Generative search optimization (GEO) basics

Generative Search Optimization (GEO) is the practice of making your content cited and recommended by AI assistants – ChatGPT, Gemini, Perplexity, Claude. It overlaps with SEO but the ranking signals are different. This page covers the foundations.

GEO vs SEO

The four GEO levers

1. Be in training data

LLMs are trained on the public web. Pages that existed and were indexed when the model was trained are candidates for citation. There's no opt-in mechanism; the path is to have your content published, indexed by Google, and ideally cited by other authoritative sources before the next training cutoff.

2. Be retrievable at runtime

Most LLM assistants now do real-time web search ("retrieval-augmented generation"). When a user asks a question, the model searches the web and grounds its answer in the results. Being SEO-discoverable on the underlying search engine (Google for ChatGPT/Gemini/Claude, Bing for Copilot, custom for Perplexity) is half of GEO.

3. Have an llms.txt that points to your best content

llms.txt is an emerging standard that lets you tell LLM crawlers which pages on your site are most important. Not all models respect it yet, but ChatGPT and Claude do, and the cost of publishing one is near zero. See "Writing llms.txt for your site".

4. Write in citation-friendly shapes

LLMs prefer to cite passages that have specific structure: clear questions answered directly, short paragraphs, bulleted lists, named sources, dated stats. See "Improving your AI-readability score".

How to measure GEO success

• Daily snapshots of your tracked prompts across all four engines.
• Citation rate (% of runs where your domain is cited).
• Citation position (where in the source list).
• Share of voice vs competitors on the same prompts.
• Referral traffic in GA4 from `chatgpt.com`, `perplexity.ai`, `gemini.google.com` referrers.

Realistic timelines

GEO moves slower than SEO. New content that drops today won't show up in ChatGPT citations consistently for 4-8 weeks. Don't expect daily-rank-tracker-style movement; expect monthly-trend movement.

Common mistakes

• Optimizing for one model. Each LLM has slightly different preferences; optimize for the median.
• Stuffing keywords into pages hoping LLMs will pick them up. They don't work that way – clarity and citability win.
• Ignoring traditional SEO. If you're not findable on Google, you're probably not findable to ChatGPT either.
• Not having an llms.txt because "no one knows if it works yet". It's free; ship it.