Do CVs Written with ChatGPT Pass ATS? Risks and Best Practices
By Umut Barış Terzioğlu · · Updated
An applicant tracking system does not ask who wrote your CV. Its job is to extract the text into fields such as experience, education and skills, and to let recruiters search them. A CV drafted with ChatGPT goes through the same process as any other. What can hurt it is not the tool but what the text says, and how it ended up in the document.
Where AI drafts go wrong
Left to its defaults, a chatbot produces text that sounds polished and says little. Four problems come up again and again:
- Generic phrases. "Results-driven professional", "proven track record", "dynamic team player". None of these can be checked, so readers skip them, and they rarely match the terms a recruiter searches for.
- Invented details. Ask a model to "make this stronger" and it will often add a percentage, a tool or a team size you never mentioned. Every one of those is a line you may be asked about in an interview.
- Keyword stuffing. Asked to "optimise for ATS", models tend to repeat the job ad's terms in every bullet. A search finds them, but the person who reads the CV next notices the repetition.
- Formatting leftovers. Text copied from a chat can bring markdown asterisks and hash signs, unusual bullet characters, or headings the model made up, such as "Professional Journey".
Before and after
Here are the same notes run through a default prompt and through one with clear rules.
Your notes: "ran the weekly newsletter for 2 years, started testing subject lines, open rates got better"
| Version | Bullet |
|---|---|
| Default prompt | Spearheaded a transformative email strategy, boosting engagement by 45% and driving significant revenue growth |
| Prompt with rules | Ran the weekly customer newsletter for two years and introduced A/B tests for subject lines; open rates rose over the following months |
The first version contains a number and a revenue claim that are not in the notes. The second keeps only what you wrote, and you can explain every word of it. If you know the actual open rate, add it; if you do not, leave the line as it is.
A prompt that keeps the facts yours
Give the model your own notes for one role at a time, together with the job ad, and set the rules up front. For example:
- "Rewrite these notes as three to five CV bullets. Use only facts that appear in my notes."
- "Do not add numbers, tools, employers or results. Where a result is missing, write [number] as a placeholder."
- "Use a term from the job ad only where my notes show I have done that thing."
- "Start each bullet with a verb. No adjectives such as dynamic, passionate or results-driven. Plain text, no markdown."
The XYZ formula, "accomplished X, as measured by Y, by doing Z", is a useful shape to ask for, as long as X, Y and Z all come from you. Then read every bullet and ask one question: could I explain this in an interview? If not, cut it or rewrite it.
You do not need to paste your phone number, home address or email into a chat to get help with bullets. Leave them out.
Getting the text into your CV
Paste the result as plain text ("Keep Text Only" in Word, "Paste without formatting" in many other editors), then rebuild the bullets with your editor's own list tool. Remove stray asterisks and hash signs, and use standard section headings such as "Experience" and "Skills" instead of whatever the model suggested. Export the PDF directly from your word processor and check that you can select the text.
How this site handles AI
On this site, AI explains and suggests; the score is deterministic, and AI never changes it. Every AI output passes a grounding check that looks for numbers, technologies or organisations that are not in your CV. The page on AI-assisted fixes explains the three model options, including one that runs on your own machine.
If you turn on a model, tailor mode rewrites only bullets you already have and leaves placeholders such as [X%] where a measurable result is missing. No skill goes into the CV until you confirm you have it.