A catalog of resumes that ChatGPT wrote badly, the recruiter-grade rewrites that fix them, and line-by-line annotations explaining what changed and why.
This repo is a reading resource, not a tool. 33 full worked cases across 10 role categories, each one paired with a recruiter annotation. If you've ever asked "what's actually wrong with my AI-generated resume," every case here is one answer.
The same logic in this repo powers cvpage.org:
These cases were curated against the same credibility heuristics used by the production pipeline at cvpage.org — open-sourced here for anyone who'd rather learn from worked examples than blog posts.
| Folder | Cases | Roles covered |
|---|---|---|
| examples/engineering/ | 6 | Junior, mid, senior, staff, platform, frontend |
| examples/product/ | 3 | Junior PM, senior PM, growth PM |
| examples/marketing/ | 3 | Demand gen, content, junior |
| examples/sales/ | 3 | SDR, AE, sales leadership |
| examples/design/ | 3 | Product designer, senior designer, junior |
| examples/operations/ | 3 | RevOps, BizOps, People Ops |
| examples/data-analyst/ | 3 | Junior, mid-level, ML-leaning senior |
| examples/finance/ | 3 | FP&A, IB analyst, controller |
| examples/customer-success/ | 3 | CSM, onboarding, technical CS |
| examples/engineering-management/ | 3 | EM, senior EM, director |
| annotations/ | 2 docs | How recruiters read · What AI gets wrong |
Read these three to understand the pattern. The 18 others apply the same logic to different roles.
Before (ChatGPT default):
Spearheaded the implementation of a robust, scalable microservices architecture, driving operational excellence and optimizing system performance for cross-functional stakeholders.
After (recruiter-grade):
Split the checkout monolith into three services (orders, payments, fulfillment); owned the API contract with the mobile team.
Why the after works:
- Names the three services (recruiter can ask "what were the orders-service boundaries?")
- Names the actual collaborator (the mobile team)
- "Split" is a trusted, low-risk verb compared to "spearheaded"
- Zero buzzwords, 21 words instead of 28
Full annotated case: examples/engineering/02-senior-backend.md
Before:
Drove a 47% improvement in conversion through innovative, data-driven product enhancements and strategic cross-functional initiatives.
After:
Rewrote the trial-to-paid conversion flow; A/B test showed a +47% lift in paid conversion over four weeks (significance: p<0.01).
Why the after works:
- 47% is now anchored to a specific test, with the duration and significance level
- "Innovative, data-driven product enhancements" → named the actual artifact (the conversion flow)
- "Cross-functional initiatives" → deleted
- A recruiter reading this can ask exactly one follow-up: "what was the variant?" and the candidate can answer
Critical note: if the 47% was a guess, the after should be:
Rewrote the trial-to-paid conversion flow as one of three Q3 onboarding experiments.
This is the single most-faked PM bullet shape. Full annotated case: examples/product/03-growth-pm.md
Before:
Leveraged consultative outbound prospecting techniques to drive a robust pipeline and consistently exceed activity quotas.
After:
Booked 24 meetings per quarter on average across FY24 (team baseline: 18); top of the team for two of four quarters.
Why the after works:
- "Consultative outbound prospecting" → named the deliverable (meetings booked)
- Anchored against the team baseline (18) — a credibility multiplier
- "Top of the team for two of four quarters" — defensible, non-fabricated, recruiter-meaningful
- "Robust pipeline" → deleted; meeting count is the real KPI
Full annotated case: examples/sales/01-sdr.md
Each case file follows the same five-section structure:
- The role context — who wrote it, what they were applying for
- The ChatGPT-default version — what came out of "ChatGPT, write me a resume for X"
- The recruiter-grade version — what the rewrite looks like
- Line-by-line annotations — what changed in each bullet and why
- What still wouldn't survive an interview — the bullets that are still risky even after the rewrite
The fifth section is the one most resume-rewriting resources skip. It's where the real recruiter judgment lives.
If you only have ten minutes:
- annotations/how-recruiters-read.md — the actual reading path a recruiter uses in the 6-second scan, with annotated examples
- annotations/what-ai-gets-wrong.md — the eight most common LLM failure modes on resumes, with examples of each
- Job seekers who got their resume back from ChatGPT and aren't sure why it feels off
- Recruiters and resume writers building their own example library
- Bootcamp and career-coaching curriculum builders (the examples are CC-friendly under MIT)
- Researchers studying LLM failure modes on resume tasks
- Not a templates library. Resume templates are a different problem. This repo is about rewriting content.
- Not a fictional showcase. The "ChatGPT-default" versions are real outputs from GPT-4 on real resume rewrite prompts. The "after" versions are the kind of rewrite a senior recruiter would produce.
- Not a guide to faking experience. Several cases explicitly call out where a metric was invented and how to honestly retract it.
- Add data / analytics roles (3 cases)
- Add customer success roles (3 cases)
- Add engineering management roles (3 cases)
- Add career-switch case studies (junior PM coming from teaching, junior SWE coming from finance, etc.)
- Per-case downloadable PDF showing the before, after, and diff side by side
Contributions welcome. New cases should follow the existing five-section structure and respect the credibility rules from annotations/what-ai-gets-wrong.md.
Part of an open resource set for resume credibility, ATS, and recruiter-grade rewriting:
- ai-resume-humanizer-prompts — prompt library for rewriting AI-sounding resumes
- chatgpt-resume-before-after-examples — annotated bad-vs-improved resume samples
- recruiter-resume-audit-framework — 5-step framework recruiters actually use
- ats-resume-keyword-dataset — structured ATS keyword dataset by role
- awesome-ai-resume-resources — curated list of prompts, tools, and recruiter resources
Try the full recruiter-grade audit pipeline (Analyze → Critique → Rewrite, no fabricated metrics):
Specific tools:
- Resume Humanizer — rewrite ChatGPT-sounding bullets
- ATS Resume Checker — keyword + parse audit
- Resume Credibility Checker — recruiter-style audit
- ChatGPT Resume Fix — fix AI-generated drafts
- Rewrite Resume Bullets — line-by-line rewriter
Deep-dive articles that pair with this repository:
- Why recruiters hate AI-generated resumes
- Signs your resume sounds like ChatGPT
- AI resume vs human-written resume
- Why AI resumes feel generic
- How recruiters spot fake resumes
- What recruiters notice in 6 seconds
- Resume credibility signals
MIT — free to fork, adapt, and use commercially. Attribution appreciated but not required.
Issues and PRs welcome. Please keep the quality bar high: examples must be realistic, prompts must be tested, and no keyword-stuffed contributions.