# ChatGPT Resume: Make or Improve It Without Losing Keywords > We scored three versions of one resume: a generic ChatGPT rewrite kept 1 of 15 posting keywords. The prompts that keep them, and what to strip before pasting. Canonical: https://hiredfolio.com/blog/chatgpt-resume Published 2026-11-06 · 11 min Keywords: chatgpt resume, how to use chatgpt to write a resume, chatgpt resume prompts, improve resume with chatgpt, ai resume Every guide to writing a resume with ChatGPT says the same two things: give it good prompts, and check the output for made-up numbers. Both are true. Neither tells you what the most common prompt, "make my resume more impactful", does to the words a recruiter searches for. So we measured it. We took one experience block, wrote it three ways, and ran each through the keyword matcher behind HiredFolio's ATS checker. The generic rewrite kept one keyword out of fifteen. This article starts with that result, then gives you the prompts that build or improve a resume while keeping the words, and the one thing to do before you paste anything personal into a chatbot. ## What a generic prompt does to your keywords, measured The job: Account Manager at a B2B SaaS company. The requirements block of the posting reads: > 3+ years as an Account Manager or Customer Success Manager in B2B SaaS. Own a portfolio of 60 mid-market accounts: renewals, upsell and cross-sell. Run onboarding for new customers and drive adoption. Track churn, NRR and other KPIs; weekly reporting in Excel. Daily use of HubSpot (CRM). Strong negotiation and customer service skills. Fluent English; Spanish is a plus. We built the keyword list the way our free [job description keyword extractor](/tools/job-description-keyword-extractor) does, from the terms it recognizes in that text (account manager, KPIs, reporting, negotiation, customer service, English, Spanish, Excel, HubSpot, CRM), and added five requirement words it does not classify: onboarding, renewals, upsell, churn, NRR. Fifteen keywords. Then three versions of the candidate's last job. The first two are not outputs of a specific tool. We wrote them for this article: one as the candidate would type it, one following the pattern a "rewrite this to sound more impactful and professional" prompt reliably produces, one under a strict instruction. What is measured is the text, not the chatbot. **Version 1, raw.** What the candidate had written. > Account Manager, Northwind Software (B2B SaaS), 2022 to 2026 > > - Managed a portfolio of 60 mid-market accounts > - Renewals and upsell > - HubSpot, Excel reporting > - Onboarding new customers **Version 2, inflated.** The generic-prompt pattern, with no extra information. > Account Manager, Northwind Software, 2022 to 2026 > > - Spearheaded strategic relationship management across a diverse client base to drive sustainable revenue growth and long-term loyalty > - Proactively orchestrated retention and expansion initiatives, delivering exceptional value to key stakeholders > - Leveraged cutting-edge sales technology and advanced analytics to optimize performance visibility for leadership > - Championed a seamless customer journey from initial engagement through full product adoption **Version 3, fact-constrained.** Same facts, plus the numbers the candidate was asked to supply, with the instruction "one action verb and one result per bullet, keep these words verbatim, add nothing I did not give you". > Account Manager, Northwind Software (B2B SaaS), 2022 to 2026 > > - Managed a portfolio of 60 mid-market accounts ($2.1M ARR); renewal rate 92% in 2025 > - Drove renewals and upsell: +$310K expansion revenue in 2025 through quarterly business reviews > - Tracked churn and NRR KPIs in HubSpot and built the weekly Excel reporting pack for the sales director > - Ran onboarding for about 40 new customers a year; time to first value cut from 6 to 3 weeks The matcher scores each version against the fifteen keywords. It folds accents and case, accepts plurals and a list of known aliases, and otherwise looks for the words as written. | Version | Keywords found | Rate | What is missing | | ------------------- | -------------: | ---: | ------------------------------------------------------------------------------ | | 1. Raw | 7/15 | 47% | KPIs, negotiation, customer service, English, Spanish, CRM, churn, NRR | | 2. Inflated | 1/15 | 7% | Everything except "Account Manager", which only survived in the job title line | | 3. Fact-constrained | 10/15 | 67% | Negotiation, customer service, English, Spanish, CRM |

1 of 15

7 of 15 before the rewrite

posting keywords left in the experience block after a generic "more impactful" rewrite. The fact-constrained rewrite of the same facts kept 10.

Source: HiredFolio keyword matcher (the keyword component of the ATS checker), run October 2026 on the three texts above

Read the inflated version again and look for HubSpot. It became "cutting-edge sales technology". Excel and the KPIs became "advanced analytics". Onboarding became "a seamless customer journey from initial engagement through full product adoption". Every replacement is longer, sounds more senior, and is a phrase nobody types into a search field. The raw version, dry as it is, beats it on every keyword. Note what the fact-constrained version still misses. Negotiation, customer service, English and CRM are not in it because the candidate never gave those facts. No prompt fixes that. Only the candidate can say "negotiation: 14 renewals closed above list price" or "English, daily, with the US team", and once they do, in the posting's own words, the matcher finds them. ## Before you paste: your resume is personal data A resume is your name, phone number, address, employers and dates. France's data protection authority, the CNIL, published a Q&A on using generative AI systems on 18 July 2024 that puts the rule plainly: users should "ne jamais partager d'informations confidentielles telles que des données personnelles, des données de l'entreprise ou de l'administration", never share confidential information such as personal data, company or administration data. Its guidance for users adds that they "ne devraient soumettre que des informations qu'ils sont autorisés à partager", should only submit information they are allowed to share. Both quotes are from [the CNIL's questions and answers on generative AI](https://www.cnil.fr/fr/les-questions-reponses-de-la-cnil-sur-lutilisation-dun-systeme-dia-generative). That does not mean you cannot use ChatGPT for your resume. It means three habits: 1. **Strip the header.** Delete your name, address, phone and email before pasting. The chatbot does not need them to rewrite a bullet, and your contact block is the part that identifies you. 2. **Turn off training on your chats.** In a note dated 16 October 2025 on [how to object to the reuse of your data for training conversational agents](https://www.cnil.fr/fr/ia-comment-sopposer-la-reutilisation-de-ses-donnees-personnelles-entrainement-agent-conversationnel), the CNIL documents the path in ChatGPT: account settings, then the data management tab (« Gestion des données » in the French interface), then the option to improve the model for everyone (« Améliorer le modèle pour tous »), which you uncheck. Look for the equivalent in your language's settings. 3. **Think of your current employer.** Client names, revenue figures and internal project names are company data under the same rule. Use "a 60-account mid-market portfolio" rather than the client list. None of the ranking guides on this topic mentions it. It takes thirty seconds. ## Making a resume from scratch with ChatGPT If you have no resume yet, the order matters more than the prompts. Facts first, posting second, writing third, layout last. ### 1. Get the facts out of your head first ChatGPT cannot know how many accounts you managed or what your budget was. Make it ask. This prompt turns it into an interviewer: > "I am going to write my resume for a [job title] role. Ask me about my last three jobs, one question at a time: team size, budget or volume, tools I used daily, one result per job with a number. Do not write anything until I have answered all of them." Answer in fragments. "60 accounts, HubSpot, renewal rate 92%" is enough. What you are building is the fact sheet every later prompt will draw from, and the chatbot's questions are a good way to remember the numbers you never wrote down. ### 2. Pull the keywords out of the posting Paste the requirements block of the job you want (not the company blurb) into the [keyword extractor](/tools/job-description-keyword-extractor), or ask ChatGPT: "List the tools, methods, qualifications and skills this posting asks for, one per line, in the posting's exact words." Keep that list open. Our count of what [real postings ask for](/blog/resume-keywords) shows that the terms that matter are the ones the ad repeats, and they are usually in that block. ### 3. Write the experience section, one job at a time Now the writing prompt, with the two constraints that produced version 3 above: > "Here are the facts for one job: [paste from your fact sheet]. Write 3 to 5 resume bullets. One action verb and one measurable result per bullet. Keep these words exactly as written: [paste the keywords that are true for this job]. Do not add any skill, tool or number I did not give you. No adjectives." Read every bullet against the fact sheet. If a number appears that you did not supply, delete it; the model fills gaps with plausible figures when the facts run out, which is the hallucination every guide warns about and the reason the constraint "do not add" goes in the prompt rather than the clean-up. ### 4. Write the summary last The summary is three lines: your title and years, one proof with a number, what you are looking for. Ask for it only once the bullets exist, from the bullets: "Write a 3-line summary for a [target title] from these bullets, no adjectives, one number." Our [resume summary examples](/blog/resume-summary) show the shape for twenty situations. ### 5. Move the text into a one-column document ChatGPT gives you text, not a file, and that is the right division of labour. Paste it into a one-column template in Word or Google Docs, standard headings, no text boxes. The [formatting rules that survive extraction](/blog/resume-formatting) are short, and the two that matter most are one column and no header or footer for your contact details. ## Improving an existing resume: prompts that hold up Most people arrive with a resume already written. These are the prompts that improve it without the version-2 effect. ### Rewrite one bullet from a fact Never "improve my resume". One bullet, with the facts behind it: > "Here is a bullet from my resume: 'Onboarding new customers'. The facts: about 40 new customers a year, 2023 to 2025, I ran the kickoff and the first 30 days, and time to first value went from 6 weeks to 3. Rewrite it as one line, action verb first, keep 'onboarding' and the numbers. Add nothing."

Example one bullet, two prompts

"Make it more impactful"

Championed a seamless customer journey from initial engagement through full product adoption, driving satisfaction and long-term retention.

From the facts, words kept

Ran onboarding for about 40 new customers a year; time to first value cut from 6 to 3 weeks.

### Find the number you forgot When a bullet has no figure, do not ask for one. Ask for the questions: "Here is a bullet with no numbers. Ask me five questions whose answers would give it a measurable result." Then answer them. The number is yours; the model just reminded you it existed. ### Tailor to a posting without lying Paste the posting and one bullet: "Which terms from this posting describe what this bullet says I did? List them, and tell me which ones the bullet does not support." You then decide which are true. "Negotiation" is true if you negotiated the renewals; it is not true because the posting asks for it. The method for [matching a resume to a job](/blog/resume-keywords) takes twenty minutes per application and does not change the facts. ### Give the skills section one proof each A bare list of skills is the easiest section to inflate and the one recruiters skip. Ask: "For each skill in this list, write the one bullet from my experience that proves it. If none does, mark it." Delete the marked ones. The [skills section guide](/blog/skills-for-resume) covers which to keep. ### Proofread as a list, not a rewrite "List every spelling, grammar or consistency error in this text, with the line and the fix. Do not rewrite the text." A full rewrite silently swaps words; a list lets you accept each change on its own. The same applies to cutting: "Shorten this to fit one page. Keep every tool name, every number and every date." ## Which synonyms a keyword matcher forgives, and which it never does Part of why version 2 scored 7% is a question nobody in the ranking guides answers: how literal is the match? Here is how ours works, since it is the one we can read. The matcher ignores accents and capitalization, and reads plurals and a set of noun families as one word, so "KPI" and "KPIs" match, and "developer" meets "development". Verbs are not in that family: a posting that asks for "negotiation" is not satisfied by a bullet that says "negotiated", so the noun has to appear somewhere, in the skills line if nowhere else. It also carries a list of aliases for things a posting and a resume routinely name differently: "KPI" and "key performance indicators", "project management" and "project manager", "social media" and "community management", "GDPR" and its French name "RGPD", "account manager" and "key account manager". Those forgive a real difference in vocabulary. What it never does is read a periphrase. "Cutting-edge sales technology" is not HubSpot, "advanced analytics" is not Excel, "international stakeholders" is not English. No alias list can contain every way to avoid naming a tool, and it should not try: a recruiter searching their system types the tool's name.

Words that survive a match and the ones a rewrite replaces them with

Not found

"Cutting-edge sales technology"

"Advanced analytics for leadership"

"International stakeholders"

"Retention and expansion initiatives"

Found

"HubSpot"

"Churn and NRR KPIs, weekly Excel reporting"

"English, daily, with the US team"

"Renewals and upsell"

One rule under all of them: a tool, a metric or a language is named, never described. The matcher forgives plurals and known synonyms; it does not forgive elegance.

## Two questions people ask **Can employers tell I used ChatGPT?** What a recruiter sees is the pattern: bullets of identical length, adjectives instead of numbers, a summary that could open anyone's resume. A resume written from your facts with the words kept does not have the pattern. Our guide to [how ATS filters actually work](/blog/ats-resume-filters) covers what the software does read. **Should I upload my PDF or paste the text?** Paste the text. Open your PDF, select all, copy, and paste that. It is the text an extraction step gets from your file too, so if a column comes out interleaved or a heading is missing, you have just found a formatting problem before any software did. A resume saved as an image (a scan, or a design tool export) yields no text at all, and the scanner below refuses it rather than scoring it. ## Check what the rewrite kept with HiredFolio The experiment above is one you can run on your own resume. HiredFolio's free [ATS checker](/ats-checker) takes your resume as a PDF (8 MB max) and the job posting pasted as text (at least 50 characters), extracts the text from the file, and scores it out of 100. Keyword match is 40% of that score; the rest comes from required-skills coverage, formatting checks (multi-column layout, missing summary, experience without bullets) and how well the experience aligns with the role. The report lists the posting's keywords it found and the ones it did not. Without an account you get five scans a day and the first five matched and three missing keywords; signing in shows the whole list. Run it twice: once on the resume you have, once after the rewrite. If the second score is lower, you have just caught your own version 2.

Key takeaways before you paste your resume into a chatbot

01

Delete your contact block and turn off training on your chats before you paste anything; a resume is personal data.

02

Prompt one bullet at a time with the facts, the numbers and the words to keep verbatim, and the instruction to add nothing.

03

Check that every tool, metric and language from the posting is still written by name in the final text, not described.

The check takes two minutes: score your resume against the posting and see the keywords it lost.

## Read next - [How to write a CV: the UK, Irish and academic version, and what the law keeps off it](https://hiredfolio.com/blog/how-to-write-a-cv) - [How to list work experience on your resume (with examples)](https://hiredfolio.com/blog/work-experience-resume) - [How to list education on your resume: 10 examples, and what a keyword matcher reads in each line](https://hiredfolio.com/blog/resume-education-section)