AI-Search SEO Kit: Get Cited by ChatGPT, Perplexity & Google

Make your site the easiest source for AI search to quote, and prove it with PASS/FAIL checks.

Most SEO skills hand your agent a checklist. This one ships the tooling: tell your agent "Audit my site for AI search and fix what you can" and it scores every page for AI-citation readiness, generates the files ChatGPT search, Perplexity, Google AI Overviews, Copilot and Claude read, validates every one of them, writes answer-first rewrite briefs for the copy, and injects the markup into your HTML with a backup. On the bundled demo site the score goes 32 → 65 with the scripts alone, and 99 after the answer-first rewrite.

What you get

• Audit of a local site folder, a saved page or (only when you ask) a live URL: 30 page checks and 12 site checks in four pillars (AI answer-readiness, crawl & index, structured data, trust/E-E-A-T). Answer-first intros, question-style headings, fact density, citations, bylines, visible dates, FAQ blocks, canonical, meta, headings, alt text, broken links, sitemap coverage.
• A report you can hand to a client: self-contained, printable HTML with score ring, pillar bars, before → after deltas, an AI-crawler access table, a page-by-page matrix and every fix ranked by its simulated point gain with the exact snippet to paste. Plus audit.json and fixes.md.
• AI-crawler policy, done right: 21 crawlers mapped from the operators' own docs (OAI-SearchBot vs GPTBot, PerplexityBot, Claude-SearchBot vs ClaudeBot, Google-Extended, Applebot-Extended, CCBot...). Search bots in so you can be cited; training bots follow your choice. RFC 9309 matching, and your private paths survive.
• Generators: schema.org JSON-LD per page (Organization/LocalBusiness with 100+ business subtypes, WebSite, Article/BlogPosting with author, FAQPage, HowTo, Product with Offer/AggregateOffer, BreadcrumbList, Person, AboutPage), llms.txt + llms-full.txt (llmstxt.org format), sitemap.xml and robots.txt, all from a small YAML profile and your real pages. Nothing is invented.
• Validators that print PASS/WARN/FAIL: JSON-LD against Google's required/recommended properties, URL/date/price/currency formats, @id integrity, and a check that FAQ answers are really visible on the page; llms.txt linter; robots.txt checker; sitemap validator.
• Answer-first rewrite briefs per page: the question to answer, suggested question headings, current fact density, and acceptance checks. Your agent writes the copy; the audit verifies it.
• Safe inject: dry run, backup outside the site, idempotent marker blocks, never overwrites your own tags, one-command restore.
• One-command demo on a 5-page sample business, so your first run shows the whole pipeline in about a second.

Why it beats prompting

Honest by design

Every rule in references/ cites its source (Google Search Central, OpenAI/Anthropic/Perplexity/Apple/Amazon/Meta crawler docs, llmstxt.org, RFC 9309, sitemaps.org, the GEO research paper). Heuristics are labelled as heuristics. It tells you that Google retired FAQ and How-to rich results and that llms.txt is a proposal. No promises of rankings or citations, and no invented facts, reviews or dates.

What's inside

SKILL.md (10-step workflow, report template, quality checklist) · 9 Python scripts (audit engine, report, generators, validators, injector, demo) · 6 reference guides (AI crawler table, schema requirements, llms.txt spec notes, AI-citation checklist with evidence labels, scoring, profile format) · demo site, sample rewrites, demo profile and pre-built before/after reports.

Requirements

An agent that can run code with Python 3.8+. Standard library only: no installs, no API keys. Offline except the optional live-URL audit.