ChatGPT wrote my resume — what should I check before I send it?

Check two things, in this order. First truth — read every line and delete or rewrite any claim you could not defend in an interview, especially numbers you did not supply; then parsing, confirming the file still reads back correctly as text. Both checks are free and take about ten minutes.

Using a model to draft is fine. Sending it unchecked is not.

There is nothing wrong with drafting this way, and the output is usually better-structured prose than most people write cold. The risk is not the tool. It is that the tool was asked to describe work it had no record of.

You gave it an old resume and a job posting. It did not have your projects, your numbers, or what your manager said in a review. Faced with that gap it produces plausible sentences in the shape of achievements — which is exactly what a fabricated metric is.

Check one: is every line true of you?

Go line by line and put each claim into one of three piles. This is tedious and it is the part that matters.

True and yours — keep it, even if the phrasing is not how you would say it.
True but overstated — usually a scope word. "Led" where you contributed, "owned" where you helped. Downgrade it to what happened.
Not something you can defend — delete it. Any number you did not personally supply belongs here by default: if you cannot say where it came from, an interviewer asking "how did you measure that?" is a question with no good answer.

Check two: does the file still parse?

A rewrite often arrives inside a new template, and templates are where parsing breaks. Multi-column layouts, text boxes and tables all look fine on screen and change what a text extractor pulls out.

Our own measurements make the point: the same resume as a two-column PDF extracted to 19 lines instead of 25, with the sidebar interleaved through the work history — and it still scored 98. A clean score is not proof of a clean parse, so read the extracted text, not just the number.

The fix for next time is a record, not a better prompt

People reach for prompt tricks here. The more reliable fix is upstream: if you keep a running record of what you actually did, the model has real material to work from and stops inventing.

CareerDiary is our product and that is the seat it sits in — we are not trying to be a better writer than the model you already have. We are the place your work is recorded, and the check you run on what the model wrote. The checker is free and needs no account.

Frequently asked questions

Can employers tell my resume was written by AI?

AI-detection claims are unreliable and we will not pretend otherwise. The practical risk is not detection, it is the interview: a line you cannot speak to is a problem regardless of who wrote it. Fix the substance and the question becomes moot.

Is it dishonest to use ChatGPT for my resume?

Using it to draft and edit is a writing tool, no different in kind from asking a friend. It becomes dishonest at the point where the document claims experience you do not have — and that line is crossed by content, not by which tool produced the sentence.

What about the numbers it added?

Treat every metric you did not supply as unverified until you can source it. If a real number exists, look it up and use it. If it does not, describe the outcome qualitatively — "cut the manual step out of the weekly close" is credible; an invented percentage is a trap you set for yourself.

Do I need to rewrite it from scratch?

Usually not. Most AI drafts are structurally reasonable and fail on specific claims, so line-by-line editing is both faster and better than starting over. Keep the structure, fix the substance.

How do I check the file itself?

Select all the text in your PDF, copy it, and paste it into a plain text editor — that is close to what gets stored. Our free checker does the same extraction in your browser, shows the recovered text, and requires no upload or account.