What is AI search optimization?

It is the work of getting AI answer engines such as ChatGPT, Google's AI Overviews and AI Mode, Perplexity, Gemini, Copilot and Claude to cite your pages. In TIN's measurement of everyday questions, most cited sites appeared in only one engine's answers, so it is several channels, not one.

What is AI search optimization called?

It goes by several names: generative engine optimization (GEO), answer engine optimization (AEO), LLM SEO and AI search SEO. TIN treats them as one practice.

How does AI search engine optimization work?

An answer engine fetches a page, splits it into blocks, scores each block and shows the model a few of them. In a controlled test of six models on B2C product review posts, topic match, position in the model's context, a recent date and a stated price decided which source was cited; formatting showed no consistent effect.

Is SEO still worth it in 2026?

On TIN's data, yes, as a correlation rather than a proven lever. Across AI and automation agencies, ranking keywords and AI citations rose together across all AI platforms, and in the US some of Google's top-10 organic domains were also cited in the AI Overview. Most domains the AI Overview cited were outside the top 10, so ranking is not enough on its own.

Does schema markup or llms.txt help AI search optimization?

Not on the evidence TIN has read. In an Ahrefs study of pages that added schema, the changes on AI Mode and ChatGPT were statistically indistinguishable from zero and AI Overviews declined. Most llms.txt files received no traffic, and that study did not measure citations. Google lists llms.txt, special markup and overfocusing on structured data among things you can ignore for Google Search.

Should I optimize for each AI engine separately?

Yes. In TIN's study of everyday questions across seven engines, most cited sites appeared in only one engine's answers, and each engine leaned on a different kind of source. The study's own advice is not to optimise for AI as one channel.

AI search optimization: what it is, what is measured, and where to start, The Internet Ninja

Q: How do I know if AI search optimization is working?

Measure each engine on its own, with repeated runs per prompt. Martinez's survey reports that Schulte et al., on a small set of Swiss queries, observed daily source-level Jaccard scores of approximately 0.34 to 0.42, and proposed seven to eight repetitions per prompt as a starting point.

The short answer

AI search optimization is the work of getting AI answer engines (ChatGPT, Google’s AI Overviews and AI Mode, Perplexity, Gemini, Copilot, Claude) to cite your pages. It is not one channel. TIN asked seven engines the same 60 questions in 10 countries and coded all 1,395 cited sites: 1,091 of them (78%) appear in only one engine’s answers source. A controlled test favours matching the question and a current date source; crawler logs say to serve the text without JavaScript source.

That study covers a sample of everyday searches, not business-to-business questions, and it is one weekly run, “a snapshot, not a trend” source.

What AI search optimization covers, and what it is called

The practice has one goal: when someone asks an AI engine a question your business can answer, the engine retrieves your page, or a page that names you, and cites it in the answer. A search ranking puts a link in a list. A citation puts a source inside the answer.

The mechanism, as TIN’s retrieval post describes it: “An answer engine fetches your page, splits it into blocks, scores each block on its own, and shows the model a handful of them” source.

It travels under several names: generative engine optimization (GEO), the academic name from the KDD 2024 paper that named it source, answer engine optimization (AEO), LLM SEO, and AI search SEO or AI search optimisation. TIN treats them as one practice.

Each engine is its own channel (TIN’s measurement)

TIN’s AI citation sources study, edition 2026-10-02, put the same questions to seven engines and sorted every cited site into ten kinds of source source. The engines do not share a source diet.

Engine	Citations	Largest category	The study’s advice line
ChatGPT	299	Government (50.5%)	“For ChatGPT, be cited by the institutions in your field.”
Perplexity	1,099	Company sites (38.9%)	“For Gemini and Perplexity, your own site can be the source.”
Gemini	218	Company sites (58.3%)	“For Gemini and Perplexity, your own site can be the source.”
Copilot	523	Company sites (33.3%)	“Comparison sites stand between brands and buyers … Being listed fairly on them is part of AI visibility.”
Claude	47	Comparison (40.4%), on few citations	”Comparison sites stand between brands and buyers … Being listed fairly on them is part of AI visibility.”
AI Overviews	184	Company sites (39.1%)	“For Google’s AI answers, publish where people watch and discuss.”
AI Mode	487	Company sites (38.6%)	“For Google’s AI answers, publish where people watch and discuss.”

Source for every row: source. Claude’s 47 citations are too few to carry a claim on their own.

“ChatGPT sends 68% of its citations to government and associations; Gemini sends 58% to company sites” source. The 68% is government (50.5%) and associations (17.1%) together source. “YouTube is the most-cited site in Google AI Overviews and AI Mode; ChatGPT cites no social or video platform” source. “ChatGPT shares at most 1 of its 25 most-cited sites with any other engine” source.

Copilot is the outlier on small publishers: it “sends 28% of its citations to small single-topic sites, more than any other engine” source. Across all seven, “Company sites are the largest category overall, with 37% of all citations” source, so a company’s own pages are the single most common kind of source, but not on ChatGPT, whose largest category is government at 50.5%, or on Claude, whose largest is comparison sites at 40.4% source.

The study’s own advice: “Do not optimise for “AI” as one channel” source.

The business-to-business picture

The citation sources study does not cover business-to-business questions source. TIN’s agency study is the counterweight: 238 AI and automation agencies listed in seven partner directories, counted in one index, Ahrefs, on 2026-09-26 source. “161 of 238 agencies were cited at least once by an AI platform, but the median agency was cited 3 times” source.

The engine where an agency is invisible differs. Counted by platform, the agencies never cited were:

Platform	Agencies never cited
All AI platforms	77 of 238
Google AI Overviews	140 of 238
ChatGPT	173 of 238

Source for every row: source.

Citations also concentrate: “The 24 most-cited agencies hold 78.4% of all 29,717 AI citations” source. The study’s own advice: “Expect ChatGPT to be the last place you show up” source.

What the evidence supports, and what it does not

The ledger below grades each tactic by the strongest evidence TIN has read for it.

Tactic	What was measured	Strength	Engine	Verdict
Match the topic of the question	Six models, 18 factors, 252,000 trials; unanimous across all six models source	Controlled test	Six language models in a retrieval testbed (Vishwakarma et al.)	Decisive in that testbed
A recent date against a stale one	Vishwakarma et al., SIGIR 2026; unanimous across all six models source	Controlled test	Six language models in a retrieval testbed of B2C product review posts	Decisive in that testbed
A stated price, on product queries	Vishwakarma et al., SIGIR 2026; unanimous across all six models source	Controlled test	Six language models in a retrieval testbed of B2C product review posts	Decisive in that testbed
Organised sections against dense prose	An odds ratio of 0.78 to 1.68 depending on the model in Vishwakarma et al., which TIN’s GEO post reads as a range that straddles no effect source	Controlled test	Six language models in a retrieval testbed	No consistent effect
Text readable without JavaScript	”None of the major AI crawlers currently render JavaScript.” source	Crawler logs	The major AI crawlers	Put the text in the HTML
Mentions over backlinks	Across 75,000 brands, YouTube mentions correlated at “~0.737” across ChatGPT, AI Mode and AI Overviews; branded web mentions “(0.709 vs. 0.664 and 0.656)”, AI Mode against ChatGPT and AI Overviews; link metrics were “very weak” source	Correlation	AI Overviews, AI Mode, ChatGPT	Associated, not proven
Search authority	”Ranking keywords and AI citations rise together (Spearman 0.92)” source	Correlation	All AI platforms combined, agencies only	Associated, not proven
Schema markup	1,885 pages that added schema: AI Mode +2.4% and ChatGPT +2.2%, “Statistically indistinguishable from zero”; AI Overviews a 4.6% “Small but statistically significant decline relative to matched controls” source	Pages tracked after adding schema	AI Mode, ChatGPT, AI Overviews	No major uplift on any engine; a small decline on AI Overviews
llms.txt	Of 137,210 domains, 28% publish an llms.txt file; “97% of those files received zero traffic in May 2026” source	Request counts	Not engine-specific	Rarely fetched; effect on citations not measured

TIN’s reading of each outside study is in its generative engine optimization post, with the source tables source. The controlled experiment is Vishwakarma, Kumar and Jamidar, presented at SIGIR 2026 source. Its fourth unanimous factor, position in the model’s context, is not something a publisher sets directly, so the ledger leaves it out.

On block length, the retrieval post calls 150 to 300 words “the best-supported range”, and says it is “not a measured optimum” source.

One more finding sets the ceiling on all of it. In a study of 112 startups across 2,240 queries, ChatGPT recognised a product by name 99.4% of the time and Perplexity 94.3%; asked the buyer’s question without the name, the products appeared 3.32% and 8.29% of the time source. Being known to the model is not the same as being recommended by it.

Where Google and the ChatGPT record disagree

Google’s own guide to its generative AI features, last updated 2026-07-10, lists things “you can ignore for Google Search”: llms.txt and other “special” markup, chunking content, rewriting for AI, inauthentic mentions, and overfocusing on structured data source. On chunking it says: “There’s no requirement to break your content into tiny pieces for AI to better understand it” source.

That is Google speaking about Google. TIN’s retrieval post records ChatGPT retrieval scoring blocks one at a time source. Both can be true, because they describe different engines. TIN’s inference, not a measured finding: a page written in self-contained, well-sourced blocks costs nothing on Google and fits what is observed on ChatGPT.

Does SEO still count?

Partly, and the size of “partly” depends on the engine. For Google AI Overviews in the US: “30.7% of top-10 organic domains are also cited in the AI Overview (152 of 495, across 69 searches)”, and “63.0% of the domains US AI Overviews cite are not among that search’s top-10 organic domains (259 of 411)” source. Across AI platforms, agencies with more ranking keywords drew more AI citations (Spearman 0.92) source.

Both are correlations in one snapshot, not causes source. Keep the SEO that builds authority and crawlable pages; do not expect a ranking to deliver a citation on its own.

The studies behind this page

A Recon study is TIN’s own measurement across many sites against one business question. The method and its amendments are published on the study page, and the dataset is downloadable, free and ungated.

Study	Question	Edition	Sample	Dataset
AI citation sources	Which kinds of site do AI engines cite?	2026-10-02	1,395 cited sites from 60 questions in 10 countries on 7 engines	data.csv
Agency AI citations	How often do AI platforms cite AI and automation agencies, and what tracks it?	2026-09-27	238 agencies from 7 partner directories	data.csv
Google top 10 vs AI Overviews	How often is a Google top-10 site also cited in the AI Overview?	2026-10-02	438 buyer searches in 7 markets	data.csv

In the citation sources study, “The method, the category scheme and the audience layer were committed before the data they govern was read” source. The agency study ran a pre-registered analysis plan, “amended three times (v1.1 to v1.3)”, with the population “frozen before any Ahrefs call” source. The Google study froze its buyer intents “before any search phrase was chosen” and publishes five dated amendments source.

How to start, by where you are

These paths are TIN’s inference from the studies above, not a measured finding.

You already rank on Google. Start with Google AI Overviews, where the gap between your ranking and the citation is measurable search by search.

Your buyers ask ChatGPT. Start with ChatGPT, and expect it to take longest. ChatGPT’s sources lean on institutions, so the work is getting named by the regulators, trade bodies and standards groups in your field, not rewriting your own blog.

You do not know which engines matter to your buyers. Start by tracking. Run your buyers’ real questions on each engine, repeatedly, before you change a page. Without a per-engine baseline you cannot tell a gain from run-to-run noise. Martinez’s survey reports that Schulte et al., on a small set of Swiss queries, observed daily source-level Jaccard scores of approximately 0.34 to 0.42, and proposed “seven to eight repetitions per prompt as a starting point” source.

What this page cannot tell you

Business-to-business questions in general. The citation sources study used 60 everyday questions and does not cover business-to-business questions. The agency study covers AI and automation agencies only source.
Trends. Each study is one snapshot: one weekly run, one index on one date, one crawl window on 2026-10-01 and 2026-10-02 source.
One engine from another’s data. The Google study is Google AI Overviews only. The agency correlations are across all AI platforms combined, not any one engine.
Causes. All three studies measure associations. The agency study says so directly: “Nothing here shows that raising Domain Rating, rankings or case-study quality raises AI citations” source.
That this works for TIN. TIN measures how engines cite; it does not claim they cite TIN. In its own words, “our AI visibility baseline is 0%” source.

Sources

TIN, “AI citation sources” study, edition 2026-10-02. https://theinternetninja.com/recon/ai-citation-sources-2026/
TIN, “Agency AI citations” study, edition 2026-09-27, Ahrefs AI index 2026-09-26. https://theinternetninja.com/recon/agency-ai-citations-2026/
TIN, “Google top 10 vs AI Overviews” study, edition 2026-10-02. https://theinternetninja.com/recon/google-top10-vs-ai-overviews-2026/
TIN, “Generative engine optimization: what the controlled evidence shows”. https://theinternetninja.com/blog/generative-engine-optimization-what-the-controlled-evidence-shows-2026/
TIN, “ChatGPT ranks passages, not pages”. https://theinternetninja.com/blog/ai-search-optimization-what-the-retrieval-record-shows-2026/
TIN, “ChatGPT stopped citing Reddit”. https://theinternetninja.com/blog/chatgpt-stopped-citing-reddit-substitute-platforms-2026/
Google Search Central, guide to optimizing for generative AI features, last updated 2026-07-10. https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
Vishwakarma, Kumar and Jamidar, SIGIR 2026, controlled experiment on citation factors. https://arxiv.org/abs/2605.25517
Vercel with MERJ, “The rise of the AI crawler”, 17 December 2024. https://vercel.com/blog/the-rise-of-the-ai-crawler
Ahrefs, AI brand visibility correlations, 75,000 brands. https://ahrefs.com/blog/ai-brand-visibility-correlations
Louise Linehan, Ahrefs, “We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved.”, 11 May 2026. https://ahrefs.com/blog/schema-ai-citations/
Ahrefs, llms.txt study. https://ahrefs.com/blog/llmstxt-study/
Amit Prakash Sharma, “The Discovery Gap: How Product Hunt Startups Vanish in LLM Organic Discovery Queries”, preprint, January 2026. https://arxiv.org/abs/2601.00912
Olivier Martinez, “Optimizing Visibility in Generative Engines: A Critical Survey of Generative Engine Optimization (2023-2026)”, 2026-07-15. https://arxiv.org/abs/2607.14035
Aggarwal et al., “GEO: Generative Engine Optimization”, KDD 2024. https://arxiv.org/abs/2311.09735
TIN case file, Google AI Overviews and the glue-on-pizza answers, 2024. https://theinternetninja.com/stories/google-ai-overviews-glue-pizza-eat-rocks-scaled-back-2024/

AI search optimization: what it is, what is measured, and where to start