AI & ChatGPT · 2026-08-17 · 15 min read

How to make a CV with ChatGPT (step by step)

A practical method to draft a résumé with ChatGPT or Gemini: inventory, prompts, catching invented metrics, and why chat output is not an ATS PDF.

If you are searching how to make a CV with ChatGPT, the expensive mistake is asking it to “write an amazing résumé” and pasting the result into a document. The model does not know your career, cannot see the vacancy unless you paste it, and does not produce a file that ATS parsers handle well. It can help you write. It cannot replace facts, layout, or a human final pass.

This guide is a method that works: first an evidence inventory, then a prompt that includes the job ad and your current CV, then a pass to catch invented metrics and generic phrasing. The same flow works in Gemini and Claude. Finally the text has to live in a selectable-text PDF with standard headings and truthful keywords. That is where a chat falls short and a tool like ClarCV (clarcv.com) finishes the job.

You will not “hack” ChatGPT. You will use it as a fast writer you must supervise. If you have no CV yet, the inventory is your raw material. If you already have a weak PDF, do not throw it away—improve it. Starting from a blank chat often yields something that looks professional and collapses in the interview.

What ChatGPT does well (and what it does not)

ChatGPT is useful for turning tasks into achievements, shortening a summary, suggesting action verbs, and aligning vocabulary with a posting. It can also produce a more natural English draft than word-for-word translation. Gemini and Claude perform similarly if you give the same context: CV + ad + clear rules.

• Good: rewriting bullets with verb, scope, and outcome from facts you provide.
• Good: extracting 8–15 keywords from an ad and marking which ones already appear in your CV.
• Good: cutting a summary to 3–4 lines aimed at the target role.
• Bad: inventing percentages, team sizes, or tools you never used.
• Bad: exporting an ATS PDF, choosing a parseable template, or checking how the file parses.
• Bad: knowing your market better than you do if you do not paste the real posting.

The model completes patterns. With thin context it fills in clichés: proactive, results-oriented, excellent communication skills. That does not pass filters or persuade a recruiter. The quality of a ChatGPT CV is the quality of what you paste and what you delete afterwards.

Step 1: inventory (15 minutes, no chat)

Open a blank document. List roles (title, employer, month/year), 3–5 facts per role (even if poorly written), real tools, education, languages, and links. Include projects, internships, and volunteering if you are junior. Do not design. Do not ask the model to invent impact yet.

Add honest numbers: book of business, frequency, team size, ticket volume, approximate budget, deadlines. If you lack KPIs, use proxies: “about 25 calls a day”, “inventory of 300 SKUs”. You will have to defend every figure in interview. ChatGPT will not be in the room.

• One target role family (for example B2B Customer Success, not “whatever comes up”).
• One or two real job ads pasted as text, not a vague memory of the posting.
• What you will cut: irrelevant hobbies, dead courses, soft skills with no proof.
• What not to paste into chatgpt.com: national ID, full street address, photo, passwords.

Step 2: a ChatGPT prompt that actually works

Do not start from an empty chat. Paste inventory or current CV, the ad, and rules. A useful prompt: Act as a recruiter who also thinks about ATS filters. I will paste 1) my CV or inventory 2) a job ad. Rewrite the CV in the language of the ad, one page if junior or mid-level, sections Contact, Summary, Experience, Education, Skills, Languages. Start each bullet with a verb. Do not invent metrics, tools, or job titles. If a fact is missing, write [MISSING] instead of filling the gap. Avoid clichés (proactive, passionate, results-driven). Return CV text only, no layout.

If you already have a PDF, paste the text (copy from the file; if you cannot select text, neither can an ATS). Then: “Do not change dates, employers, or titles unless I ask. Improve clarity and align truthful keywords from the ad.” That constraint reduces hallucinations.

Second pass in the same thread: “List which sentences sound generic, which ad keywords are missing, and which claims I never gave you data for.” Use that list to edit; do not blindly accept the second draft. A ready-made pack for ChatGPT, Gemini, and Claude is in the AI prompts for CVs guide on this blog.

The same method in Gemini (and Claude)

Making a résumé with Gemini is not a different system: same inventory, same prompt, same review. Gemini may let you attach the PDF if extraction works; check that it does not drop columns or the header email. Claude often follows “do not invent” more tightly. None of the three produces an ATS layout on its own.

If you switch tools mid-way, paste the ad and inventory again. Chats do not share memory across products. Do not assume a “smarter” model detects lies: a stronger model invents more fluently if you leave gaps.

Step 3: catch inventions and template language

Read every bullet out loud. Ask: can I tell this story in 60 seconds? If ChatGPT wrote “reduced churn 23%” and you never measured churn, delete it. Replace with a real proxy or the process you actually ran. Same for Salesforce, SAP, or Python if you only saw them in a two-hour course.

Before (typical chat): Responsible for improving customer experience and optimizing processes with a data-driven mindset. After: Managed 40 accounts in HubSpot, cut first-response time from 12 hours to 5, and documented onboarding in Notion for the team. Same person, usable signal. The ATS finds HubSpot and Notion; the human understands the work.

Watch inflated titles. “Lead” and “Head of” look fine in chat and bad against a LinkedIn check. Dates, employers, and titles must match your public profile. A ChatGPT CV that contradicts LinkedIn creates instant distrust.

Step 4: the chat is not the file you send

Pasting the text into Canva or a two-column template and exporting PDF is the classic failure. Many of those files are images or fragile tables. An ATS needs selectable text, standard headings (Experience, Education, Skills), and dates next to each role. Test: open the PDF, select three paragraphs, paste into a plain-text editor. If nothing comes out, do not send that file.

ChatGPT can return markdown. That is not a résumé. You need an A4 document, a professional filename (Ana_Lopez_CSM.pdf), and a reasonable file size. For most European job portals, a clean single-column PDF is safer than a “premium” layout that breaks when parsed.

• One main column; avoid sidebars where skills exist only as icons.
• Contact as real text at the top (email, phone, city), not only in a graphic header.
• Ad keywords inside achievements, not repeated in white text at the bottom.
• Match the ad language: do not mix Spanish and English in one PDF unless asked.

Privacy: consumer ChatGPT is not a CV product

Pasting a full CV into chatgpt.com is convenient and risky. A résumé carries name, phone, email, sometimes a photo, and a detailed career. In consumer ChatGPT, data handling depends on your account and settings; it is not the same framework as a business API with a processor agreement. Do not paste national ID, a full street address, or identity documents. Strip contact details if you only want writing help.

On ClarCV your CV is not pasted into chatgpt.com: it goes through OpenAI’s API for your analysis or rewrite, is not used to train general-purpose models, and is covered by a DPA. That does not mean “stored on no server”: the product needs an account and file so you can edit and download. It means we do not sell your data or use it for ads. Detail in the privacy and AI guide on this blog and at clarcv.com/privacy.

When to move from ChatGPT to ClarCV

Use the chat for a text draft. Use ClarCV for the next step: upload PDF or Word, get analysis of clarity and structure, and regenerate a version with templates built for ATS and A4 print. Start free; Pro is €7/month if you iterate often. UI in Spanish and English; output can follow the language of each application.

The honest shortcut: inventory → prompt with the ad → catch inventions → upload to ClarCV for a parseable file. Skipping the inventory and asking ChatGPT for a “senior data-driven CV” is the fastest way to a document that looks like AI and does not get interviews.