I let three AIs argue over my resume. Here's what they caught.

Asking the same AI to check its work

Earlier this year I was job hunting and doing what everyone does: paste the resume, paste the job, ask the AI to make me shine.

It made me shine a little too brightly. It used words I never use, like "spearheaded", made claims a size bigger than what I'd done, and echoed the job post so neatly it felt like cheating on a test.

Then I asked the same AI to check its work. It said the draft was strong.

Researchers have measured this. In one 2026 study, AI judges could be more than 50% more likely to pass a failed answer when the answer was their own.

Sixteen prompts and a notes file

So I got a second opinion: a different company's AI, in a fresh chat, told to read the draft like a skeptical hiring manager.

It was far more willing to point out problems. So I kept going, all summer:

• May 2026: prompts to decode the job, screen my resume, and rewrite the weak bullets, run in three different AIs.
• June and July 2026: all three reviewed the merged result, and two of them researched the role itself.

The resume I ended up with was the best I'd had. It also took hours, three browser tabs, and a notes file just to remember which reply came from which AI.

Nobody else was going to spend hours on that, so I turned it into a skill.

What Resume Tribunal does now

A skill is a set of instructions your AI app loads when it needs them. You paste one message into ChatGPT, Gemini, Claude or any AI, and it knows the whole routine. Try it: you do two steps, and the AIs do the rest.

Your turn at steps 1 and 5. The AIs do the rest.

Before anything goes to a second AI, your name and contact details are swapped for stand-ins. The facts you confirm can be saved on your computer, or handed to you as a block to paste, so your next job takes a few questions instead of a full interview.

Testing the page on someone who isn't technical

I asked AI reviewers to pick the project apart, each playing a different kind of reader. The most useful one played someone whose whole AI experience is typing a question into ChatGPT.

That reader gave up within 30 seconds, because the first thing on my page was a terminal command. So I rebuilt the page around one message you paste into any AI chat.

What the second AI caught

I then had a fresh AI run the whole thing on a made-up candidate. The other AIs suggested more than twenty things the candidate never said. A few:

• "from 4 hours to 40 minutes", when the candidate said "half a day"
• "framework adopted by leadership", for a spreadsheet one supervisor used
• dropping "simple" from "simple SQL queries", four separate times

All of them were turned down, because the candidate's own words decide. You can check every line yourself.

Three things I'd tell anyone using AI for a resume

1. Get a second opinion from a different AI. The one that wrote it tends to approve it.
2. Never keep a number you didn't give it. If you can't explain it in an interview, it can hurt you.
3. Keep your own words. Words like "spearheaded" stand out to readers, and they're probably not words you use.

Resume Tribunal is free and open source. Try it in your AI app, and tell me what broke.