What is Generative Engine Optimization (GEO)?

More buyers now ask an AI assistant before they search, compare or call. Instead of ten links, they get a written answer that names a few businesses — and often cites the pages it relied on. GEO is how a business makes sure that answer can include it, describes it correctly, and points to its own pages.

How AI assistants decide what to say about a business

The short answer: they repeat what they can find and check. When an assistant answers a question such as “which bookkeepers near me handle payroll for a 20-person company?”, it typically works in one of two ways.

• With live search. The assistant searches the web, reads a handful of pages — your site, competitors' sites, directories, reviews, articles — and writes an answer from them, often citing the sources. ChatGPT search, Claude with web search, Gemini with Google Search, Perplexity and Google's AI Overviews all work this way for many questions.
• From what the model learned. For some questions the assistant answers from its training data, without searching. That knowledge changes only when the model is updated.

In both cases the assistant can only say what it could find. If no readable page states your prices, service area, what is included or who you are a good fit for, the assistant either takes those details from someone else's page or leaves you out. Different assistants also use different sources and settings, so the same question can get different answers on each — which is why GEO measures each one separately.

GEO and SEO: what's the difference?

GEO builds on search engine optimization (SEO) but measures a different result. SEO asks where your link ranks; GEO asks whether an AI-written answer names you, cites your page, describes you accurately and recommends you for the buyer's needs.

Google says the same search fundamentals apply to its AI features, and that no special markup or file is required for them ^[1]. In practice, good SEO is a prerequisite for GEO, and GEO adds the checks SEO does not make: what the answers actually say.

What GEO measures

A useful GEO measurement asks real buyer questions on each AI model and records four outcomes for every business in every answer:

1. Named — is the business mentioned at all?
2. Cited — is one of the business's own pages among the sources?
3. Presented favourably — is it offered as an option, positively or neutrally, rather than as a warning?
4. Recommended for complex needs — when the buyer states several requirements (budget, size, location, timing), is it recommended as a fit?

Accuracy matters as much as visibility: a positive answer that states the wrong price or invents a service is not a success. Results should always show how many answers were counted and how many could not be measured. How Echo Tuner measures.

What helps: the GEO checklist

There is no published formula that guarantees an AI recommendation. These are the practices that are supported by the AI providers' own documentation or by how these systems demonstrably work. We apply all of them to this website.

1. Let AI search crawlers read your pages

Check your robots.txt and firewall. AI providers use separate crawlers for search and for training: OpenAI's OAI-SearchBot powers ChatGPT search, while GPTBot collects training data ^[2]; Anthropic's Claude-SearchBot and Claude-User are separate from ClaudeBot ^[3]. Blocking the search crawlers can keep you out of answers; whether to allow training crawlers is a separate business decision.

2. State the facts buyers decide on, in plain words

For each product or service, say how it is priced, what is included and what costs extra, who it is for (and not for), where you work, how to book or buy, how long it takes, and relevant qualifications. Put each fact on the page about that offering, in text — not only inside images or PDFs.

3. Give each offering its own clear page

One page that lists every service gives an assistant little to cite for any of them. A dedicated page per offering, with a descriptive heading, is easier to find and to quote.

4. Keep your business identity consistent

Use the same business name, address, phone, hours and categories on your website and every listing. Inconsistent details make it harder to tell that sources describe the same business.

5. Earn independent corroboration

AI answers draw on directories, reviews, articles and other sites. Accurate listings, genuine customer reviews and coverage by relevant independent sources give the assistant something to cite besides your own claims. For local businesses, Google weighs relevance, distance and popularity, including reviews ^[5].

6. Add structured data that matches the page

schema.org markup for your organization, services, products and FAQs helps machines read facts precisely. It must describe what the page visibly says; testimonials you choose and host yourself are not independent reviews ^[6].

7. Write answer-first, and date your pages

Answer the question in the first sentence, then explain. Show who published the page and when it was last updated. A clear, attributable, current statement is the easiest kind to quote.

What doesn't help

• Promises of a guaranteed AI ranking. No one outside the AI providers controls their answers.
• An llms.txt file as a shortcut. It is a harmless summary file, but Google says its search systems do not use it ^[1]; it cannot replace clear pages.
• Keyword stuffing or rewriting every heading as a question. Write for the buyer; formatting tricks are not a substitute for missing facts.
• Mass-produced pages. Generic pages created to target every possible question add little an assistant can rely on.
• Fake reviews, bought mentions and invented claims. They mislead buyers, break platform rules and can be repeated back inaccurately.

The research that named the field found that some content changes increased a source's visibility inside generated answers in experiments ^[4]. That is a reason to test changes and measure them — not a guaranteed uplift for any particular business.

Who needs GEO?

Any business whose customers might ask an AI assistant for a recommendation before buying. For small businesses, the biggest gains usually come from stating basic facts that are missing today. For larger and multi-location companies, the work is about measuring each offering and market against named competitors and keeping many pages and listings accurate.

How to get started

1. Write down ten questions your best customers ask before they buy.
2. Ask them in two or three AI assistants and note who is named and which pages are cited.
3. Check whether your own pages answer each question plainly.
4. Fix the missing facts first, then measure again.

Or let us do it properly, on several AI models at once, with your competitors alongside you: see the GEO audit.

Sources

1. Google Search Central — guidance on optimizing for Google's AI search features
   Google's guidance for AI Overviews and AI Mode: the usual search fundamentals apply; no special AI markup or llms.txt is required.
2. OpenAI — Overview of OpenAI crawlers
   OAI-SearchBot (search results in ChatGPT) is controlled separately from GPTBot (model training).
3. Anthropic Privacy Center — Does Anthropic crawl data from the web, and how can site owners block the crawler?
   Claude-SearchBot and Claude-User are distinct from ClaudeBot, which collects training data.
4. Aggarwal et al., “GEO: Generative Engine Optimization”, KDD 2024
   The research paper that named the field; it measured visibility of sources inside generated answers in an experimental setting.
5. Google Business Profile Help — Tips to improve your local ranking on Google
   Relevance, distance and popularity — including reviews — drive local results.
6. Google Search Central — Review snippet structured data documentation
   Reviews a business selects and hosts about itself are treated differently from independent reviews.

All the sources behind our GEO practices