Generative Engine Optimization

Generative Engine Optimization (GEO)

Getting cited by AI answer engines is a different game from ranking on Google. This skill is the playbook for making a site the source an LLM quotes when a user (or their agent) asks a category question like "what are custom domains for SaaS" or "best tool for X", not just when they search the brand name.

The core reframe to give the user: branded discovery ("find MyProduct") is easy; intent discovery ("find a solution to my problem") is the valuable, hard part, and it now spans three surfaces at once — classic search results, AI answers (ChatGPT/Perplexity/AI Overviews), and AI agents shopping for a tool.

Ground truth (cite these; they kill common myths)

Lead with evidence, because most GEO advice online is hype. The load-bearing facts (verify with a quick web search if challenged, they shift):

• Crawlability is lever #1. If robots.txt blocks AI bots or the site is client-rendered with no HTML, nothing else matters. On-demand fetchers (ChatGPT-User, Claude-User, Perplexity-User) obey robots.txt.
• Bing indexing gates ChatGPT. ChatGPT search runs on Bing; studies show ~87% of its citations match Bing's top organic results. Google rank does NOT drive ChatGPT citations. So verify + submit the sitemap in Bing Webmaster Tools, not just Google Search Console.
• llms.txt is largely decoration for AI chat. Server-log studies (Ahrefs, OtterlyAI) show ~97% of llms.txt files get zero requests, and Google says it does not use it. Its one real consumer is coding agents (Claude Code, Cursor). Publish it for docs/agents, not to get cited in chat. Do not oversell it to the user.
• Structured content gets extracted disproportionately. AI pulls comparison/HTML tables ~81% vs ~23% for prose, and FAQ-schema pages earn roughly 3x more citations. First ~30% of a page is where most citations come from — so answer first.
• Reddit is the #1 cited source across major engines (~40% frequency), and G2/Capterra are top-cited for B2B. Off-site presence matters as much as on-site.
• There is no magic force-index button. Google's Indexing API is officially JobPosting/BroadcastEvent-only; off-label use risks revocation. Sitemap "ping" was removed in 2023. Do not chase these.

The workflow

1. Map the demand (keyword-for-intent research)

Do not start from the brand. Enumerate the queries buyers and their agents actually type, by awareness stage:

• Problem-aware: "how do I let users use their own domain", "SSL cert stuck pending".
• Solution-aware: "custom domains as a service", "managed X for SaaS".
• Vendor-aware: "X alternative", "best X tools 2026" (respect any no-competitor-names brand rule; you can still win the category term without naming rivals).
• Dev/agent: "API to do X", "MCP server for X".

Research the real phrasings with web search: Google autocomplete, "People Also Ask", related searches, and community threads (Reddit, Hacker News, Indie Hackers, Stack Overflow). Note who currently ranks and who AI cites for the top queries — that is who you displace or get named alongside. Fan this out across a few subagents if available.

2. Build answer-first content that AI can lift

The cornerstone is a pillar page for the head category term, plus supporting clusters. Structure every page so an LLM can extract a clean answer:

• Open with a 40-60 word direct answer (BLUF) before anything else.
• Use question-shaped H2s that mirror the target queries, each opening with a 1-2 sentence answer, then detail.
• Include at least one HTML <table> (build-vs-buy, feature/price, decision matrix).
• Add stats + dates + a cited source where relevant.
• End plainly; no keyword stuffing (Google explicitly names AI-specific rewriting and tiny-chunk fragmentation as ineffective).

Content assets that rank and get cited, in priority order: pillar guide, glossary/definitions (own the category vocabulary), per-use-case pages, dev how-to guides, and (if the brand allows) comparison/alternatives pages with tables.

3. Add schema.org (machine-readable = citable)

Put a JSON-LD @graph in the page. The high-value types:

• FAQPage (mirror the on-page Q&A, 8-12 real buyer questions) — the ~3x citation lever.
• SoftwareApplication (applicationCategory, operatingSystem, description, provider, featureList; add offers/PriceSpecification only if you have real prices — do not invent them).
• Organization (with sameAs to LinkedIn/Crunchbase/GitHub) and WebSite (SearchAction).
• Article/BlogPosting with author + dateModified on posts.

4. Guarantee crawlability + submit

• robots.txt: no Disallow: /, and a Sitemap: directive with the canonical host. For max reach, give AI bots explicit Allow groups (ChatGPT-User, Claude-User, Perplexity-User, OAI-SearchBot, PerplexityBot, GPTBot, ClaudeBot, Googlebot, Bingbot) plus a Cloudflare-style Content-Signal: search=yes, ai-input=yes, ai-train=yes line.
• Sitemap: canonical, same-host, indexable URLs only; accurate <lastmod>.
• Owner-only console steps (say clearly these need the user's login): submit the sitemap in Google Search Console AND Bing Webmaster Tools; URL-Inspection "Request Indexing" for the homepage + top ~10 pages.

5. Earn off-site citations

Because Reddit + review sites dominate AI citations: seed genuine, non-spammy answers on the Reddit/Indie Hackers/DEV threads AI already cites for the category; get listed on G2/Capterra in the right category; and get included in the "best X" roundup listicles that rank. This is slower but often the highest-ceiling lever.

6. Measure

Track rankings + which URLs get cited: Bing Webmaster's AI Performance report (Copilot citations), and prompt ChatGPT/Perplexity with the buyer queries and log which domains they cite. Iterate on the gaps.

Common pitfalls

• Optimizing only for Google while ignoring Bing (which gates ChatGPT).
• Over-investing in llms.txt expecting chat-citation lift.
• Prose-only pages with no tables/FAQ schema (poor extraction).
• Client-side-rendered pages AI can't read (verify with view-source, not devtools).
• Treating "OKF" (Google Open Knowledge Format) as an SEO tool — it is an internal-agent knowledge spec, unrelated to indexing.

Related skills

For a full crawlability/indexing setup behind a reverse proxy, see reverse-proxy-cms-indexing. For making an MCP server findable by agents, see mcp-server-discoverability. For an "Ask AI about us" on-site CTA, see llm-deeplink-widget.