A doctor's visit for your AI visibility problem

What it does

GEO Doctor is a bilingual (Chinese/English) agent skill for AI assistants like Claude Code, Codex, Kimi Code, and WorkBuddy. It runs an eight-step workflow: audit whether AI crawlers can actually read your site, build a verified fact card from your own materials, generate 20–30 real user questions, sample how target AIs currently answer them, then diagnose the gaps — which competitors get recommended, how often, and which sources the AI cites. From there it batch-writes content (one piece per gap, facts only), publishes to your site or a generated GitHub Pages fact-site, and re-tests to measure whether your recommendation rate moved.

The interesting bit

The methodology is unusually honest for the GEO/LLM-SEO space. It insists you sample answers inside the actual chat apps — because that’s what your users see, and API answers often differ — and asks each question twice to track occurrence rate rather than a single ranking. The FAQ explicitly refuses to promise recommendations and warns against “GEO poisoning” (spammy mass posting), with scripts that block duplicate rephrasing and unverified claims.

Key highlights

• Zero-dependency audit script that simulates AI crawler identities to check if your firewall blocks them — runs fully offline, no server, no data uploaded
• Competitor recommendation leaderboard with per-question gap analysis and cited-source tracking
• geo-studio workbench: local browser UI for reviewing drafts, confirming facts, and managing a channel-by-channel publishing matrix
• Optional weekly “autopilot” via GitHub Actions: re-sample, draft new gap content, open a PR for your review
• If your site is unreadable to AI, it generates a plain fact-site with llms.txt and structured data, deployable to GitHub Pages

Caveats

• Doubao, DeepSeek, and other plain chat apps can’t install skills — you diagnose in a skill-capable assistant, then manually sample in the target app (15–20 minutes per round)
• Third-party platforms like Zhihu and WeChat are deliberately human-in-the-loop: it formats drafts, you click publish
• Currently supports software/apps/websites/online products only; a local-storefront version is “on the way”

Verdict

Worth a look if you run a product and suspect AI assistants are misquoting or ignoring it — this is one of the few GEO tools that shows its methodology and its limits. If you wanted a one-click “guaranteed AI recommendation” machine, the README will disappoint you, which is exactly the point.