Tailoring your résumé with AI? Make it list the gaps instead.
The vague prompt
Tested in GPT-5.5 (OpenAI API), Oct 2026
A common way: paste a résumé and a job post, then type this. Jordan Reyes, a fictional office coordinator, lists: phones and booking, a supply spreadsheet, onboarding 4 staff, weekly reminder emails, fixing scheduling conflicts.
What the vague prompt wrote
Real output · GPT-5.5 (OpenAI API) · Oct 2026
Jordan's original says “Helped onboard 4 new front-desk staff.” The AI turned that into training, and added a result nobody measured. (“…” marks text cut for space.)
Four things the résumé never said
Checked against the 5 original bullets, the vague prompt added all four of these. It also dropped “Answered phones” and offered Excel, budget and Salesforce bullets “if you have experience.”
Why it fills in the blanks
With only “match this job post” to go on, the AI filled the gap. Anthropic's guide (written for Claude) says to use only the documents you give it. We tested that idea in GPT-5.5.
Source: Anthropic Claude Platform Docs, Oct 2026
The prompt that sets limits
Tested in GPT-5.5 (OpenAI API), Oct 2026
This version limits the facts, asks for a gap list and makes the AI quote the original behind every bullet.
Every bullet has a receipt
Real output · GPT-5.5 (OpenAI API) · Oct 2026
Every fact now has a quoted original behind it, and 8 gap items from the job post's requirements went on a gap list instead of into the bullets.
The safe version is flatter
It merged three bullets into one long line and left “Helped onboard” unchanged. It also listed training as a gap, so whether onboarding counts is your call.
Try it: Also: remove personal details before pasting, then check each bullet against its quote.
Sources and assumptions
- Reduce hallucinations, Claude Platform Docs (Anthropic): 'External knowledge restriction: Explicitly instruct Claude to only use information from provided documents and not its general knowledge.' 'Allow Claude to say "I don't know": Explicitly give Claude permission to admit uncertainty.' 'Verify with citations… find a supporting quote… If it can't find a quote, it must retract the claim.' 'while these techniques significantly reduce hallucinations, they don't eliminate them entirely. Always validate critical information.' (checked 2026-10-15)
Assumptions:
- The résumé (Jordan Reyes, Brightline Dental Group) and the job post (Harbor Lane Clinics) are fictional samples written for this test.
- The tests ran once each in GPT-5.5 through the OpenAI API on 2026-10-15/16 UTC, not in the ChatGPT app. A rerun may word things differently.
- The Claude CLI could not run from the research session (both claude attempts failed, recorded with output null), so no Claude result is claimed. The Anthropic doc is cited for the general technique only, not as proof of a Claude result.
- Counting 'helped onboard' as a training gap is the model's conservative call. Whether onboarding counts as training is the reader's honest judgment.
The short version
- Tell it to use only your résumé
- Ask for a quote behind every bullet
- Have it list gaps, not fill them
- Still check every bullet yourself


