How to write a resume with ChatGPT without it making things up

Give ChatGPT your own facts first, tell it to use only those and to mark anything it adds, then check every number, date and title against your records. OpenAI's help center itself says to verify important information.1

Written 11 October 2026. CitedCV. How we write our guides

Check your CV for ATS problems, no account needed

Drop a PDF, DOCX or text file, or paste the text. You get a score, the problems a parser trips on, and your CV as a parser files it.

Runs in your browser. The file and its text never leave this device.

Why ChatGPT adds things to a resume

OpenAI's help center article Does ChatGPT tell the truth? says ChatGPT can produce incorrect or misleading output and can sound confident while it is wrong. OpenAI calls this hallucination, and two of its examples matter on a resume: incorrect dates or facts, and fabricated quotes, studies or references. Its advice is to use ChatGPT as a first draft, not a final source, and to check important information.

OpenAI researchers explain the cause in Why Language Models Hallucinate (Kalai, Nachum, Vempala and Zhang, September 2025): models guess when unsure, because training and evaluation reward a guess over an admission of uncertainty. Their example is close to a CV. Asked for the title of one author's PhD dissertation, ChatGPT, DeepSeek and Llama each answered with confidence, and none gave the right title or year.

What that means for your resume

The paper ties hallucinations to facts that appear rarely in training data, such as a birthday mentioned once. Your job titles, dates and results appear nowhere at all. When your prompt leaves a gap, such as a team size or a percentage, a plausible guess can fill it. The guide to why AI CV builders hallucinate covers the mechanism; this page is the routine that keeps guesses out of your resume.

In the first minute: write a fact sheet

Before you open ChatGPT, open a plain text file and list what is true, one job at a time. This is the material you will tell the model to use, and the record you will check its draft against.

Employer name, and your job title as your contract or payslip writes it.
Start and end month and year.
Three to five things you did there, in plain words.
Any number you can put on them, with where it comes from: a report, a dashboard, a review.
Education, certifications and tools, each with a date.

What to leave out of the chat

This is our advice, not OpenAI's: ChatGPT does not need your street address, date of birth or any ID number to draft your experience section. Add contact details yourself in the final document. OpenAI's data settings are covered further down.

Step by step: from facts to a draft

Paste the fact sheet, then the job posting.
Tell ChatGPT to use only the facts in the sheet and to list anything missing instead of writing it.
Ask it to mark every line it cannot trace to the sheet.
Ask for the resume one section at a time, so each part is short enough to check.
Check every number, date and title against the sheet yourself.
Put the text into a single-column document and paste-test the file.

Example prompt 1: use only my facts

Example prompt, to adapt: 'Below are my fact sheet and a job posting. Write the Experience section of my resume using only facts from the fact sheet. Do not add numbers, tools, team sizes or job titles that are not in it. If the posting asks for something my facts do not cover, list it under Missing instead of writing it.'

Example prompt 2: mark what you added

Example prompt, to adapt: 'Go through your draft line by line. For each line, quote the part of my fact sheet it comes from. If you cannot find one, mark the line ADDED and do not rewrite it.'

Anthropic's documentation for its own model, Claude, recommends the same techniques: tell the model to use only the documents you provide, give it permission to say it does not know, and have it find a supporting quote for each claim and drop the claim when it cannot. The same page says these techniques reduce hallucinations but do not eliminate them, so the marking in prompt 2 is a help, not the check.

Example prompt 3: tailor without inventing

Example prompt, to adapt: 'Compare my fact sheet with this posting. Which of my facts answer which requirements? Do not write new bullet points; just map facts to requirements.'

MIT's career office (CAPD) suggests prompts of this kind for tailoring: asking how to align your resume with the language of the posting, or which parts of the job your resume suggests you would do well in.

Check every number, date and title

MIT CAPD gives the reason plainly: you do not want to be asked in an interview about an experience the AI created and you never had. Do this check yourself, even after prompt 2.

What to check	Against	If it does not match
Every number: percentages, money, team sizes, counts	The source you wrote next to it in the fact sheet	Delete the number and keep the sentence
Job titles	Your contract, payslip or offer letter	Use the title on the document
Start and end dates	Your fact sheet, a payslip or a reference letter	Correct them; do not round to whole years
Tools, certifications, languages	What you could discuss in an interview	Remove them
Verbs such as led, owned, managed	What you actually did	Use the verb that matches your role

Why the model's own list is not enough

Asking ChatGPT to mark its additions catches some of them. It is still the same model checking its own text, and OpenAI's help center notes that the model may express high confidence even in incorrect answers. The record you compare against has to be one you wrote. The comparison of CitedCV and ChatGPT has a longer checklist for stock phrases and overstated verbs.

Paste-test the final layout

ChatGPT gives you text; the file you send is a separate job. Put the text into a single-column document with standard headings, export a PDF or DOCX, and test what a parser gets from it.

Try to highlight a line of the exported PDF. If you cannot select the text, Lever's help page says the document is likely not parseable.
Select all, copy and paste into a plain text editor. Read from the top: name and contact details first, then each job with its employer, title and dates together.
Look for symbols carried over from the chat, such as asterisks or hash signs, and remove them.
Upload the file to the employer's form and read the fields it fills in. Greenhouse's candidate FAQ says you can review and edit parsed information before you submit.
Run the file through the free checker at the top of this page.

What the checker can and cannot tell you

CitedCV's free CV checker runs in your browser, needs no account and calls no language model. It reads a PDF, DOCX or TXT of up to 5 MB and flags what a parser trips on, such as more than one column, text boxes or pages that are pictures, plus missing headings, dates it cannot read and missing contact details. It cannot tell whether a number is true. That part is the check above.

Your data in ChatGPT: what OpenAI's settings say

From OpenAI's help center, read on 5 October 2026:

For services for individuals, such as ChatGPT, OpenAI says it may use your content to train its models.
You can opt out under Settings, Data controls, by turning off Improve the model for everyone. New conversations are then not used for training; past chats stay in your history.
Temporary chats are not used to improve the models and do not appear in your history; they may be kept for up to 30 days for safety purposes.
If you give feedback, such as a thumbs up or down on a reply, the whole conversation may be used for training even after you opt out.
By default, OpenAI does not use content from ChatGPT Business, Enterprise or Edu workspaces to train its models.

MIT CAPD points to the same Improve the model for everyone setting in its advice on using AI for resumes.

Who this approach is not for

You do not have your own records. Without contracts, payslips or dates there is nothing to check the draft against; gather them first.
The employer forbids AI help. Prospects reminds applicants to respect each employer's AI policy, and says failing to disclose AI use where an application asks can be seen as dishonest.
You want every line traced to a source without checking it by hand. That is what a grounded builder is for; see the alternatives below.

Alternatives, named honestly

Plan details come from each vendor's own pages, read on 5 October 2026; check them again before you rely on them.

Option	What it does for a resume	Fits you if
ChatGPT Free	Drafts and rewords from your prompt; OpenAI's pricing page lists a Free plan with limited uploads	You will run the checks on this page
ChatGPT paid plans	The same task with more uploads and messages	You already pay for it for other work
Claude or Gemini	The same kind of draft; Anthropic documents the same prompting to reduce hallucinations in Claude	You prefer them; every check here applies equally
A university career service	Free resume guides from MIT CAPD and Harvard, and advisers if you are a student	You are a student or a recent graduate
CitedCV	Writes only from facts you give it; each line carries the ids of the facts it came from, and lines whose numbers or names cannot be verified against them are dropped	You want each line traced to a source instead of checked by hand

If you want each line traced to a source

CitedCV costs credits: a new account starts with 6, enough for one quick draft of a general CV, and more come in one-time packs from $9 for 30 credits, listed on the pricing page. The comparison of CitedCV and ChatGPT sets the two side by side, and the guide to why AI CV builders hallucinate explains what grounding means. If you already have a ChatGPT draft, you can upload it as a fact source; you confirm each fact it finds.

Questions people ask

Can ChatGPT write my resume?

Yes, it can draft one, but it writes from your prompt plus general knowledge of jobs, so it can add details you never gave it. Give it your facts first and check every number, date and title before you send.

Is ChatGPT free for writing a resume?

Yes, there is a Free plan. OpenAI's pricing page, read on 5 October 2026, lists a Free plan with limited uploads; the paid plans add more uploads and messages.

Does ChatGPT make things up on a resume?

It can. OpenAI's help center lists incorrect dates or facts and fabricated references among the errors ChatGPT can make, and says it may sound confident while wrong. Gaps in your prompt are where it guesses.

What prompt should I use to write a resume in ChatGPT?

One that limits it to your facts. Paste a fact sheet and the posting, tell it to use only the fact sheet, to list anything missing instead of writing it, and to mark every line it cannot trace to your facts.

Is it safe to put my resume into ChatGPT?

It depends on your settings. OpenAI says content from individual accounts may be used for training unless you turn off Improve the model for everyone; temporary chats are not used for training. Leave out ID numbers and your address either way.

Can I use Claude instead of ChatGPT for my resume?

Yes, and the same routine applies. Anthropic's own documentation recommends limiting Claude to the documents you provide and checking each claim, and says these steps reduce hallucinations but do not remove them.

Will a ChatGPT resume pass an ATS?

That depends on the file, not on who wrote the words. ChatGPT gives text; a single-column file with selectable text and standard headings is what parsers read cleanly. Paste-test it or run the free checker on this page.

Should I tell employers I used ChatGPT?

Only if the application asks or the employer has a rule. Prospects advises respecting each employer's AI policy, and says failing to disclose where asked can be seen as dishonest.

How do I check a ChatGPT resume for mistakes?

Compare it line by line with records you wrote yourself. Every number needs a source, every title must match your contract and every date your payslip or fact sheet; delete what you cannot back up.

Sources

OpenAI Help Center: Does ChatGPT tell the truth?help.openai.com/en/articles/8313428-does-chatgpt-tell-the-truth
OpenAI Help Center: Data controls in ChatGPThelp.openai.com/en/articles/7730893-data-controls-faq
OpenAI Help Center: How your data is used to improve model performancehelp.openai.com/en/articles/5722486-how-your-data-is-used-to-improve-model-performance
Kalai, Nachum, Vempala and Zhang: Why language models hallucinate (arXiv 2509.04664, 2025)arxiv.org/abs/2509.04664
Anthropic, Claude documentation: Reduce hallucinationsplatform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations
MIT Career Advising and Professional Development: AI uses for resume writingcapd.mit.edu/resources/ai-uses-for-resume-writing/
Prospects: Using AI in job applicationswww.prospects.ac.uk/careers-advice/getting-a-job/using-generative-ai-in-job-applications/
ChatGPT: Pricingchatgpt.com/pricing
Lever Help Center: Understanding resume parsinghelp.lever.co/s/article/Understanding-Resume-Parsing
Greenhouse Support: MyGreenhouse FAQ for Candidatessupport.greenhouse.io/hc/en-us/articles/43418495049499-MyGreenhouse-FAQ-for-Candidates