What is AI SEO? How ChatGPT, Gemini and AI Overviews pick sources

AI SEO usually means optimizing a website so AI search, such as Google's AI Overviews and AI Mode, ChatGPT search, Perplexity and Gemini, finds its pages and cites them in answers. The same name also gets used for doing SEO work with AI tools, which is a separate job with its own risk. The first kind is mostly ordinary SEO, plus letting AI crawlers in, putting your content in the HTML and writing passages a model can quote. The second is fine by Google as long as you don't mass-produce pages that add nothing, which its spam policies call scaled content abuse.

When someone pitches you AI SEO, ask which one they mean. The deliverables have almost nothing in common.

What does AI SEO mean?

AI SEO has two meanings, and they share little beyond the acronym.

The first meaning is the same job as generative engine optimization. Our guide to generative engine optimization covers the research behind it, and AEO vs GEO vs SEO sorts out the rest of the acronyms.

How AI search engines pick the sites they cite

AI search engines run a search before they write, then cite the pages that back up the answer.

1. The engine turns the question into one or more searches. Google's guide to AI features calls this query fan-out. OpenAI's ChatGPT search help page says ChatGPT rewrites a message into narrower queries for its search providers.
2. Each search pulls results from an index. Google's guide to optimizing for generative AI says AI Overviews and AI Mode use its core ranking systems to fetch pages from the Search index. The Gemini app grounds its answers in the same index, according to Google's crawler documentation.
3. The model reads passages from the results and writes one answer.
4. It links the pages whose passages support what it wrote.

That gives you two gates. Retrieval decides whether your page is among the candidates, and that is classic SEO plus crawler access. Citation decides whether your passage is the one the model uses, which comes down to how plainly it answers a sub-question. A page that ranks sixth for the head term can still win the citation because it states the price or the limit a fan-out search went looking for.

Past the first gate, nobody outside Google knows how sources are ranked inside an answer. Google's guide says no third-party tool has access to its ranking or AI systems, so be wary of anyone who claims to know the AI Overviews algorithm.

Is AI SEO different from SEO?

For Google, barely. Its guide says optimizing for generative AI search is still SEO, and the AI features page lists no extra requirements to appear in AI Overviews or AI Mode beyond being indexed and eligible to show a snippet.

Google also lists what you can skip. You don't need AI text files, Markdown copies of your pages or special schema.org markup, and you don't need to cut content into tiny chunks. An llms.txt file does no harm, but Google Search ignores it.

Outside Google, more changes. ChatGPT search and Perplexity rely on their own crawlers, OAI-SearchBot and PerplexityBot, so being open to Googlebot is not enough. OpenAI documents its training crawler, GPTBot, separately, so you can block training and still appear in search. Our post on LLM SEO covers that split.

Most AI crawlers also skip JavaScript. A December 2024 study by Vercel and MERJ found that none of the major AI crawlers rendered it. The exceptions were Gemini, which runs on Googlebot's infrastructure, and Applebot. A pricing table that loads client-side shows up for Google and comes back blank to ChatGPT's crawler.

The last change is whose pages get read. Ask for the best tool in a category and the engine retrieves comparison articles, review sites and forum threads along with vendor pages, so what they say about you ends up in the answer.

What you control in AI search

You control whether engines can fetch and use your pages, and what those pages say. You don't control which source a model picks or how it words the answer.

Start with robots.txt, because a mistake there blocks everything after it. This pattern is common and looks harmless:

User-agent: Googlebot
Allow: /

User-agent: *
Disallow: /

It lets Google in and turns every other crawler away, OAI-SearchBot included. OpenAI's crawler documentation says a site that blocks OAI-SearchBot won't appear in ChatGPT search answers, apart from plain navigational links. Perplexity's crawler page says PerplexityBot exists to surface and link sites in its results, not to train models. To appear in both, give each bot its own group with Allow: /. Then check your CDN or firewall, because a bot rule there can return a 403 to a crawler that robots.txt invites in.

Since August 31, 2026, Search Console also has a sitewide Search generative AI setting under Settings. It defaults to include. Exclude keeps the site out of AI Overviews, AI Mode and Discover's generative AI features, and Google says it does not affect ranking or training. Leave it on include, but check it once. An exclude that someone set and forgot removes the site from Google's AI answers while rankings look normal.

The page-level controls nosnippet, data-nosnippet, max-snippet and noindex limit what AI features show too, and they cut your regular snippet as well. To keep one block out of AI answers, put data-nosnippet on that block rather than nosnippet on the whole page.

How to start with AI SEO on your own site

Fix access first, then content, then measure.

1. Confirm the pages you care about return 200 and are indexed in Google.
2. Allow Googlebot, OAI-SearchBot and PerplexityBot in robots.txt and your CDN.
3. Put key facts like prices, plan limits and specs in the HTML your server sends, not in content that JavaScript loads.
4. Open each section with a direct answer that names your product and states one specific fact.
5. Watch Search Console's Generative AI performance report, open to all sites since August 31, 2026. It counts impressions in AI Overviews and AI Mode combined, with no clicks.

Our step-by-step AI search engine optimization playbook walks through each step in detail.

Using AI tools for SEO: what Google allows

Using AI to do SEO work is allowed, and Google judges the result, not the tool. Its February 8, 2023 post on AI-generated content said so directly. "Appropriate use of AI or automation is not against our guidelines."

Google's guidance on generative AI content says AI helps most with researching a topic and adding structure to original content. It asks you to fact-check everything a model writes before publishing, including titles, meta descriptions, structured data and alt text, since those appear in search results too.

The line is volume without value. Google announced its scaled content abuse policy on March 5, 2024. It covers many pages made mainly to manipulate rankings, and its examples include using generative AI to produce many pages that add nothing for users. Google said the policy applies whether automation, people or both made the pages.

Use AI for the tedious middle of SEO work. Group a 2,000-row keyword export into topics, draft JSON-LD you then validate, pull bot hits out of a server log. It gets sites in trouble when it's used as a page factory, such as 300 city pages with the town name swapped. Those pages make poor AI search sources anyway, because a model has no reason to cite a passage that says what every other page says.

Check whether AI search can read your site

Our website check tests the access and content gates above on your own pages. The home page preview costs no credits and takes about 20 seconds. It covers crawling and indexing, on-page SEO and structured data, including whether robots.txt lets in GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended and other AI crawlers, and scores SEO readiness and AI access from 0 to 100.

The full check runs on up to three pages and adds content and answer structure, rendering, which flags content that only loads with JavaScript, performance, images, a site structure crawl of up to 150 pages and brand identity. Its price is the sum of the checks you pick and appears before it runs. The four default checks on one page come to 67 credits, and each question you add to sample ChatGPT, Perplexity and Gemini answers costs 120. Rerun it after you publish fixes and it reports the percent of issues fixed.