I sent 47 applications with my old resume. 0 callbacks.

Then I tried 10 ChatGPT prompts I'd been collecting. I sent the next 12 applications with rewritten resumes. 4 callbacks. Here are the prompts that actually worked, and the ones that wasted my time.

The pattern that worked: every prompt that produced something useful was specific about output format and gave the model permission to leave things blank. Every prompt that failed was vague.

Prompt 1: STAR with [needs metric] markers

"Rewrite my resume bullets using the STAR method (situation, task, action, result). Use the job description below as context. If a bullet doesn't have a measurable result, mark it [needs metric] instead of inventing one."

This was the unlock. The "[needs metric]" instruction stopped ChatGPT from hallucinating numbers. Half my old bullets came back marked, which forced me to actually go check Slack archives and find real outcomes. The other half came back tightened.

Prompt 2: Keyword gap analysis (don't add — identify)

"Read the job description. List the 10 most important keywords (skills, tools, methodologies) it uses. Then check my resume and tell me which keywords I'm missing or underusing."

Don't ask the AI to add keywords. Ask which ones are missing. Then you decide which ones genuinely apply.

This is critical for ATS without resume-stuffing — the bot detects keyword density above ~2-3% and downgrades you for it.

Prompt 3: 3-sentence summary cap

"Rewrite my summary in 3 sentences max. Sentence 1: who I am professionally. Sentence 2: my biggest measurable achievement. Sentence 3: what I'm looking for next."

The 3-sentence constraint is what makes this work. Without the cap, ChatGPT writes corporate filler ("results-driven professional with a passion for...").

Prompt 4: ❌ "Make my resume more impressive"

The one that failed for me. Vague prompts produce vague output. Skip.

Prompt 5: Multi-dimension scoring

"Compare my resume to the job description. Score the match 0-100 across these 5 dimensions: keyword overlap, experience relevance, seniority match, industry fit, recency. Explain each score in one sentence."

Forces structured feedback you can actually act on. "Your industry fit is 30/100 because the JD mentions 'fintech' 4 times and your resume has none" is fixable. "Your resume could be better" is not.

Prompt 6: Weakness audit

"What's the weakest bullet on my resume? Why is it weak? Suggest a rewrite if I provide more context — but if you don't have enough context, ask me 1 specific question."

The "ask me a question" branch is what makes this useful. Most ChatGPT outputs are confidently wrong. This forces it to admit when it needs more info.

Prompts 7-10

I'll spare you the full list. The structure is what matters:

• Specific output format (numbered, capped, marked)
• Permission to leave things blank ([needs metric], "I don't know")
• Cap on length (3 sentences, 5 bullets, 100 words)
• Compare-don't-fix framing (score it before fixing it)

After spending a weekend running these manually...

I built ResumeAI to do all 10 automatically. Same prompts (refined), career-situation branching (switcher / returning / new grad / senior), and 23-criteria ATS scoring built in.

It's $9 once. No subscription.

If you want to do it manually, the 5 prompts above are enough. If you want it in 30 seconds, that's the link.