LLM SEO: How to Get Cited by ChatGPT, AI Overviews and Perplexity in 2026

What is LLM SEO, and how is it different from SEO, AEO and GEO?

LLM SEO is the work of getting your pages quoted and linked inside answers written by large language models. Classic SEO wins a ranking. LLM SEO wins a citation: a sentence in the answer with your link on it. The base is the same page, crawlable and indexed. The extra work happens at the passage level, because these systems quote passages, not pages.

The names overlap because the field is three years old. Answer engine optimization (AEO) grew out of featured snippets: state the answer, then explain it. Generative engine optimization (GEO) is the academic name, from a 2024 Princeton paper that tested content edits and measured visibility gains of up to 40%. AI search optimization is the umbrella term most agencies use. We say LLM SEO because it names what reads your page.

One thing hasn't changed. Google's AI features guidance, updated December 2025, says there are no additional requirements to appear in AI Overviews or AI Mode, and no special file, markup or schema. If a page can't be crawled and indexed, nothing can cite it.

How do ChatGPT, AI Overviews, Perplexity and Claude choose what to cite?

Each of them runs a search first and writes second. The search step decides which pages can be cited; the writing step decides which passages get used. The indexes differ, so how to get cited by ChatGPT is a different question from how to rank in AI Overviews. The crawler you allow is the first fork.

Google's AI features use what its docs call query fan-out: multiple related searches across subtopics, combined into one answer. That's why the citations moved off page one. In Ahrefs' March 2026 study of 863,000 keywords, 37.9% of cited pages ranked in the top 10. Another 31.2% ranked 11 to 100, and 31.0% didn't rank in the top 100 at all. In July 2025 the top-10 share was about 76%. AI Mode sits further out: SE Ranking measured 14% URL overlap with the organic top 10 and 12.6 links per answer.

ChatGPT has its own index. OpenAI's crawler docs name three agents. OAI-SearchBot "is used to surface websites in search results in ChatGPT's search features." ChatGPT-User fetches a page when a user asks about it. GPTBot crawls for training. Disallow OAI-SearchBot and you're out of ChatGPT search results. If you want to know how to show up in ChatGPT, that one robots.txt line comes first.

Perplexity works the same way. Its docs say PerplexityBot "is designed to surface and link websites in search results on Perplexity," and Perplexity-User visits pages mid-conversation. Anthropic, in a note updated April 2026, lists Claude-SearchBot for search and Claude-User for pages Claude opens on a user's behalf. If you've connected Claude to your CRM, Claude-User is the agent fetching the pages you point it at.

The platforms also disagree with each other. In Semrush's 13-week study of 230,000 prompts, Wikipedia appeared in roughly 55% of ChatGPT responses at one point and about 2% of AI Mode responses. AI Overviews and AI Mode share only 10.7% of their URLs, per SE Ranking. One page can win on one platform and be invisible on another.

What does zero-click search do to a small business?

It removes the click, not the customer. Pew Research watched 900 US adults make 68,879 Google searches in March 2025. With an AI summary on the page, users clicked a result in 8% of visits; without one, 15%. They clicked a link inside the summary in 1% of visits. For a local service business, the answer gets read on Google, and the phone call is the click.

Ahrefs put numbers on the loss by position, comparing 300,000 keywords in December 2023 and December 2025. The top result lost 58% of its clicks when an AI Overview appeared; position 10 still lost 19.4%. Zero-click search predates AI, though. SparkToro and Datos measured 58.5% of US Google searches ending without a click in 2024. AI widened a gap that already existed.

Two things soften it. First, the visitors who still click are further along. Semrush's July 2025 model puts an LLM visitor at 4.4x the value of an organic visitor, across digital-marketing topics. Ahrefs reported that 0.5% of its visitors came from AI search and drove 12.1% of signups, a 23x gap. Both are one-company or one-industry figures: direction, not a forecast for your business.

Second, the volume is still small. Ahrefs estimates ChatGPT gets about 12% of Google's search volume for Google-style queries, and Google sends 190x more traffic to websites. So the 2026 job is to be the business the answer names while classic search pays the bills. AI referrals arrive with intent, so we pair this work with a fast reply on the other side.

Is AI-written content safe with Google?

Yes, if a person checked it and it adds something. Google's helpful-content page says using automation, including AI, "to produce content for the primary purpose of manipulating search rankings" violates its spam policies. The spam policy itself, updated August 2026, defines scaled content abuse as many pages made to manipulate rankings, "no matter how it's created." The tool isn't the problem. Volume without review is.

The same page suggests disclosing AI help wherever a reader might ask how the page was made. We do, in the last section. The failure patterns are consistent: pages that restate what already ranks, invented statistics, near-duplicate location pages by the hundred, and no named author. Those fail in AI answers too. A passage with a fabricated number is a passage nobody wants to quote.

How to rank in AI Overviews: what the 2026 data says

Answer the question in the first paragraph, on a page that already ranks somewhere for a related query, and get mentioned elsewhere on the web. That's the short version of how to rank in AI Overviews, and each part has a number behind it.

Question queries are where AI Overviews live. Ahrefs found they trigger on 21% of all keywords but 57.9% of question queries (January 2026). Fan-out means you don't need position one. You need to be the best answer to one of the sub-questions Google asks itself. That's the case for question headings, each followed by a paragraph that stands alone.

Off the page, the strongest signal Ahrefs found across 75,000 brands was web mentions, at a correlation of 0.664 with AI Overview visibility. Backlinks came in at 0.218. Brands in the top quartile for mentions averaged 169 AI Overview mentions against 14 for the next quartile. For a small business that means reviews, directory listings, local press and podcasts count, even without a link.

Behavior differs by surface. Kevin Indig's May 2026 clickstream study covered 846,000 US sessions. In 88% of AI Mode tasks users took the AI's shortlist as-is, 74% picked the first item, and 64% clicked nothing. In AI Mode the shortlist is the whole game. In AI Overviews, users still compare, so the page under the answer earns its keep.

The LLM SEO on-page checklist

Here's the on-page half of LLM SEO as we run it, in order. None of it is exotic. Most of it is what a careful writer did anyway, made consistent and checked every time.

1. 01Answer first, in about 60 words. The first paragraph under every heading answers that heading's question completely, so one passage can be quoted alone.
2. 02Headings as questions. Phrase 60 to 70% of H2s the way a buyer would ask. Question queries trigger AI Overviews nearly three times as often as keywords overall.
3. 03One number, one source, one link. Every statistic carries a tier 1 to 3 source linked beside it, with the page's date. Unsourced numbers come out.
4. 04Tables with source captions. Tables lift cleanly into answers, and the caption travels with the data. A comparison with more than two numbers becomes a table.
5. 05Structured data you already have. Article, Person, Organization and FAQ schema, kept accurate. Google says no special schema is needed, so don't invent any.
6. 06Entity consistency. The same name, address, phone and description on your site, Google Business Profile, directories and review platforms. Mentions correlate with visibility; mismatches erode them.
7. 07Original data. One thing per post only you know: a price table you compiled, a process you run, a number from your analytics. The GEO paper measured visibility gains of up to 40% from content edits, and found the effect varies by domain.
8. 08Crawler check. robots.txt allows Googlebot, OAI-SearchBot, PerplexityBot and Claude-SearchBot. Check your CDN's bot rules too.

The tools we run this list with, including the fetcher that verifies sources, are in our small-business AI tool stack.

Does llms.txt help?

Not in 2026. llms.txt is a proposed plain-text index of your site for language models. Ahrefs checked 137,210 domains against May 2026 traffic and found that 97% of llms.txt files receive zero requests. Its conclusion was that no AI system goes looking for one. Google's AI features page says you don't need to create "AI text files," and it never mentions llms.txt by name.

It costs nothing to publish, so keep one if you have it. Just don't count it as work done. The hour goes further on a sourced table or a question heading, both of which demonstrably get read.

How do you measure AI citations?

Three ways, cheapest first: AI referrals in GA4, a monthly prompt panel you run by hand, and a paid tracker once the first two show movement. Search Console won't separate the AI surfaces. Google's docs say AI Overview and AI Mode traffic is included in the Web search type of the Performance report.

The panel matters more than it sounds. Semrush saw Reddit's ChatGPT citation share fall from close to 60% inside the study's 13 weeks, and not on every platform at once. A single check tells you nothing; a dated monthly log shows a trend. If your CRM is GoHighLevel, we pull the how-did-you-hear answers with Claude through the MCP server, one of the ten use cases we run.

How we write this blog to get cited

We draft these posts with Claude and check them by hand, and the checking is the part that earns citations. Each rule below exists because the shortcut failed at least once. What follows is the mechanism, not a claim of results.

A brief sets the H2s first: seven to nine of them, 60 to 70% phrased as questions, the primary keyword in two. Then every number is fetched live before a word is drafted. The fetcher returns the page's verbatim text, and that quote goes into a sources file with the page's date. One of the old posts carried a "30 to 40%" range for the GEO paper; the paper says up to 40%. Only the verbatim rule caught it.

The draft goes into the site's typed content model: an answer-first paragraph under each heading, sourced tables, charts from real figures. Two reviewers score it, one for voice and one for SEO, before it goes to Samar for approval. Nothing publishes on the strength of the message that says it's fine.

What we don't do: quote client numbers outside the results page, invent a case study, or report a citation we haven't seen. We'll add the first logged citation here when the panel records one. Our roundup of the 2026 small-business AI numbers is where we send readers who want the wider picture. If you'd like the same treatment for your site, that's our SEO service, and it starts with a conversation.