GPT-6 Resume Prompts: Tested on Astra, Sol and Luna

Since October 7, GPT-6 has been rolling out as the model behind ChatGPT’s chat tab, on free and paid plans alike. So we ran the same resume through it to see what actually happens when you ask it to tailor a resume to a job posting. We used three prompts and all four GPT-6 models: Luna, Sol, GPT-6.1 Sol and Astra.

The short version: none of the models invented a single number, employer or tool. But a one-line “tailor my resume to this job” also barely changed the bullets. Below are the two prompts we recommend running in order, a gap check and then an honest rewrite, ready to copy, along with what each GPT-6 model did differently.

Which GPT-6 model are you actually using?

The model menu in ChatGPT just says “GPT-6”. Which model answers you depends on your plan. Here is what OpenAI’s October 7 announcement, pricing page, help center and API docs say:

OpenAI announced Astra on September 3, 2026, Sol and Luna on September 22, and GPT-6.1 Sol on September 29. GPT-6 started rolling out to the ChatGPT chat tab for Plus, Pro, Business and Enterprise on October 7 (for Enterprise it depends on admin settings) and to Free and Go on October 8. OpenAI’s own help pages still disagree in places about the Free plan, so check the model menu in your app.

How we tested GPT-6 on a resume

We wrote a realistic resume for a fictional marketing coordinator, Maya. It lists duties, not results: a monthly email newsletter, Mailchimp, weekly reports, social media. It has no results or metrics: no open rates, list sizes or percentages. Then we paired it with a fictional job posting for an Email Marketing Specialist at a home goods brand. The posting asks for Klaviyo, automated email flows, A/B testing, list segmentation and a 400,000-subscriber list. Maya’s resume shows none of those.

That gap is the point. A model that tries to please can fill it: invent an open rate, claim Klaviyo experience, add “increased revenue by 25%”. So the test was simple. Does the tailored resume contain anything that isn’t true?

We ran three prompts through the OpenAI API on October 14, 2026, with default settings and one run per prompt, on all four GPT-6 models. That made 12 runs. ChatGPT adds its own instructions on top of the model and answers vary from run to run, so treat the timings as a rough guide, not a benchmark. OpenAI also doesn’t say whether these API models are the exact versions inside the ChatGPT app, so your results there may differ.

Results: no invented numbers or tools, but not much tailoring either

Two things stood out across all 12 runs.

GPT-6 didn’t invent numbers, tools or employers. Even with no guardrails in the prompt, every model kept Klaviyo, A/B testing and segmentation off the resume, and all four hinted at the missing skills. Only GPT-6.1 Sol and Astra named Klaviyo outright. OpenAI says GPT-6 is “better at recognizing and communicating the limits of its own capabilities, including when it lacks the information” it needs (source). OpenAI publishes no test on resumes, though, and a few bullets did get small additions the resume doesn’t support, such as Luna’s “using Mailchimp” or Sol’s “track record of collaborating with designers”. Read every line before you send it.

A one-line prompt can’t tailor what isn’t there. With no results to work with, the models mostly polished Maya’s duties and added a summary aimed at the job. In GPT-6 Sol’s version, the first bullet reads “Manage the monthly email newsletter for members.” instead of “Run the monthly email newsletter for members”. No model could write “Grew the newsletter’s click rate from 2.1% to 3.4%”, because Maya never said so. Honest tailoring needs facts from you, and the routine below helps you find them first.

The two prompts to run: gaps first, then the rewrite

We tested each prompt on its own, on Maya’s original resume. We didn’t run the full routine with an updated resume, but the logic is simple: the model can only use facts you give it.

What most people type (and what it gets you)

Tailor my resume to this job.

JOB POSTING:
[paste the full job posting]

MY RESUME:
[paste your resume]

Every GPT-6 model returned a tidier version of the same resume plus advice to add results if they’re true. Luna’s version, for example, ended with “If accurate, consider adding specific results”. It’s fine as a formatting pass, but it’s unlikely to show a recruiter proof of the posting’s must-haves.

Step 1: the gap check prompt

Run this first. It doesn’t rewrite anything. It shows you which requirements your resume proves, which it only hints at, and which you simply don’t have.

Compare my resume with this job posting. Return a table with three columns: Requirement from the posting | Evidence in my resume (quote it, or "none") | What to do (rephrase / add a number / it's a real gap). Then list the 10 keywords from the posting I should use, marking the ones my resume can honestly support. Do not rewrite the resume.

JOB POSTING:
[paste the full job posting]

MY RESUME:
[paste your resume]

All four models produced a useful table. On Maya’s resume, all four flagged at least these gaps: lifecycle flows, Klaviyo, A/B testing and segmentation. Most also listed smaller ones, such as basic statistics, SMS and HTML/CSS for email. They also found rows where a number would help: list size, campaign cadence, the metrics in her weekly reports. Most of them also noted that Mailchimp counts toward the posting’s “Klaviyo or a similar ESP” requirement, so Maya should rephrase that line rather than treat it as a gap.

Before you move on, answer every “add a number” row with real figures and update your resume text. How many subscribers? How often did you send? What were the open and click rates? Did you ever test two subject lines? If you don’t know a number, leave it out.

Step 2: the honest rewrite prompt

You are a resume editor. Rewrite my resume for the job posting below.

Rules:
- Use only facts from my resume. Do not invent employers, tools, numbers or results.
- Where a number would help, write [ADD NUMBER: what to measure] instead of guessing.
- Mirror the posting's exact wording for skills I actually have.
- If the posting asks for something I don't have, list it under "Gaps to address" at the end instead of adding it to the resume.
- Keep it to one page, plain text, reverse-chronological, 3–5 bullets per role starting with a strong verb.

JOB POSTING:
[paste the full job posting]

MY RESUME:
[paste your updated resume]

Every model kept to the rules on numbers, tools and employers, and put the missing skills into a “Gaps to address” section. Luna added one small detail the resume doesn’t state (“using Mailchimp”). Sol, GPT-6.1 Sol and Astra put the placeholders right inside the bullets. Astra’s version reads “Run the monthly member email newsletter for [ADD NUMBER: newsletter subscribers].”, so you know exactly where your numbers go. Luna listed them at the end instead.

The downside of strict rules is that the rewrite stays close to your original wording. That’s why Step 1 matters: the real facts you add before Step 2 are what the rewrite can build on. For a manual method, see how to tailor your resume to a job description.

Model by model: what to expect

• GPT-6 Luna (Free and Go in ChatGPT). Quick and cautious. It added no numbers or tools, though it padded a few bullets, so compare its version with your original. Use the gap check prompt so it tells you what’s missing.
• GPT-6 Sol (Plus and Pro in ChatGPT). The quickest overall in our runs, with clean structure and a short list of things to add. Its output was as usable as Astra’s.
• GPT-6.1 Sol (API, ChatGPT Work). Clear coaching: it listed what it left out and why, then four things to add if they’re true. Astra did much the same.
• GPT-6 Astra (Pro option, Work and Codex). The most careful reasoning. For example, it noted that job dates alone don’t prove two years of email work, and it marked partly supported keywords separately. It also put a past job in the present tense twice, so proofread tense and dates. It was also the slowest overall, and it costs the most in the API.

GPT-6 resume tips, from OpenAI’s guidance and our test

• Give it a clear assignment. OpenAI’s guide to GPT-6 recommends stating “the result you want, who it’s for, the relevant context and constraints, and what counts as done”. For a resume, that means the full job posting (not just the title), one page, plain text, and no invented facts.
• Don’t over-specify style. The same guide notes that “overly specific guidance can now hinder results”. Set the rules that matter (truth, length, format) and let the model handle the wording.
• Upload a text-based file. ChatGPT reads PDF, DOCX and TXT files, but on non-Enterprise plans it extracts the text, not the images (OpenAI help). A resume built as a picture or a fancy two-column design may come through garbled. Pasting plain text is safest.
• Keep the gaps list. The “Gaps to address” section is your interview prep. Be ready for questions about the items on it.

Is it safe to put your resume into ChatGPT?

Your resume holds your name, contact details and work history. Here is what OpenAI’s help center says about where that goes:

• On Free, Plus and Pro, your chats may be used to train models unless you turn off Improve the model for everyone under Settings → Data controls (source).
• Even after you opt out, giving a thumbs up or thumbs down on an answer can send that whole conversation for training, resume included.
• A Temporary chat isn’t used for training and doesn’t create memories, but it may be kept for up to 30 days for safety reasons (source).
• Uploaded files can stay in your Library after you delete the chat. Delete them there too.

If you’d rather not share your phone number or address, replace them with placeholders before you paste. The model doesn’t need them to tailor your bullets.

Skip the prompts: how Dovira does the same job

The routine above works, but it’s a lot of copying and pasting for every application. Dovira is our product, built for this step. You add your resume once. Then you paste a job posting, and Dovira writes a resume for that posting: it picks the skills and keywords the employer asks for and rewrites your bullets around them.

• Your own rules. Add rules once, for example “don’t invent numbers”, and Dovira passes them to the AI with every resume. Read the result before you send it.
• Edit by chatting. Ask for “shorter bullets” or “more focus on email marketing” and see the change in the document.
• The rest of the application. Get a cover letter for the same job, answers to common interview questions and a board that tracks every application.
• Ready to send. Download a PDF.

Your first 3 tailored resumes and cover letters are free, with no credit card. Try it on the next job you want to apply for.

FAQ

Is GPT-6 good for writing a resume?

In our small test, yes, especially for honest edits. No GPT-6 model invented numbers, tools or employers, and all of them flagged skills the resume was missing. It can’t add achievements you didn’t tell it about, so give it real numbers.

Which GPT-6 model is best for resumes?

In our test, GPT-6 Sol, which Plus users get in ChatGPT, was the fastest and gave clean, usable rewrites. GPT-6.1 Sol and Astra gave the most detailed coaching, and Astra the most careful judgment. Luna, which Free users get, works well if you run the gap check first.

Can ChatGPT make up experience on my resume?

It can. GPT-6 didn’t invent numbers, tools or employers in our 12 API runs, but a few bullets got small unsupported additions, and OpenAI makes no promise about resumes. Check every number, tool and date before you send.

Do I need ChatGPT Plus to tailor my resume?

No. According to OpenAI’s October 7 announcement, Free accounts get GPT-6 Luna, and they can upload 3 files a day. Pasting your resume as text avoids the upload limit.

What’s the best ChatGPT prompt for a resume?

Run two prompts in order. First the gap check, which compares your resume to the posting without rewriting it. Then the honest rewrite, which forbids invented facts and uses placeholders where a number belongs. Both are above.