What Is GEO? The New Rule for Staying Visible in the Age of ChatGPT and Gemini

In short: GEO (Generative Engine Optimization) is the work of getting your brand cited as a source when ChatGPT, Gemini, Perplexity, Claude or Google AI Overviews answer a question. It doesn't replace SEO; it builds on it, because AI answer engines also pick content that is crawlable, answers the question clearly and rests on trustworthy sources. In practice GEO comes down to four jobs: giving AI crawlers access, answering the question in the first paragraph of every page, backing up claims with sources, and describing your brand with the same facts everywhere.

When a user asks a question, they no longer get ten blue links; more often than not, they get a single answer straight away. That answer is written by ChatGPT, Gemini, Perplexity or Google's AI-powered search experience. The problem is this: is your brand inside that answer?

What is GEO and where did it come from?

The term caught on with the paper "GEO: Generative Engine Optimization" (Aggarwal et al.), published in 2023 by researchers from Princeton and IIT Delhi and accepted to the KDD 2024 conference. The researchers measured, on a multi-domain benchmark, how small changes to content affect a page's visibility in generative answers. Three findings matter for us:

• The methods that worked best were adding citations to reliable sources, adding quotations and using statistics instead of qualitative description; the paper reports visibility gains of up to 40%.
• Keyword stuffing, the classic SEO reflex, brought little to no improvement in generative answers.
• The effect varied by domain, and sites ranked lower in search results gained more from these changes.

These are results from a controlled experiment, not a guaranteed effect in any particular AI product. But the direction is clear: AI prefers content that carries evidence and is easy to read.

How do AI answer engines choose their sources?

There are two ways an assistant can mention you. The first is the model's training data: the model "knows" you from the text it was trained on. That knowledge only updates with new model versions, and it isn't something you can influence directly. The second is live search: before answering, the assistant searches the web, reads a few pages, bases its answer on them and cites them. That second path is where GEO really plays.

On Google's side the picture is clear. According to Google's guide to AI features and your website, there are no additional requirements or special optimizations needed to appear in AI Overviews or AI Mode. The page has to be indexed and eligible to be shown with a snippet. Google also says these features may use a "query fan-out" technique that splits a question into several related searches. That raises the odds that pages answering the side questions of a topic get picked as sources.

The main things answer engines look for when choosing a source:

1. A direct answer: text that answers the question plainly, in the very first paragraph.
2. Semantic coherence: a structure that breaks the topic into headings and ties each section to a single sub-question.
3. Evidence: sourced data, dates and concrete examples.
4. Trust signals: an author, a last-updated date and internally consistent information. AI steers clear of contradictory sources.

The difference between GEO and SEO

SEO aims to move you up the search results page; GEO aims to move you into the answer itself. They aren't rivals; think of them as two floors standing on the same foundation.

You may also come across AEO (Answer Engine Optimization); it's where the two meet: content that gives a clear, direct answer.

GEO step by step: a checklist for your business

1. Open the door to AI crawlers

An assistant that can't read your page can't cite it. Each company runs more than one crawler, some for model training and some for search. Knowing the difference lets you make the "should we block AI?" decision deliberately:

These details come from the companies' own documentation: OpenAI's crawlers, Anthropic's crawlers, Perplexity's crawlers and Google's list of crawlers. Agents such as ChatGPT-User, Claude-User and Perplexity-User open a page on the spot when a user asks something. OpenAI and Perplexity note that because these visits are user-initiated, robots.txt rules may not apply to them.

If you want to be visible, don't block the search bots. If you don't want your content used for training but do want to stay in search, you can close off just the training bots:

User-agent: GPTBot
Disallow: /

User-agent: ClaudeBot
Disallow: /

User-agent: *
Allow: /

Checking robots.txt isn't enough. Google also reminds site owners that crawling has to be allowed not only in robots.txt but also by their CDN and hosting infrastructure. Your firewall's "bot protection" setting may be quietly shutting the door robots.txt left open. On our own site, we leave all of these crawlers open.

2. Answer one question in the first paragraph of every page

Answer engines usually quote a paragraph, not a whole page. Put a 2–4 sentence summary at the top of the page that makes sense on its own. Write headings the way users actually phrase their questions, and let each section answer a single sub-question. Put important information in the text, not inside an image or a PDF.

3. Back up your claims with sources

The strongest methods in the GEO paper were adding sources, quotations and data. Google's guide to helpful content also asks whether content provides original information and clear sourcing. Instead of "we're the industry leader", write how many projects you've delivered, how long you've been working and what you measure and how. If you use outside data, link to where it came from.

4. Set up structured data correctly

Schemas such as Organization, Article and BreadcrumbList help machines interpret the page correctly. Two caveats. First, Google states plainly that no special schema.org markup is needed for AI features; schema isn't a magic button. Second, the markup has to match the content visible on the page. Marking up content readers can't see goes against Google's structured data policies. Also, since May 2026 Google no longer shows FAQ rich results in Search, so don't treat FAQ schema as a shortcut to visibility.

5. Describe your brand with the same facts everywhere

AI doesn't learn about a brand from a single page; it pieces it together from many sources. Your company name, address, phone number, founding year and "what we do" sentence should be identical on your site, your Google Business Profile, LinkedIn and industry directories. Link your official profiles together with the sameAs field in your Organization schema. A model that sees contradictions either won't mention you or will get you wrong.

6. Add llms.txt, but don't expect miracles

llms.txt is a file proposed by Jeremy Howard in September 2024. It sits at the site root and gives language models a Markdown summary of your site and its key pages. The only required section is a heading with the site's name. Setting it up takes about an hour, and we use one on our own site. But it's a proposal, not an official standard, and Google's guide to AI features doesn't list it as a requirement. Add it alongside the five steps above, not instead of them.

How do you measure GEO?

Measurement is GEO's weakest link. These are the tools available today:

• Google Search Console: data for your pages that appear in AI Overviews and AI Mode is included in the Performance report under the "Web" search type. There's no separate filter, but you can follow the trend in impressions and clicks.
• Bing Webmaster Tools: the AI Performance report, launched in public preview in February 2026, shows how often your pages are cited as sources in Microsoft Copilot and Bing's AI-generated summaries, and which queries surfaced them. It doesn't include click data.
• Analytics: OpenAI says ChatGPT automatically adds the utm_source=chatgpt.com parameter to links from ChatGPT search. You can separate visits from other assistants by referring domain.
• Regular question tests: pick 20–30 questions your customers would ask and put them to ChatGPT, Gemini, Perplexity and Claude every month. Record whether your brand is mentioned, which of your pages is cited and where competitors pull ahead. Answers shift a little every time, so look at the trend over months rather than a single test.

Common mistakes

• Blocking every AI bot wholesale. Closing a training bot is a choice; closing a search bot means giving up visibility.
• Mass-producing separate pages "for AI". Generating many pages to manipulate rankings is listed in Google's spam policies as "scaled content abuse".
• Schema that doesn't match the page. Marking up a rating or an FAQ that isn't on the page doesn't build trust; it erodes it.
• Content with no date and no author. An answer engine trusts text less when it can't see how current it is or who stands behind it.
• Treating GEO as a one-off project. Assistants, competitors and questions keep changing; measurement has to be repeated every month.

To check the technical foundation, see our technical SEO checklist, and for why page speed still matters, our article on Core Web Vitals. If an agency has sent you a GEO proposal, our article on SEO service pricing explains what to look for in it.

Search has changed, and so has the rule for staying visible. At Kodakod, we build your site to be cited as a source in both classic search and AI answers, combining technical SEO and GEO into a single strategy. You'll find the scope on our SEO and GEO services page.