6 prompts that turn one job ad into a tailored CV, cover letter and interview prep
One chat, one job ad, six prompts: map the requirements, build an evidence bank, rewrite the bullets, draft the letter, prepare the interview, then rehearse. We ran each prompt once before publishing.
In this article
Most people open a chatbot, paste a CV and type "make this better". The model has no idea what the job wants, so it polishes the same generic lines that were already there. The fix is to make the ad the brief and your own history the only evidence the model is allowed to use.
That is the whole method: one chat per application, the job ad and your CV in front of the model, then six prompts in a fixed order. The first three stop it inventing things. The last three turn what survives into a letter and an interview.
Start here
Keep one conversation per job. In ChatGPT, make a Project for the role and upload the ad and your CV to it. OpenAI's help page for Projects says it accepts PDFs, spreadsheets, docs and images, and that the free plan allows 5 files per project. The same page says your previously saved memories are not referenced in project chats, which is what you want here: this chat should be about this job, not about everything you have ever told the assistant. The page was updated on 19 October 2026.
Two rules before the first prompt. First, the chat may only use facts you supply; every prompt below says so, and you should still check the output line by line. Second, strip what the application does not need: no ID numbers, no references' phone numbers, no full home address. A CV is personal data, and a project chat is not a sealed vault.
1. Map the ad before you touch the CV
What it does: separates what the employer says it needs from what it merely lists, and shows where your CV has nothing to offer. The gaps are the useful part.
Act as the hiring manager for this role. From the job ad and my CV, list every requirement. Mark each must-have or nice-to-have, quote the ad's wording, and name the closest real evidence in my CV. Write GAP where none exists. Invent nothing.
Tested: on a fictional junior analyst ad and a fictional CV it produced a nine-row table, quoted the ad, and marked two requirements GAP: a spreadsheet function and delivery-data cleaning. It also refused to treat the CV's "4,000 SKUs" as evidence of the employer's "4,000 parcels a day". Both refusals are the feature, not a failure.
2. Build an evidence bank
What it does: turns the must-haves into one concrete story each, with a number and a result, so the writing later has something true to draw on.
Build an evidence bank. For every must-have, pull one real example from my CV or notes: what I did, the number, the result. Use only facts I gave you. If I have no example, ask me one question instead of guessing.
Tested: it returned a three-row table with the numbers it could find and asked exactly one question about the missing spreadsheet skill instead of filling the hole with plausible fiction. Answer that question honestly, or leave the gap and plan how to address it in the interview.
3. Rewrite five CV bullets
What it does: makes the strongest lines match the ad's language without changing the facts.
Rewrite my five strongest CV bullets for this ad. Keep every number and fact exactly. Use the ad's vocabulary only where it is true. Start each with a verb, keep it under 20 words, and show before and after side by side.
Tested: five before-and-after pairs, all numbers preserved. With a thin sample CV the changes were small, and that is the correct result. A prompt that produces dramatic new bullets from a thin CV has probably invented them.
4. Draft the cover letter
What it does: three paragraphs that answer the employer's actual problem instead of praising the company.
Write a 120-word cover letter. Paragraph 1: the problem this team is hiring to fix, in their words. Paragraph 2: one proof from my evidence bank with a number. Paragraph 3: what I would do in week one. No 'passionate' or 'excited'.
Tested: it opened with the ad's own phrase "reporting is still manual", used the 18% figure from the CV, and proposed a first-week plan. It ran to roughly 160 words, so treat 120 as a target and cut it yourself. A letter that fits on one screen gets read.
5. Prepare the eight questions
What it does: predicts the interview from the ad, hardest question first, using only evidence you actually have.
List the eight questions this interviewer is most likely to ask, hardest first. For each, give a two-line answer built only from my evidence bank, in STAR shape. Mark any question where I lack evidence and say what to prepare.
Tested: eight questions, five marked GAP with a specific thing to prepare, and the remaining answers built from the CV's one measurable win. The GAP list is your study plan for the night before.
6. Rehearse out loud
What it does: turns reading into speaking, with a score after every answer.
Run a 20-minute mock interview for this role. Ask one question at a time and wait for my answer. After each, score it 1-5 for specificity, relevance and length, give one fix, then ask a harder follow-up. End with my three weakest answers.
Tested: it opened with why this role, asked for a 60-90 second answer, and waited. It works best typed, one answer at a time, or spoken if your app supports voice. The scores are a rehearsal device, not a prediction of the interview.
The order that saves time
Run prompts 1 and 2 first, then fix the real gaps in your CV before running 3. Do the letter with 4, and keep 5 and 6 for the day before the interview. If you have five minutes, run 1 and 5: the requirement map and the question list give you the most per minute.
What we did not test
Each prompt was run once, in a fresh chat, on an OpenAI GPT model through the Codex command line, against a fictional job ad and a fictional CV. We did not run them inside the ChatGPT app, on a phone, or on Claude or Gemini, so treat the exact formatting you get as variable. Plan limits and project features change; the Projects details above were checked on 21 October 2026.
More importantly, the model cannot check whether your achievements are real, cannot see the employer, and does not know the recruiter, the applicant tracking system or your local job market. It also cannot tell you whether you are a good fit. It can only stop you sending the same generic application to thirty jobs.
No prompt gets anyone hired. What this does is remove the excuse for a generic application, and give you a rehearsal partner that never gets tired of the same question.
Our articles help you explore AI and build practical skills. For a correction or a question, contact us. Contact
