What Is GEO? How to Get Cited by ChatGPT, Perplexity and Google AI Overviews

TL;DR: GEO (generative engine optimization) is the work of getting your pages found and quoted by AI answer engines such as ChatGPT search, Perplexity and Google's AI Overviews. Google says it is still SEO, and for Google's surfaces that holds: a page has to be indexed and eligible for a snippet before any AI feature can use it. The GEO-specific parts are narrow. Let the AI search crawlers in, put your content in the HTML rather than behind JavaScript, and write answers that still make sense when they are quoted alone. Start by running your site through the free AI visibility check.

Every few years search marketing gets a new acronym and a new round of people selling it. GEO is this cycle's. The good news is that underneath the noise there is a short list of checkable things, and most of them are the same things that made a site rank in the first place.

What is GEO?

Generative engine optimization (GEO) is the practice of making a website easy for AI answer engines to find and cite. When someone asks ChatGPT, Perplexity, Microsoft Copilot or Google's AI Mode a question, the system searches the web, reads a handful of pages and writes an answer with links back to its sources. GEO is about being one of those sources. The term comes from a 2023 research paper, GEO: Generative Engine Optimization, which tested how rewriting content changed its visibility inside generated answers. In practice most of the work overlaps with classic SEO: the page must be crawlable, indexed, fast and clearly written. GEO adds attention to three things AI systems depend on more than a results page does. Their crawler has to be allowed in, the text has to be present in the HTML without running JavaScript, and each passage has to stand on its own when it is lifted out and quoted.

Is GEO different from SEO?

For Google, barely. Google's page on AI features and your website says the same SEO best practices apply to AI Overviews and AI Mode, and that there are no additional requirements beyond being indexed and eligible to show with a snippet. There is no separate AI index to submit to and no special markup that unlocks a citation. If your page does not rank for anything today, an AI feature built on the same index is unlikely to discover it tomorrow.

Outside Google the picture changes a little, because each assistant runs its own crawler. OpenAI's crawler documentation says sites that block OAI-SearchBot will not be shown in ChatGPT search answers. Perplexity runs PerplexityBot to build its own index, as described in its crawler documentation. So GEO is SEO plus a few extra doors that each need to be unlocked, and a few habits in how you write.

How do AI assistants choose which pages to cite?

None of the companies publish their ranking, but the general mechanism is documented well enough to act on. An assistant receives a question, runs one or more searches against an index, fetches the most promising pages, and picks passages that answer the question. Google describes one part of this as "query fan-out": the system issues several related searches across subtopics and then assembles an answer from what comes back. That means one question can pull a page in through a sub-question you never targeted directly.

Two practical consequences follow. First, the retrieval step is still search, so everything that decides whether a page is found and trusted in normal search still applies. Second, the final choice is made at the passage level. A long page with one great paragraph buried in the middle competes with a short page whose first paragraph answers the question outright, and the short page often wins because its answer is easier to extract without losing meaning.

What does a citable passage look like?

A citable passage answers one question in its first sentence, names its subject instead of saying "it", and carries the specific fact a reader came for. Here is an example of the kind we see all the time, on a made-up dental clinic's FAQ. The answer to "How much does a crown cost?" opened with two sentences about how every mouth is different. The rewrite leads with the price range instead: a porcelain crown at the clinic costs between 3,500 and 4,500 shekels, depending on the tooth and whether it needs a root canal first. That one sentence can be quoted by an assistant as-is, and it still makes sense without the page around it.

There is no official length for a passage. We aim for roughly 130 to 170 words under each question-style heading, because that is enough room for a full answer with its main caveat and short enough to be lifted whole. Put the answer first and the edge cases last. If a paragraph only makes sense after reading the one above it, an extractor will either skip it or quote it wrong.

Some habits that help:

• Write headings as the question a person would type or say: "How much does a website audit cost?" rather than "Pricing".
• Put numbers, prices, dates, names and places in the text. An assistant cannot quote a price that only exists inside an image.
• Keep one idea per paragraph. Tables work well for comparisons, because each row stands on its own.
• Say who wrote it and when it was last checked. A byline and an honest "updated" date are part of how people and machines judge a source.

Which AI crawlers need access to your site?

This is where most GEO problems actually live. Sites block AI crawlers by accident all the time, usually through a CDN bot setting someone switched on or a robots.txt rule copied from a 2023 blog post about protecting content from training. The table lists the crawlers that matter for being cited, with the companies' own descriptions.

The training crawlers are a separate decision from the search ones. OpenAI states that its settings are independent, so you can block GPTBot and still allow OAI-SearchBot. Google says Google-Extended does not affect inclusion or ranking in Google Search. Anthropic describes its three bots in its crawler help article. We go through the rules line by line in llms.txt and robots.txt for AI crawlers.

Does your content need to render without JavaScript?

For AI crawlers, assume yes. Googlebot renders JavaScript, although rendering can happen later than the first crawl, as Google explains in its JavaScript SEO basics. The AI companies' crawlers are less documented on this point, and the safe reading is that many of them take the HTML they are given and move on. If your product descriptions, prices or FAQ answers are injected by a script after the page loads, some assistants will see an empty template.

Testing takes a minute. Open the page, view the source (not the inspector, the raw source) and search for a sentence from your main content. If it is there, you are fine. If the source is mostly script tags and an empty <div id="root">, the fix is server-side rendering or static generation, which every modern framework supports.

Does structured data help you get cited?

It helps machines be sure what a page is about, which is not the same as a guaranteed citation. Google's guidance is plain: structured data should describe content that is visible on the page, and there is no special schema for AI features. Where schema earns its place is in removing ambiguity. An Organization node that names your business and links its profiles, a Person node for the author, and an Article or Product node that repeats the facts on the page all make it easier to connect your page to the entity a question is about.

Nobody outside the AI companies can show that adding a given schema type raises citation rates. We add it because it is cheap and it has made the facts on every site we have audited easier to extract, not because it is magic.

What about llms.txt?

llms.txt is a proposal from 2024, documented at llmstxt.org, for a Markdown file at the root of a site that summarises it for language models. It is a reasonable idea. So far no major AI search provider has said it reads the file when choosing sources, and Google's AI features guidance says no special AI files are needed. Publishing one takes minutes with the free AI intro file tool, so it is fine to have. Just do not expect it to change what you are cited for.

How do you measure GEO?

There is no Search Console for ChatGPT yet, so measurement is part manual and part analytics. Once a month, ask the main assistants the ten questions your customers ask most and write down whether you are cited and who is cited instead. It is tedious, and it is also the only direct measure there is.

In analytics, segment referral traffic from chatgpt.com, perplexity.ai, copilot.microsoft.com and gemini.google.com. Links clicked in ChatGPT often arrive tagged utm_source=chatgpt.com, which makes them easy to filter. For Google, clicks from AI Overviews and AI Mode are counted inside the normal Search Console performance report, according to Google's AI features page, so they do not show up as a separate line.

A GEO checklist you can run this week

1. Run your site through the free AI visibility check and allow the search crawlers you want to be cited by.
2. Check your CDN or firewall bot settings too, because a block there overrides a friendly robots.txt.
3. View the raw source of your five most important pages and confirm the main text is in it.
4. Confirm those pages are indexed in Google Search Console and Bing Webmaster Tools.
5. Rewrite the first paragraph under each main heading so it answers the heading in the first sentence.
6. Move prices, opening hours, service areas and other facts out of images and into text.
7. Add Organization, Person and Article or LocalBusiness structured data that matches what the page shows.
8. Add a byline and a real "last updated" date to pages that give advice.

If you would rather have all of this checked across the whole site, with a score per category and the fixes in order, that is what our full report does. For the do-it-yourself version, the free SEO + GEO guide has every check with a free tool next to it.