence is a liability, however good it looks on the page.
A reliable test: read each bullet aloud and try to say the same thing to a friend in ordinary language. Anything you cannot translate, cut.
6. It quietly drops something that mattered
Rewrites shorten. In shortening, a model discards what it judges least important, and it has no idea that the certification it removed is the mandatory one for the role you are applying for.
This is the failure people never check for, because proofreading naturally looks at what is on the page rather than what has gone from it.
The rule that prevents all six
Let AI change how something is written. Never let it change what is claimed. Every one of the six above is a case of the second thing happening while you thought the first was.
In practice that means working in small steps and comparing. Rewrite one section, put the old and new side by side, and check the facts specifically — employers, titles, dates, qualifications, every figure — rather than reading for whether it sounds better. It always sounds better. That is what it was optimised for.
That applies whatever you use: a ChatGPT CV session, a dedicated AI CV tool or an AI CV checker that suggests edits. The tool changes; the check does not.
What a safer setup looks like
This is the problem our rewriting was built around, so the design is opinionated. There are four explicit modes rather than an on-off switch. The lowest makes no AI call at all — it fixes layout and nothing else, so if you want a clean document with your exact words, that is available and it is free. Tidy corrects spelling, grammar and punctuation while leaving your wording alone. Polish improves phrasing. The highest is a full rewrite.
Underneath every mode there is a check that runs afterwards. Employers, job titles, dates and qualifications are restored word for word from your original. Every figure in the rewrite has to exist in your source document and stay within the same job — which is exactly the failure in point 2. Company and school names that were not in your CV are rejected outright. When a rewrite fails that check, you get your reviewed document back unchanged along with the reason, rather than a polished version containing something invented.
What it does not do — and we would rather say so — is judge whether the rewritten prose still means what you meant. That is points 4 and 5, and no automated check can settle them. So every change is shown as a before-and-after you approve, and you can put it back with one click.
You can try it on your own CV with the free conversion, which includes Tidy and needs no card. Polish sits on the Starter plan at £15 a month.
If you are weighing a paid writer against an AI tool, CV rewrite services versus AI compares the UK costs and the risks of each.
FAQ
Q01Can employers tell if I used AI on my CV?
Detection tools are unreliable and most employers are not running them. What people do notice is the register — generic professional prose that says little — and, at interview, claims you cannot explain. Both are consequences of letting AI decide what is claimed rather than how it is written.
Q02Is it dishonest to use AI on a CV?
Not inherently. Using it to fix grammar or tighten a sentence is no different from asking a friend who writes well. It becomes dishonest at the point the document asserts something you did not do — which, as the six failures above show, can happen without anyone intending it.
Q03What should I check after any AI rewrite?
Four things, in this order: every employer name, every job title, every date, and every number. Then check for what has disappeared — compare the section length before and after and account for anything that has gone. Reading for quality comes last, because quality is the one thing the model definitely improved.
Q04Can ChatGPT write a CV?
It can draft one, and it is good at tightening wording. It cannot know your figures, dates or what you actually did, so it fills gaps with plausible guesses. Use ChatGPT on a CV you have already written, one section at a time, and check every employer, title, date and number it touches.
Q05Which AI is best for a CV?
The one that shows you exactly what it changed. Any capable model can improve wording; the difference that matters on a CV is whether you can see a before-and-after for every edit and whether facts are protected. A general chatbot gives you neither by default, so the checking is on you.
Q06Is lying on a CV illegal?
It can be. In England, Wales and Northern Ireland, dishonestly making a false representation in order to gain something, such as a job or a higher salary, can be fraud by false representation under section 2 of the Fraud Act 2006, which carries a maximum of ten years. Most CV lies end in a withdrawn offer or dismissal rather than a prosecution, but an invented figure that an AI tool added is still your representation once you send it. Scotland has its own fraud law.
ConnectIQ Editorial documents practical recruitment operations and the controls implemented in the ConnectIQ product. This guide is general operational information, not legal advice.
