AI & ChatGPT · 2026-08-17 · 14 min read

ChatGPT, Gemini, and Claude prompts to write a CV

Copy-paste prompts to improve a CV, tailor it to a job ad, rewrite achievements, and extract keywords—with rules so the model does not invent facts.

A ChatGPT prompt for a CV is not a magic sentence. It is a brief: who you are, which posting, which rules, and which output format. If you only type “make me a résumé”, the model fills gaps with clichés and often with metrics that never happened. This guide is a pack to copy, paste, and edit.

It works in ChatGPT, Gemini, and Claude. Change little in the prompt text; change the context you paste underneath. Always: inventory or current CV + job ad as text + “do not invent”. The full method (inventory, review, ATS PDF) is in how to make a CV with ChatGPT. Here we focus on the instructions.

Before pasting anything into chatgpt.com, strip national ID, a full street address, and photos. If the goal is a file ready to send, the chat drafts; ClarCV formats and analyses. Prompts do not replace facts. If you mark [MISSING], fill it yourself—do not ask the model to “estimate a realistic percentage”.

Rules that belong in every prompt

• Output language = language of the job ad (do not mix Spanish and English).
• Do not invent metrics, tools, titles, dates, or employers.
• If a fact is missing, mark [MISSING]; do not fill the gap.
• Bullets: verb + what + scope or tool + outcome if you have one.
• Banned: proactive, passionate, results-oriented, excellent communication skills, data-driven as decoration.
• Do not ask for layout, columns, icons, or a “premium visual CV”.

Prompt 1: rewrite the CV you already have

Copy this, then paste your CV as text and, if you can, a typical target job ad. Prompt: You are a résumé editor. I will paste my current CV. Improve clarity and density. Keep dates, employers, and titles. Turn tasks into achievements only when I provided a result or a proxy. Remove clichés. One page if my career is short; otherwise say what you would cut. Structure: Summary (3–4 lines), Experience, Education, Skills, Languages. Return the CV only.

Second message in the same thread: Make a short table: original line → new line → why. If you added any number I did not give, list it separately. That table is your quality control. ChatGPT and Gemini often “improve” by inserting round figures; the table exposes them.

Prompt 2: tailor the CV to a job ad (highest impact)

This prompt moves interviews because it forces alignment. Prompt: I will paste A) my CV B) a job posting. Extract 10 real keywords from the ad (tools, domain, responsibility verbs). Tell me which already appear in my CV and which are missing. Rewrite the summary, the five most relevant bullets, and the skills list using only keywords I can defend. Do not copy requirements I do not meet. Do not invent the ad’s flagship tool if I have not used it. Output: 1) keyword list 2) tailored CV 3) three remaining gaps and how I would address them honestly in interview.

If you apply to two role families (for example operations and customer support), do not reuse the same thread. Start a new chat per family. Mixing ads produces a Frankenstein that convinces no one.

Prompt 3: achievements from weak tasks

When your bullets are “responsible for…” or “collaborated on…”. Prompt: I will paste weak bullets and, separately, raw facts (volume, tools, frequency, team). Rewrite each bullet with an action verb. If a fact is not enough for an outcome, state scope without inventing a KPI. Max 20 words per bullet. Professional tone, no marketing.

Example of raw facts worth pasting: “family shop, summer 2023, till and restocking, about 25 customers per shift, Excel for orders”. The model can return: Updated Excel orders and served about 25 customers per shift in peak season. That is defensible. “Increased sales 40%” is not if you never measured it.

Prompt 4: CV with little experience (projects, internships, volunteering)

Prompt: I am a junior candidate. I will paste studies, projects, internships, and informal work. Do not disguise classwork as paid employment. Order: Summary, Projects or relevant experience, Education, Skills, Languages. One page. Each project: problem, my role, tools, outcome (academic is fine). Flag what you would cut if space is tight.

Then ask: “Propose six technical skills that actually follow from what I gave you, not a trendy list (Python, SQL, Figma) if they are absent from my projects.” That line cuts the padding that sinks junior AI CVs.

Prompt 5: English version

Do not translate word for word. Prompt: Localize this CV into English for the job ad I paste. Keep facts and figures. Use destination-market job titles, not calques. Professional European tone, not Silicon Valley jargon if the ad is local. Mark [CHECK] on any title you adapted so I can validate it.

Then compare dates and employers with the Spanish version. Inconsistencies across languages look like exaggeration in interview. Keep one source of truth (inventory) and generate ES and EN from it.

How to paste the job ad (so the prompt does not hallucinate)

Do not paste a screenshot. Copy the ad text: title, requirements, responsibilities, and stack. Cut the “we are a values-led industry leader” block if it adds no keywords. If the posting sits behind a cookie wall, expand “show more” before copying. An excellent prompt with a half-copied ad produces a half-copied CV.

If the ad wants five years of Salesforce and you used an internal CRM for two, say so in the inventory: “daily internal CRM, not Salesforce”. The model will stop stuffing Salesforce into every bullet. That honesty saves an awkward interview and a possible ATS reject for lying.

Five-minute checklist after the prompt

• Can I explain every figure in an interview without looking at the chat?
• Do dates and employers match LinkedIn?
• Is the language the ad’s language from top to bottom?
• Did I remove proactive, passionate, and “results-oriented”?
• Does the text already live in a selectable PDF, not only in the thread?

Tweaks by tool: ChatGPT, Gemini, Claude

ChatGPT: works well in a long thread if you paste the ad again when the model “forgets” and returns to clichés. Remind it every few messages: “still do not invent numbers”. Skip third-party GPTs that promise an “automatic ATS CV” without you seeing the final file.

Gemini: useful if you attach the PDF; verify extraction (email, dates, columns). If the attachment is incomplete, paste plain text. Same prompt as ChatGPT. Gemini sometimes over-summarises: add “do not shorten bullets that contain metrics; keep the figures I gave”.

Claude: often follows “do not invent” and “mark [MISSING]” more tightly. Use it for the scepticism pass: paste the CV ChatGPT gave you and ask for “a list of claims not supported by the original text”. That audit is gold before you send.

• One chat per ad or role family; do not mix ten postings.
• Paste the ad every time you request a new tailoring.
• Keep the inventory outside the chat (notes or a sheet). The thread is not your master file.
• Never ask it to “invent believable achievements to cover this requirement”.

Prompts you should not use

“Make me the best CV in the world”, “act as a Google recruiter and give me 98/100”, “hide keywords in white at the bottom”, “inflate seniority a bit”. That yields AI-sounding text, sometimes detectable tricks, and a document you cannot defend. Do not ask for a Canva layout inside the chat either: you will get markdown or fragile HTML.

If the model hands you a “perfect” CV in one shot, be suspicious. A good result is a draft with [MISSING], a keyword list, and your decisions. Then move the text into a parseable PDF. On ClarCV you upload the file, see clarity and structure issues, and regenerate with ATS templates. Start free; Pro is €7/month. Spanish and English.