LLM SEO: How ChatGPT, Perplexity, and Google's AI Pick Their Sources

LLM SEO means making your pages easy for AI assistants to find, trust, and cite. LLM stands for large language model, the kind of AI behind ChatGPT, Gemini, and Perplexity. Many people call the same work GEO, short for generative engine optimization.

Most of it is ordinary SEO done with care. Google says so directly: optimizing for its AI search features “is still SEO.” The difference is the goal. Besides ranking in a list of links, you want to be named or linked inside the AI’s answer.

This page covers what is known about how AI tools pick sources, what we saw in our own tests, and what to do about it. Where the evidence is thin, we say so.

What does LLM SEO mean?

Classic SEO tries to rank a page in a list of results. LLM SEO tries to get that page used as a source when an AI writes an answer.

The term GEO comes from a 2023 research paper by researchers from Princeton and other groups. They built a test set of questions and measured which page edits got content used more often in AI answers.

How AI assistants pick their sources

From what the companies publish, most AI search tools work in two steps. First they run web searches. Then they read the top results and write an answer with links.

Google calls the first step query fan-out: the AI runs many related searches at once to gather material. OpenAI says ChatGPT search rewrites your question into one or more targeted queries and sends them to search partners. Perplexity runs its own crawler, PerplexityBot, to find pages for its answers.

That has a plain consequence. If a tool’s crawler can’t reach your page, or your page never shows up in its searches, you can’t be cited.

What we saw in 11 prompts

On October 9, 2026, we sent 11 questions from our own field (SERP APIs, rank tracking, and website traffic) to two AI tools. We used the API versions, Perplexity’s sonar model and OpenAI’s gpt-5.4-mini with web search, through DataForSEO. We also pulled 56 Google results pages for similar keywords, US desktop.

A few patterns stood out.

• Perplexity favors whoever publishes the comparison table. Most of the “best SERP API” pages it cited sat on the blogs of companies that sell scraping or SERP tools.
• ChatGPT looks up vendors it already knows. Its searches read like official DataForSEO Google SERP API pricing features. It then cited the pricing and docs pages it found.
• Google cites past page one. Only 24 percent of the URLs cited in AI Overviews also ranked in the organic top 10. YouTube was the most cited site, with 52 of 464 citations.

Serplify itself was not cited in any of these answers. That is a useful reminder that a good product page alone does not get you named.

Treat these numbers as a sketch. The sample is small, it covers one niche, and the consumer apps can behave differently from the API versions. We tested Gemini once, but its links came back as redirect addresses, so we left it out. Our study of Google AI Overviews has a larger Google dataset.

What the research says, and where it is thin

The GEO paper is the main academic study so far. In its lab benchmark, adding quotations, statistics, and cited sources raised how much of a page the AI used. The best methods beat the baseline by about 41 percent. Keyword stuffing did worse than doing nothing.

That is useful, but it was mostly a lab setup with a test engine the team built. It is not proof of how ChatGPT or Google behave today.

Google’s guidance is more direct, at least for its own products:

• There are no extra requirements to appear in AI Overviews or AI Mode. The page has to be indexed and allowed to show a snippet.
• You don’t need special schema markup.
• You don’t need new AI text files. Google says its Search doesn’t use them.

Beyond that, nobody outside these companies knows the exact rules. A tool that promises you a top spot in ChatGPT is guessing.

How to do LLM SEO: six practical steps

1. Get named on other sites. ChatGPT searches for brands it already knows, and Perplexity cites lists and comparisons. Being included in fair “best X” lists, reviews, and forum answers is how you enter that set. Don’t pay for placements that hide that they are paid.
2. Write facts an AI can lift. Put a one or two sentence answer right under each question heading. Give numbers with units and dates. Use tables for comparisons. The same format wins featured snippets.
3. Keep your official pages sharp. Pricing, docs, and spec pages get cited when an AI checks a fact. Make the key number easy to find, and keep it current.
4. Still rank in normal search. AI tools search the web first, so pages that rank well get read more. But don’t stop there: 24 percent overlap means many cited pages sat outside the top 10.
5. Let the right crawlers in. Allow OAI-SearchBot for ChatGPT search and PerplexityBot for Perplexity in your robots.txt. For Google’s AI features, Googlebot and your snippet settings are what count.
6. Publish in the formats each tool cites. Short videos for AI Overviews, where YouTube led. Clear comparison tables for Perplexity. Clean docs and pricing for ChatGPT.

llms.txt is a 2024 proposal for a plain-text map of your site for AI tools. It costs little to add, but there is no public proof that it changes citations. We explain it in what is llms.txt.

How to measure LLM SEO

Each engine needs its own check.

Google AI Overviews and AI Mode. Search Console’s Generative AI performance report counts how often links to your site showed in both features. To see which keywords trigger an overview and who else gets cited, fetch the results pages on a schedule. Our guide on how to track AI Overviews walks through it. A SERP API returns each overview’s cited URLs as data, which makes this cheap to run for hundreds of keywords.

ChatGPT and Perplexity. Write a fixed list of 20 to 50 prompts your buyers would ask. Run them once a month and record whether you are named, whether you are linked, and which competitors appear. Our guide on how to track AI visibility has a simple setup.

Your analytics. Visits from AI tools often arrive as referrals from sites like chatgpt.com or perplexity.ai. Watch them in GA4 over months, not days. Our guide to referral traffic explains where they show up.

Expect noise. AI answers change from one run to the next. Judge your progress by trends across many prompts, not by one answer.