GEO guide: where to start to become visible in AI search (2026)

GEO (Generative Engine Optimization) is the work of getting a brand into the answers given by generative AI engines such as ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. This guide briefly defines the concept, explains with sources how an AI answer is built, and lists six steps a brand can take today. The basis for every figure in the guide is in the Sources section at the end of the post.

What is GEO, in brief?

GEO is the branch of search engine optimisation (SEO) that extends into AI answers. In Turkish sources the same field also goes by the names AEO (Answer Engine Optimization), LLM SEO and AI visibility. They all ask the same question: when a user asks an AI something, whose name appears in the answer, and which sources is the answer built from?

For a short definition of the concept and how it differs from SEO, see our What is GEO? page. This guide goes one step beyond the definition and focuses on the question "where should I start?"

The unit of measurement is the most concrete difference between the two disciplines. In SEO the unit is the rank for a keyword. In GEO the unit is individual answers: a question was asked to an engine, and the brand was either mentioned in the answer or not. This binary outcome can be counted; that is why success in GEO is measured far more directly than a ranking estimate.

How is an AI answer built?

Most of today's AI assistants do not answer a question from training data alone; they search the web, build the answer from the pages they find, and show those pages as sources. The engines' own documentation says so plainly:

• ChatGPT: According to OpenAI's help page, ChatGPT can search the web for questions that need up-to-date information, and answers that use web search can include source links. The same page also notes that the user's question is turned into one or more search queries and sent to search partners.
• Perplexity: Perplexity's help center says the tool searches the internet in real time and that every answer includes numbered citations to the original sources.
• Google AI Overviews and AI Mode: Google explains that these features can split a question into subtopics and run several related searches ("query fan-out"). Google Turkey announced on 18 February 2026 that AI Mode and AI Overviews were rolling out gradually in Turkey.

The consequence of this mechanism is simple: a page the engine cannot find or cannot read cannot be a source for the answer. Google says this in its own words: to appear in AI Overviews or AI Mode, a page needs to be indexed and eligible to be shown with a snippet in Google Search; beyond that there is no additional technical requirement or special optimisation.

Why now?

Search behaviour is shifting towards a pattern where the answer is given directly on the page. According to SparkToro's analysis using Similarweb data, 68.01% of Google searches in the US between January and April 2026 ended without sending a click to any site.

When clicks decline, a brand whose name is not in the answer gets no visit at all. That is why GEO is on the agenda today: if the user has got the answer, the race is over inside the answer.

What is the difference between SEO and GEO?

GEO does not replace SEO; it builds on it. Google says no additional requirement or special optimisation is needed for AI Overviews and AI Mode, which means the basic rules that apply to Search apply here too. The differences lie in the goal and in measurement:

The last row may look minor, but it changes measurement from the start. In SparkToro's January 2026 research, 600 volunteers asked 12 questions to ChatGPT, Claude and Google AI a total of 2,961 times. The chance of the same brand list coming back in two answers was under one percent on ChatGPT and on Google's AI. How often a brand appeared in the lists, on the other hand, was much more consistent than its ranking. That is why a single screenshot is not a measurement.

Where to start: six steps

The steps below follow an order: first measure where you stand, then make sure the engines can read you, and last work on content and external sources. None of them guarantees a particular outcome; OpenAI also states plainly on its own help page that appearing in search results is not guaranteed.

1. Measure where you stand today

The starting point is to write the questions your customers would ask without including your brand name, and to ask these questions to each engine separately. A question that contains the brand name ("Is brand X trustworthy?") already guarantees your name will appear in the answer, so it does not count as evidence of visibility. We explained how to measure step by step in a separate guide; the details of our method are on the How we measure page.

2. Make sure AI crawlers can reach your site

AI companies use different crawlers for different jobs, and you can manage each one separately in your robots.txt file. An easy mistake is to block the search crawler as well while trying to block model training.

The controls Google offers for AI Overviews and AI Mode are Search's own controls: nosnippet, data-nosnippet, max-snippet and noindex. So blocking Google-Extended does not remove you from Google's AI features within search; Google's documentation says these features are part of Search and are managed through snippet settings.

Next step

Is your site open to AI crawlers?

Enter your domain. The free scan measures whether AI crawlers can reach your site and whether your pages are quotable. No signup, no card, no email.

Your domainScan for free

3. Do not bury important information in JavaScript

According to an analysis Vercel published in December 2024 based on traffic on its own network, none of the major AI crawlers execute JavaScript, including OpenAI's crawlers, Anthropic's ClaudeBot and PerplexityBot. ChatGPT and Claude crawlers download JavaScript files but do not execute them. The same analysis notes that Gemini uses Googlebot's infrastructure and can render JavaScript.

The practical consequence: if your prices, product features or contact details are loaded onto the page only by code that runs in the browser, these crawlers may never see that information. Critical information should be in the first HTML the page receives from the server.

4. Write to be quoted

One of the field's foundational academic works is the GEO paper published at KDD 2024. On its own benchmark set, the study reports that adding quotations, statistics and sources to sources already pulled into the answer increased that source's visibility in the answer. It measures that keyword stuffing, by contrast, brought almost no improvement.

Independently of the paper, there are a few well-known rules for writing to be quoted:

• Give the answer in the first sentence. An engine may quote a single paragraph from your page. That paragraph should be correct and complete when read on its own.
• Open the paragraph with its subject. A paragraph that starts with "That is why" or "For this" loses what it refers to when quoted on its own.
• Write a figure with its unit and source. Not "68 percent" but something like "68.01% of searches in the US between January and April 2026".
• Match structured data to visible content. Google's structured data guidelines ask that content not visible on the page should not be marked up, and that the markup accurately reflect the page content.

We prepared our checklist to gather these rules on one page.

5. Do not forget what is outside your site

Most of the pages the engines read about you are not on your site. When we combined the 60 analyses we ran up to 3 August 2026, 93.6% of the 75,914 source links the engines showed were outside the brand's own site (this is a set made up of our own client and comparison brands, not a sample of the market). In Ahrefs' analysis of 75 thousand brands, the strongest relationship with a brand being mentioned in Google AI Overviews was shown by the number of pages on the web where the brand's name appears (0.664). The relationship with the number of backlinks was much weaker (0.218).

This is a relationship, not cause and effect; the study itself stresses this. Still, the direction is clear: what is written about you outside your site is the raw material of the answer. We wrote in detail about where this material comes from in our own measurements in our data post on whom AI reads about Turkish brands.

6. Measure again with the same questions

The only way to know whether a change worked is to ask the same questions to the same engines again. If you change the questions every time, the difference you see may come from the changed questions rather than from your work. The inconsistency mentioned above leads to the same conclusion: not a single measurement but repeated measurements and a range should be read.

What to avoid

In a new field, the most expensive mistakes are shortcuts with no evidence behind them. Three come up often:

• Looking for a special "AI file". Google states plainly that no new machine-readable files, AI-specific text files or additional markup are needed to appear in AI Overviews or AI Mode.
• Marking up invisible content. Marking up information that is not visible to readers on the page with structured data goes against Google's guidelines and can lead to a manual action.
• Relying on a single result. Appearing or not appearing once on an engine does not show a trend, because of the inconsistency above. Decisions should be made on repeated measurement.

Frequently asked questions

Does GEO make SEO unnecessary?

No. Google says there is no additional requirement for AI Overviews and AI Mode, and that it is enough for a page to be indexed and eligible to be shown with a snippet. This means a good SEO foundation is a prerequisite for GEO.

Which engine should I start with?

We recommend not settling for a single engine: engines read different sources and can mention different brands for the same question. In our measurements we report ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews separately; we do not substitute one for another.

When will results show?

No one can honestly say this in advance. Which sources the engines read and how often is not visible from the outside, and answers change even for the same question. What we can say is this: the next measurement with the same questions will show the change (an increase, a decrease or no change).