How to Tailor a Resume to a Job Description Using ChatGPT
The prompt that works is not “rewrite my resume”. It is a two-pass: compare first, then edit only what you can evidence. Here are the prompts, and when a job-link tool is faster.
Most ChatGPT resume advice fails for the same reason: the prompt is too vague. “Improve my resume for this job” produces generic corporate filler and, worse, tools and metrics you never used. A good tailoring prompt does three things. It gives the model the job description, it constrains the output so it cannot invent facts, and it asks for a comparison before any rewrite.
This is the two-pass method, with copy-paste prompts for ChatGPT and Claude, plus the one case where pasting a job link is simply faster.
How to tailor a resume to a job description using ChatGPT
Work in this order. Skipping the comparison is how the model inflates your experience.
- 1Strip contact details from the paste. ChatGPT does not need your phone number.
- 2Paste the full job description, including requirements hidden behind “see more”.
- 3Run the gap-analysis prompt below. Do not rewrite anything yet.
- 4Rewrite one section at a time: summary, then skills order, then the top bullets of the most relevant role.
- 5Run the sceptic prompt on the finished draft before you export it.
Prompt 1: gap analysis (do this first)
Copy this into ChatGPT. Replace the bracketed blocks. This is the prompt that answers “tailor resume to job description prompt” without handing the model a blank cheque.
Read the “do not add” column before you touch a sentence. Those rows are either a cover-letter topic or a sign this role is a stretch.
Prompt 2: rewrite the professional summary
Only after the table exists. Keep it to two or three lines. Name the target title and two evidenced requirements.
Prompt 3: rewrite three bullets (not the whole job)
ChatGPT handles five focused bullets far better than “rewrite everything”. Ask for before-and-after so you can see what changed.
Prompt 4: the sceptic pass
This is the one that catches invented metrics. Run it on the finished draft, not on the original.
How to tailor a resume to a job description using Claude
Use the same four prompts. Two Claude-specific habits help:
- Re-paste the job description in every rewrite message. Long chats lose details.
- Repeat the constraint in the same message as the rewrite: “Do not invent metrics, tools, titles, or employers.” Claude follows a nearby rule more reliably than one from ten turns ago.
If you already have a Claude project with your master CV loaded, still paste the specific JD and the three bullets you want changed. Project memory is not a substitute for a tight prompt.
What ChatGPT cannot do (and a job link can)
The method above works. The arithmetic does not, once you apply weekly. You still have to find the posting, copy it, paste sections, check for hallucinations, and put the result into a layout that an ATS can parse.
| ChatGPT or Claude | From a job link | |
|---|---|---|
| Reads the posting from a URL | No — you copy-paste | Yes |
| Works from a structured CV | Only if you paste it each time | Yes, uploaded once |
| Invented tools and metrics | Common unless you constrain it | Blocked: only your evidence |
| ATS-safe PDF export | You lay it out yourself | Single-column PDF |
| Cover letter from the same parse | A separate chat | Same posting, same pass |
| Sustainable at 10 applications/week | Rarely | Yes |
If you want the manual method written out without a model in the loop, use the step-by-step tailoring guide. If you want the honest ChatGPT-versus-tool split, that is the FitMyCV vs ChatGPT comparison. If the bottleneck is producing a first draft per posting, paste the job link instead.
Skip the copy-paste loop on the next role
Upload your CV once, paste the job URL, and get a tailored resume and cover letter you can edit. Same honesty rule: nothing is invented.
The checklist
- Compare before you rewrite. Unsupported rows stay off the resume.
- Edit the summary, skills order, and three bullets — not the whole document in one prompt.
- Keep your original numbers. If a metric is missing, you supply it; the model does not.
- Run the sceptic pass. Delete anything it flags.
- Export from a real editor and run the ATS checker against the same posting.
