Most lists of ChatGPT resume prompts share a flaw that makes them worse than useless: they tell you what to ask for and nothing about what to provide. "Write me a compelling resume bullet for a project manager" is a complete prompt and a guaranteed failure — the model has no idea what you managed, so it produces a confident average of every project manager bullet it has ever seen, complete with metrics that belong to no one. The prompts below are different in one structural way. Each specifies its input requirements, because a prompt without evidence is just a request for fiction.
This is the working companion to our full AI resume workflow; that piece covers the method, this one is the prompt library.
Why "write me a resume" produces garbage: the input principle
A language model predicts likely text. Given rich, specific input about your actual work, the likely text is a sharp version of your facts. Given a job title and vibes, the likely text is the statistical center of everyone's resume — which is precisely the generic tone recruiters have learned to skim past, decorated with plausible invented numbers.
So the rule for every prompt in this library: the model rewrites; you supply the reality. Where a prompt says CONTEXT REQUIRED, skipping that input does not make the prompt fail loudly. It makes it fail quietly, with fluent fabrication. That is worse.
The context block: what to paste before any prompt
Build this once, reuse it all session. Keep it in a notes file.
CONTEXT — use only these facts, never add to them: Target role: [title and seniority you are applying for] Current/recent role: [title, company type, dates, team size] Responsibilities: [plain-language list] Verified numbers: [every metric you can defend, with "approx." where estimated] Tools I actually use: [list] Things I am proud of: [2–3 specific stories, rough language is fine] Rules: if a claim needs a number I have not given, write [METRIC?]. Never invent tools, scope, or results.
The closing rules paragraph is the load-bearing part. It converts the model's gap-filling instinct into visible placeholders you can resolve with real data.
Bullet-writing prompts: five variants
1. The grounded first draft. CONTEXT REQUIRED: context block.
From the context above, write 4 resume bullets for [role], strongest first. Start each with a past-tense action verb, keep each under 25 words, and lead with the outcome where one exists.
2. The compression pass. CONTEXT REQUIRED: your existing long bullet.
Rewrite this bullet in under 20 words without dropping any factual claim: "[paste bullet]". Give 3 versions with different structures.
3. The duty-to-achievement converter. CONTEXT REQUIRED: a duty statement plus one fact about outcome or scale.
This bullet describes a duty: "[Responsible for managing the support inbox]". Using only this additional fact — "[about 200 emails a week, response time cut from 2 days to same-day]" — rewrite it as an achievement bullet.
Before: "Responsible for managing the support inbox." After: "Managed a 200-email-per-week support inbox, cutting typical response time from two days to same-day."
Same facts, different sentence. That conversion is the highest-value trick in bullet writing generally, and the model executes it well when — and only when — you supply the fact.
4. The verb variety pass. CONTEXT REQUIRED: your full bullet list.
Here are all my bullets. Flag repeated opening verbs and repetitive sentence rhythm, and suggest alternatives that do not inflate my scope. Do not change any factual content.
5. The scope-honest reframe. CONTEXT REQUIRED: a team achievement plus your actual contribution.
Our team achieved [result]. My specific part was [your part]. Write a bullet that credits my contribution accurately without claiming the whole result.
Summary and tailoring prompts: matching a real job description
6. The three-sentence summary. CONTEXT REQUIRED: context block plus target posting.
Using the context and the job posting below, write a 3-sentence summary: sentence 1 states role and years, sentence 2 gives my strongest verified proof point, sentence 3 states direction relevant to this posting. No adjectives without evidence. [paste posting]
7. The requirements extractor. CONTEXT REQUIRED: the posting.
List this posting's requirements in two columns: must-haves vs nice-to-haves, using their exact wording. [paste posting]
8. The honest gap map. CONTEXT REQUIRED: context block plus extracted requirements.
Compare my context against these requirements. Three lists: (1) requirements I clearly meet, with my evidence; (2) partial matches, with what is missing; (3) requirements I do not meet. Do not stretch.
9. The vocabulary mirror. CONTEXT REQUIRED: your bullets plus posting.
Where my bullets and this posting describe the same real skill in different words, rewrite my bullet using the posting's term. Flag rather than rewrite anything that would claim a skill I have not demonstrated.
10. The two-posting fork. CONTEXT REQUIRED: your resume plus two postings.
Here are two postings I am targeting. For each, list which of my existing bullets to emphasize, cut, or reorder — without writing new claims.
These four are the AI-assisted core of tailoring a resume in 15 minutes; prompt 8 in particular keeps the process honest by making gaps explicit instead of paper-overable.
Critique prompts: making ChatGPT your toughest reviewer
11. The brutal reviewer.
Review this resume as a skeptical hiring manager for [role] with 200 applications to screen. List every bullet you would skim past and why. Be blunt; do not compliment me.
12. The vagueness detector.
List every phrase in this resume that could appear unchanged on a stranger's resume in the same field. Those are my weakest lines.
13. The interview stress-test.
For each bullet, write the toughest follow-up question an interviewer could ask. I will practice answering; do not answer for me.
14. The six-second scan.
If you could read only the first line of each role, what impression would this resume leave? What is missing from those first lines?
A fuller set of review workflows, including score interpretation, lives in the AI resume review guide.
The hallucination check: catching invented metrics and skills
15. The claim extractor.
List every factual claim in these bullets as a checklist: each number, tool, title, and scope word (led, owned, built) as its own line.
Tick every line against reality. Anything you cannot source gets fixed.
16. The addition detector. Run after any AI rewrite.
Compare my original text and your rewrite. List every fact, number, or implication present in the rewrite that was not in the original.
This is a manual version of what a redline diff does automatically — it surfaces what generation quietly added, which is where most AI resume mistakes enter.
Prompt chains: multi-step workflows for a full revision
17–20. The full-revision chain. Run these in order, in one conversation, with your context block pasted first:
Extract requirements from this posting (prompt 7 format). [paste posting]
Map my context against those requirements honestly (prompt 8 format).
Draft bullets for my most recent role targeting the clear matches, [METRIC?] rule in force.
Now list every claim in your draft (prompt 15 format) so I can verify each one.
Chains beat single mega-prompts because each step's output is small enough to actually check. The failure mode of "rewrite my whole resume for this job" is that no human reviews four hundred words of confident output line by line; the chain forces the review into the workflow. For the cover letter equivalent, the ChatGPT cover letter prompt sequence follows the same architecture.
Frequently asked questions
Which AI model should I use for resume prompts?
Any current major chat model handles these prompts; the differences between models are smaller than the difference between good and bad input. The prompts here are written for ChatGPT but work unchanged in Claude, Gemini, and inside grounded editors like Workplacea.
Why does ChatGPT keep adding metrics I never mentioned?
Because resume-shaped text in its training data is full of metrics, and the model completes patterns. It is not lying, exactly — it is being plausible, which for your purposes is the same problem. The rules paragraph in the context block plus prompt 16 after every rewrite is the containment strategy.
Can I paste the whole job posting into a prompt?
Yes, and you should — extraction and mirroring prompts need the full text. Strip out boilerplate benefits paragraphs if the posting is very long, so requirements dominate the input.
How many rewrites is too many?
When bullets stop getting truer and start just getting different, stop. Two or three passes per bullet is typical: draft, compress, voice. Past that you are polishing sameness into your text — the model regresses everything toward its averages if you let it iterate forever.
Prompts with the guardrails built in
Everything above is workable in a chat window; it just relies on you enforcing the rules every time. Workplacea bakes them in — suggestions arrive as visible redline diffs, and the AI asks for missing numbers instead of inventing them. See what a grounded rewrite looks like in the editor, or start by running your current resume through the free resume checker.