--- name: ai-seo description: Optimize content to be cited by AI search engines and assistants (AI Overviews, ChatGPT, Perplexity, Claude, Gemini, Copilot). Use when the user says AI SEO, AEO, GEO, LLMO, answer engine optimization, AI citations, AI visibility, llms.txt, or "show up in AI answers". Not for traditional technical or on-page audits: use seo-audit. Not for structured data implementation: use schema. Not for building the content plan itself: use content-strategy. license: MIT metadata: author: TechTide AI (Alex Cinovoj) provenance: rewritten from patterns in the Corey Haines marketing-skills pack (MIT) category: SEO & AEO --- # AI SEO Traditional SEO gets you ranked; AI SEO gets you cited. AI systems extract passages, not pages, and they pick sources for structure, authority, and presence, not just rank position. When tactics conflict, write for people and organize for clarity: that satisfies Google and the extraction-driven engines at once. ## Before starting If a product marketing context file exists (`.agents/product-marketing.md` or `.claude/product-marketing.md`), read it first. Then confirm: the top 10-20 queries that matter, whether the brand appears in AI answers for them today, which competitors get cited instead, and whether the goal is optimizing existing pages or creating new ones. ## Know the two camps **Google (AI Overviews, AI Mode):** rooted in core Search ranking. No special markup or AI files required. Do not chunk content into AI-bait fragments and do not write separate "for AI" variants; that risks the scaled content abuse spam policy. Win with E-E-A-T, original information, semantic HTML, clean indexability. Also plan for query fan-out: Google generates related sub-queries under the hood, so cover the full topical cluster, not one keyword per page. **Extraction engines (ChatGPT, Perplexity, Claude, Copilot):** actively reward extractable structure, parse llms.txt and machine-readable files, and cite third-party sources heavily. Layer structural tactics on top of the Google baseline. They help these engines and do not hurt Google. ## The three pillars | Pillar | Job | Core moves | |---|---|---| | Structure | Make it extractable | Lead each section with a direct answer. Keep key answer passages 40-60 words. H2/H3s phrased as queries. Tables for comparisons, numbered lists for processes, one idea per paragraph. Definition, FAQ, and pros/cons blocks. | | Authority | Make it citable | Cite sources with links. Specific statistics with dates. Expert quotes with name and title. Named authors with credentials. Visible "last updated" date and real refreshes. | | Presence | Be where AI looks | Wikipedia accuracy, authentic Reddit and Quora participation, review platforms (G2, Capterra for B2B), industry roundups, YouTube for how-to queries. Third-party mentions often out-cite your own domain. | Authority tactics ranked by measured impact (Princeton GEO study, KDD 2024): citing sources, adding statistics, and adding quotations gave the largest visibility boosts; fluency plus statistics was the best combination; keyword stuffing measurably reduced AI visibility. Read references/citation-tactics.md when prioritizing authority work or choosing content formats. ## Machine-readable files For non-Google engines and buying agents, add to the site root: - `/llms.txt`: short overview of what the product does, who it is for, links to key pages (llmstxt.org format). - `/pricing.md`: plain-markdown tiers with prices, limits, and inclusions. Agents skip products whose pricing hides behind JavaScript or "contact sales". - Schema markup (`Article`, `HowTo`, `FAQPage`, `Product`, `Organization`): implement via the schema skill. Keep these files current. Stale pricing is worse than no file. ## Agent-readable pages Autonomous agents read sites through rendering, DOM, and the accessibility tree. Requirements: meaningful content renders without heavy JS, semantic HTML (`
`, `
`, real headings, alt text), labelled interactive elements, and pricing, specs, and contact info on public indexable pages. ## Do not 1. Write separate content "for AI". Same content serves people and machines. 2. Chunk pages into fragments. Normal headings and paragraphs. 3. Mass-generate thin variants for ranking. 4. Fabricate mentions or spam Reddit/Wikipedia. 5. Block the crawlers you want citations from (GPTBot, PerplexityBot, ClaudeBot, Google-Extended). Block training-only crawlers if you must, not search-and-cite ones. 6. Gate your most authoritative content. AI cannot cite what it cannot read. ## Monitoring Monthly, minimum. Manual method: run your top 20 queries through ChatGPT, Perplexity, and Google. Record whether you are cited, who is, and which page. Log and compare month over month. Tools (Otterly, Peec, ZipTie, LLMrefs) add share-of-voice and sentiment at scale. There is no AI-specific Search Console reporting; standard GSC metrics are the Google-side measure. ## Good vs bad **Bad:** "Our platform is the best-in-class solution trusted by industry leaders." Unverifiable, no data, no structure. Never gets cited. **Good:** "Acme reduces onboarding time from 14 days to 3 (internal benchmark across customers, updated ). Compared with Competitor X, Acme includes SSO at the base tier." Specific, dated, comparative, extractable in one passage. ## Verification Pick one optimized page. Fetch it with JavaScript disabled (curl or a text browser). Expect the main answer content, headings, and pricing to be present in the raw HTML. If not, the page is invisible to extraction engines; fix rendering before touching copy. Then check robots.txt: expect GPTBot, PerplexityBot, and ClaudeBot are not disallowed. If they are, unblock or accept zero citations from those platforms. ## Completion checklist - [ ] Target queries listed and current AI answers checked for each - [ ] Key pages lead sections with 40-60 word direct answers - [ ] Claims carry statistics, sources, or named expertise - [ ] llms.txt and pricing.md present and current (where applicable) - [ ] Schema in place via the schema skill - [ ] AI crawlers unblocked in robots.txt - [ ] Monitoring cadence set with a baseline recorded Any box unchecked: not done. Fix or say so. ## Footguns - **Treating AI SEO as separate from SEO.** Extraction tactics on a page that cannot rank or index do nothing. Fix: baseline technical SEO first (seo-audit), then layer structure and authority. - **Undated content.** AI systems weight recency; a page with no dates loses to a dated equivalent. Fix: visible last-updated dates and quarterly refreshes on competitive topics. - **Optimizing only your own domain.** Many engines cite third-party sources more than brand sites. Fix: spend real effort on reviews, communities, and roundups, not just on-page tweaks.