Writing a CV with AI: what it can do, and what it must not
A model can rewrite your experience for a specific posting in seconds. It can also make things up in exactly the same seconds, and you are the one who has to defend every line in the room.
- Use AI to reframe what you did, never to add what you did not.
- Grounding is a check against your source text, not a polite instruction in a prompt.
- Specific, measurable lines are both harder to fake and better at passing a screen.
What AI is genuinely good at
Reframing. The work is already in your history; what changes between two applications is which parts lead and which words describe them. A model is fast at finding the sentence you already wrote and saying it in the posting's vocabulary.
It is also good at the boring, valuable parts: turning a wall of duties into achievement bullets, grouping scattered skills, cutting three sentences to one, and writing the whole thing natively in another language instead of translating it.
What it must never do
Invent. A fabricated employer, an inflated title, a date that closes a gap, a metric that sounds good — every one of those is a claim you will be asked about, and some of them are grounds for dismissal after you are hired.
The subtler version is more common: a model rewrites a summary against a posting for a different profession and quietly starts describing the posting instead of you. No invented number, right language, wrong person.
How a tool can actually prevent it
Prompting a model to be honest is not a guarantee, it is a request. What works is checking the output against the source before you ever see it: every figure in the suggestion has to exist in your own text, every skill has to come from your own list, every bullet has to trace back to a bullet you wrote.
That is how Letterin works. A suggestion that fails those checks is regenerated once and then dropped in favour of your original text — and nothing is applied to your CV until you accept it, section by section.
The tells that make writing look machine-made
Recruiters do not run detectors; they notice sameness. Openings like "results-driven professional with a proven track record", adjectives doing the work of evidence, and three bullets that could belong to anyone in your field.
The cure is specificity, which is also what the posting is screening for: what you changed, for whom, and by how much. "Cut page load from 4.1s to 1.3s across the checkout flow" cannot have been written about anyone else.
A workflow that keeps it yours
Set your history down once, honestly, with the numbers you can prove. For each application, let the model propose — the summary, the bullets that matter for this posting, the order of the sections — and read every proposal beside your own words.
Accept what is truer for this job, reject what drifts, and keep the file you would be comfortable reading out loud in the interview. If you would not say a sentence to the person hiring you, it does not belong on the page.
Questions
Can a recruiter tell that AI wrote my CV?
Not reliably, and detectors are not used at the CV stage. What is noticed is generic copy — the same adjectives and the same rhythm as everyone else's. Concrete detail from your own history is what reads as human, because only you have it.
Is using AI for a CV cheating?
No, as long as every claim is true and you can defend it. A CV has always been an edited document; a model is a faster editor. The line is the same as it has always been: the facts are yours.
Does AI improve my ATS score?
It can, by using the posting's own terms for work you really did and by tightening weak phrasing. It cannot fix a broken text layer or a layout the parser reads in the wrong order — those are export problems, not writing problems.
Letterin proposes the rewrite section by section, checks every suggestion against your own text, and changes nothing until you accept it.
Try it free