A CV written by AI, and how to find what it made up
Yes, you can tell when a CV was written by AI, and not because of the style: you can tell because facts you never gave it are sitting in the document anyway. Ask a language model for a CV and it fills the gaps, because filling gaps is the whole of what it does. The style takes ten minutes to fix. What has to be checked is what it invented, and that is done line by line.
What gives away a CV written by AI
The people who spot it are not reading for style, they are reading for evenness. A CV written by a person is lumpy: some lines long, some short, one job in detail and the next in a sentence. One written in a single pass comes out level.
Every line built the same way
Action verb, what you did, result with a percentage. Once is fine. Seven in a row is not a career, it is a template being filled.
Round figures everywhere
Thirty per cent, twenty-five, forty. Real numbers are ugly: seventeen per cent, twelve minutes, four orders. If they all end in a zero, none was measured.
No synonyms
A model repeats its words: «spearheaded», «optimised», «drove». If three jobs of your life use the same verb, you did not write them.
A summary that fits anybody
«Results-driven professional with proven ability to work in a team.» A summary that would serve the next applicant is not describing you.
The long dash
The em dash, which almost nobody types by hand, mid-sentence. The cheapest tell to find, and the cheapest to remove.
A register more polished than yours
Write the CV in a voice you would never use in an email and the interview will sound like someone else. It shows more in the room.
None of them disqualifies on its own, and all can be rewritten away in an afternoon. The real problem is not that it shows. It is what is underneath.
What a generic AI invents, and why it has to
The mechanism matters, because this is not a malfunction, it is normal operation. Ask for «a CV for an office administrator with five years of experience» and what you asked for is a text that looks like a CV, not yours. It has none of your facts, so it fills in the usual ones.
The certificate you never took
The most expensive. You mention spreadsheets and a course appears; you mention marketing and an analytics certification appears. Checkable in two clicks.
The language it assumes
You studied a semester in Germany, so it gives you German B2. Nobody said you speak it, and the first minute of the interview can be in it.
The tool of the sector
An ERP if you are in logistics, a CRM if you are in sales. Those appear in those CVs, so they appear in yours. And the first technical question lands there.
The title one rung up
«Coordinator» becomes «team lead», «assistant» becomes «specialist». It sounds better, and it will not match what your old employer says.
The dates that close the gap
Eight months out of work become a contract running to the following month. That is not decoration, it is a false fact, and it is written down.
One concrete case. What a person wrote about her job in a shop: «I kept the orders spreadsheet and flagged it when something was running out». What the tool handed back: «Managed the inventory control system, reducing stock incidents by 30% through automated weekly reporting».
It is better written, and that is why it is dangerous. That line holds three things that never happened: the spreadsheet was no inventory system, nobody counted that thirty per cent, and nothing was automated. Whoever reads it cannot know. Whoever defends it can.
The demo shows the opposite path: which line came from which fact on the profile.
Open the demoHow to audit your own generated CV, by hand
About fifteen minutes, with the generated document in one window and what you actually wrote in the other, not from memory. Most expensive first: what could cost you the job is in the first two steps.
Mark every checkable fact
Underline what another person could verify: degrees, certificates, languages, dates, employers, titles. If you did not write it, delete it. Do not rewrite it, delete it.
Ask each number where it came from
For every figure, say where it comes from and who measured it. If the answer is «it sounds about right», it goes. A modest number you can explain beats a good one.
Read each line as if they pushed back
Put «tell me more about that» after every sentence. The ones you could talk about stay. The ones that end where the sentence ends are not yours.
Put your own vocabulary back
Swap the words you never use for the ones you do. If the thing is a delivery note where you work, «logistics documentation» loses. You will say this out loud.
Check the boring parts
Email, phone, links and dates. Nobody rereads those, and a model writing your contact line can move a digit. Copy them by hand.
You will finish with a shorter, uglier CV: uneven sentences, lines with no number. That is what a true CV looks like.
Watch a CV get built from nothing but a fictional profile's confirmed facts.
Try the demoWhat to ask a tool before you trust it
Not all of them fill gaps the same way, and you can tell which one you have before sending anything. Five questions, and the answer is in what the tool does.
«What does it do with almost nothing?»
The fastest test: give it a three-line profile. If a full CV comes back, it is filling in. The right output is a short document.
«Where did this sentence come from?»
Every line should trace back to something you wrote. If you cannot walk the path backwards, it is generated text, not your experience.
«Who writes my contact details?»
Email, phone and links are not writing, they are data. If the model types them, they can come out wrong, and that part has to be exact.
«Can I tell it not to use something?»
A job you would rather not show, a skill you do not want to be typecast by. Having the option is one thing; it being honoured, another.
«Does it show me what changed?»
Before and after, line by line. One that only hands you the result makes you audit all of it; one that shows the change does not.
All five answer themselves inside ten minutes of use. And they are the same question: if a tool can write something you never gave it, it will.
Where AI genuinely helps with a CV
All of that argues against having an AI write your career, not against using one. There are three jobs where it earns its keep, and all three share a trait: the material exists and it is only being moved.
Ordering
What this job asks for first belongs at the top. It is a judgement over text that already exists, and a machine that has read the ad makes it better than you at night.
Cutting
One page fills up fast. Deciding what is surplus for this ad is tedious and comes out better with the ad in view. Nothing that survives is new.
Translating into the ad's words
You say «delivery notes» and the ad says «shipping documentation». Same fact, the reader's vocabulary. It only works where it is true: swapping the word with no fact underneath is inventing.
A real example. The line an analyst had been sending to everything: «Built reports and dashboards for client operations teams». Rewritten for a mobility company: «Was the analyst on the same last-mile delivery client for two years: routes, delivery times and cost per stop for their operations team».
There is not one new fact in it. All of it was on his profile and none of it fitted the generic sentence: who the client was, how long it lasted, what he measured. The original did not lie, it hid.
That is how this is built in BeSuix, and the mechanism is worth stating instead of the slogan. The CV generator works only from your confirmed profile, and its instructions treat inventing a certificate you never earned as a critical failure: it can cost you the job for dishonesty.
Part of it is code rather than instructions. Your email, phone and links are injected from your profile after the document is generated, and the skills you flagged as «do not use» are filtered out before the model sees them. The ad is read at temperature zero and cannot add a keyword you did not paste.
The ad's keywords
The same line of your CV
Before
Built reports and dashboards for client operations teams.
For this ad
Was the analyst on the same last-mile delivery client for two years: routes, delivery times and cost per stop for their operations team.
The rule
Not one new fact. It was all on the profile, and none of it fitted in the generic sentence.
A CV written only from what you have confirmed, one job ad at a time. Free during the private beta, invite-only.
Request accessWhat happens if you get caught
Two things get confused here that are not paid at the same price. A dressed-up sentence is paid for in the interview: they ask about that thirty per cent, there is no answer, and the rest is read with suspicion.
A false fact is paid for afterwards. A degree you do not hold, an invented certificate, dates moved to cover a gap: that is not a weak answer, it is false information in the document you were hired on.
- The reference check calls and the dates do not match.
- The certificate carries a number anyone can verify.
- Somebody on the team studied where you say you studied.
- In month three they sit you in front of the tool your CV claimed.
- In many contracts, falsifying information during hiring is grounds for dismissal.
The arithmetic, without the drama: what you gain is one screen; what you risk is the whole job. It is not a good trade.
There is a consolation here. Almost nobody has a CV as thin as they think. What is missing is not facts, it is that the ones you have are told in the terms of that ad. That does not need inventing, it needs finding.
Frequently asked questions
Can you tell if a CV was written with AI?
Often, and in seconds: identical sentence shapes, round percentages, no synonyms and a summary that would fit anyone. But that tell is cosmetic. What sinks an application is the invented fact underneath, and nobody sees that one until they ask about it.
Is it wrong to use ChatGPT to write my CV?
Not in itself. What is wrong is sending what comes out without checking it against your own facts. Use it to reorder, to cut and to use the ad's vocabulary. Never to produce experience: it does not have yours, so it supplies something plausible.
What prompt should I use for a CV in ChatGPT?
No prompt fixes the real problem, which is that the model does not hold your facts. What helps is the raw material first: paste your dates, employers, tools and numbers, then ask it to reorder them for one ad and add nothing you did not paste.
Does an ATS detect that a CV was written by AI?
No. An applicant tracking system is a database with a search box: it indexes and filters, it does not judge authorship. What hurts you there is layout: a CV in two columns or saved as an image gets read badly and never surfaces.
How do I check an AI-written CV without rewriting it all?
Do the two expensive passes only. First, delete every checkable fact you did not write yourself: certificates, languages, dates, titles. Second, make each number say where it came from. Style can stay imperfect; those two have a real cost behind them.
Does the same apply to a cover letter written with AI?
More so, because a letter is mostly claims and few are checkable on the page. The invented part tends to be motivation: reasons for wanting the job that you never gave it. Read it asking whether you would say each sentence out loud.
Keep reading
Why you are not getting interviews, and which part of it you can change
Four things actually rule you out of a job. Everything else only counts against you. How to tell which is which before you send another application.
The ATS resume format, and what happens when a machine reads yours
One column, real text, ordinary headings. What actually happens inside an ATS, why two columns break a resume, and what to check before you send it.