Learn GEO: get your brand into AI answers

GEO (generative engine optimization) is the practice of making AI systems cite your content and recommend your brand when they answer questions. Where SEO competes for a position on a results page, GEO competes for a place inside the answer itself.

You will see the same discipline under several names: AEO (answer engine optimization), LLM SEO, AI SEO, or AI visibility. The labels differ; the work is the same. Throughout this guide we say GEO, and we mean: everything that raises the odds an AI assistant quotes your pages, links to them, or names your brand when a buyer asks for options.

One framing to keep from the start: GEO is not a replacement for SEO. AI systems discover, evaluate, and quote web content using infrastructure built on search. GEO is a new scoreboard on top of largely familiar work, plus a handful of genuinely new moves covered in Chapters 6 and 7. If you have not read our guide to SEO, it is the natural prerequisite to this one.

The shift is not hypothetical. In our own signup surveys, ChatGPT went from zero to 16% of new-customer discovery in six quarters, second only to Google search. Ask around and you will hear the same from other software companies: a fast-growing slice of buyers now starts with "ask the assistant" instead of "search and click".

Three properties make this channel worth deliberate effort:

• Winner-take-most answers. A results page has ten organic slots. An AI answer typically cites a handful of sources and names two or three brands. Being in or out of that set is a step function, not a gradient.
• High-intent, pre-trusted traffic. Visitors arriving from an AI recommendation have already been told you are a fit. They convert at rates paid channels rarely touch.
• An early-mover window. Most companies are not doing this work yet. The sources assistants learn to trust now will be the incumbents of this channel later.

Meanwhile Google is folding the same behavior into search itself: AI Overviews answer queries directly at the top of the results page, drawing from the same kinds of citable sources. GEO and SEO increasingly share one battlefield.

To influence AI answers you need to know where they come from. Every assistant draws on two distinct pipelines:

• Training data (the model's memory). Models memorize the web they were trained on. When an assistant answers from memory, it reproduces what the corpus said about your category months or years ago. You influence this slowly, by being consistently and factually described across the web: your own site, review platforms, comparison articles, communities like Reddit, and press.
• Live retrieval (RAG). For current or specific questions, assistants run searches behind the scenes, fetch the top pages, and synthesize an answer with citations. ChatGPT search, Perplexity, and AI Overviews all work this way. The pages they fetch come from search indexes, which is why search visibility still gates AI visibility.

The strategic consequence: retrieval is where you can win this quarter. If your page ranks for a query and answers it cleanly, it can be cited within days of publishing. Training-data presence compounds over quarters as the web's description of you solidifies. Work both, but expect results on very different clocks.

AI companies run separate crawlers for separate jobs, and treating them as one "AI bot" leads to expensive mistakes. The ones that matter:

• GPTBot (OpenAI) collects training data for future models. Blocking it keeps your content out of model memory but does not affect ChatGPT search results.
• OAI-SearchBot (OpenAI) indexes the web for ChatGPT search. Block this and you disappear from ChatGPT's cited, linked answers.
• ChatGPT-User (OpenAI) fetches pages on demand when a user asks ChatGPT to read a specific URL or browse.
• ClaudeBot and Claude-User (Anthropic) play the equivalent training and on-demand roles for Claude.
• PerplexityBot and Perplexity-User index and fetch for Perplexity, an assistant that cites sources on nearly every answer.
• Google-Extended (Google) is a robots.txt control for Gemini training only. It does not affect Googlebot, your rankings, or AI Overviews, which are powered by normal Google Search indexing.

The practical rule: never block retrieval and user-request bots (OAI-SearchBot, ChatGPT-User, Perplexity, Claude-User) unless you want out of AI answers entirely. Blocking training bots like GPTBot is a legitimate editorial choice, but for most businesses the visibility upside of being in model memory outweighs the cost. Audit your robots.txt today; many sites blocked "AI bots" wholesale in 2023 and are now invisible in a channel they want.

Because retrieval runs on search infrastructure, most SEO fundamentals carry straight over: crawlable pages, matching intent, topical authority, structured content, and earned links all still decide whether you are in the candidate set. What changes is what happens after retrieval:

• Passages beat pages. Engines rank pages; assistants quote passages. A page wins a citation when one of its sections answers the exact sub-question the model is assembling. Every section of a page is now its own contestant.
• Citations replace positions. There is no position three in an answer. You are quoted or you are not, which raises the premium on being the clearest, most quotable source rather than a marginally better-optimized one.
• Brand mentions become a ranking surface. When a buyer asks "what should I use for X", assistants weigh how consistently the web names and describes you. Reviews, comparison posts, and community threads now do work that meta tags used to do.
• Zero-click is the default. Many users never leave the answer. The consolation prize is influence: the assistant repeating your framing, your data, and your name to the buyer is marketing even without the visit.

Citation-worthy content is a craft with learnable rules. The ones with the strongest track record:

• Answer first, elaborate second. Open every page, and every H2 section, with a direct answer to the question in the heading. Models extract self-contained passages; a section that spends three paragraphs warming up loses to one that answers in its first sentence.
• Structure for chunking. Descriptive H2s and H3s, short paragraphs, bulleted lists, comparison tables. AI pipelines split pages into chunks; clean structure means each chunk survives as a coherent, quotable unit.
• Publish original data. Statistics, benchmarks, and first-hand research are citation magnets. A model assembling an answer needs numbers to attribute, and the site that generated the number is the natural citation. Our "zero to 16% from ChatGPT" stat in Chapter 2 is this tactic in use.
• Make your entity unambiguous. State plainly what your product is, who it is for, and what category it belongs to, in the same terms everywhere: your site, your schema markup, directories, and profiles. Models recommend brands they can describe without hedging.
• Add FAQs and keep content current. Question and answer pairs mirror how users prompt assistants, and retrieval favors fresh pages. Update your cornerstone pages on a schedule, not just at launch.
• Be present where models read. Comparison listicles, review platforms, and Reddit threads are heavily retrieved for "what should I use" queries. Earning honest coverage there is GEO work as much as anything on your own domain.

The technical layer of GEO is about making your content effortless for machines to fetch and parse:

• Serve real HTML. Most AI crawlers do not execute JavaScript. If your content only appears after a client-side framework renders it, assistants see an empty shell. Static, server-rendered pages are the safe architecture, and they are faster for humans too.
• Open the gates. Chapter 4's crawler audit: retrieval bots allowed in robots.txt, no bot-blocking firewall rules silently returning errors to them, fast response times so fetch budgets go further.
• Publish llms.txt. A markdown index at yoursite.com/llms.txt that lists your key pages with clean descriptions, built for AI readers the way sitemaps are built for search crawlers. It is an emerging convention with growing adoption, and it costs nothing to be early.
• Keep schema markup rich. JSON-LD for articles, FAQs, products, and your organization gives models typed facts instead of inferred ones: who wrote this, when it was updated, what the product costs. Typed facts get quoted with more confidence.
• Ping on publish. Protocols like IndexNow notify search indexes the moment content ships, which matters more when answers favor fresh sources.

GEO has no rank tracker yet, but it is far from unmeasurable. Build your scoreboard from three sources:

• Referral traffic. Visits from chatgpt.com, perplexity.ai, gemini.google.com, and claude.ai show up in any analytics tool. Track the trend monthly. The absolute numbers will start small; the growth rate is the signal.
• Assistant spot-checks. Keep a list of the 10 to 20 questions a buyer in your category would ask, and run them through the major assistants monthly. Log whether you are cited, whether you are recommended, and what is said about you. This is the closest thing GEO has to rank tracking.
• Source-of-discovery surveys. Ask new customers where they first heard of you, with AI assistants as an explicit option. Referral data undercounts this channel badly because many users read the answer and search your brand later. Our own 16% figure comes from surveys, not referrers.

Then hold the same bar you hold for SEO: traffic and mentions are means, signups and revenue are the end. Attribute what you can, survey the rest, and give the channel quarters, not weeks, to prove itself. The compounding works the same way it always has; only the scoreboard is new.