LLM SEO is making your business retrievable and credible to large language models at the moment they answer a customer's question. You cannot edit what a model learned in training; you can change what it retrieves and corroborates today. That single distinction separates the real service from the mysticism, so it is worth seeing the mechanism plainly.
How an LLM decides who to name
In AI search mode, the model does more than recite memory. It rewrites the question into several hidden search queries, retrieves candidate pages from live indexes, and weighs the candidates: does more than one independent source say the same thing about this business? Is the information dated and current? Is the public record consistent? The answer then names the businesses whose evidence held up. You cannot control the final generation, but every earlier step responds to evidence you can change, and every step is inspectable if you measure.
What you can and cannot optimise
- Can: pages that open with the answer rather than a slogan; prices and comparisons as real text, not images; visible dates; a public record that says the same thing everywhere the model checks; crawlers left unblocked (OpenAI documents its crawlers publicly, including the OAI-SearchBot that feeds ChatGPT search); independent sources that corroborate you.
- Cannot: injection into training data, paid placement in organic answers, guaranteed outcomes. Nobody controls the engines. Anyone selling certainty in this channel is selling something else.
Timelines follow from the mechanism: retrieval-level changes can show in answers within weeks of a re-crawl, because the lag is crawler re-indexing, not model retraining. Corroboration that wins competitive questions takes months. The long version, with the do-it-yourself basics, is in the ChatGPT guide.
Searched for a different name?
This discipline is young enough that the market has not settled on one name for it. The same work is sold as: AEO agency · GEO agency · AI SEO agency · AI visibility agency · AI search optimisation agency. All of them, underneath, are the same retrieval and corroboration work this page just described.
Common questions
Can you get a business into an LLM's training data?
Not honestly, and it is the wrong lever anyway. Training runs are infrequent and uncontrollable from outside, and for buying advice, live retrieval and corroborating sources count for more than anything a model memorised. The lever that works is retrieval: what the model finds and corroborates about you at answer time. Retrieval changes when your public evidence changes, which is doable in weeks, not model-training years.
Which LLMs matter for UK customers?
The core baseline in 2026: ChatGPT (retrieving via live search, with Bing's index a major source), Google AI Overviews, and Perplexity; those three are what our published disagreement data covers. Microsoft Copilot is worth adding where your buyers live in Microsoft tools, and Bing's index influences both ChatGPT retrieval and Copilot. The engines do not agree with each other: they disagree on 29% of questions in our data, so visibility on one says little about the others.
Is LLM SEO measurable?
Yes, if it is done with repeats. A single answer is a coin flip dressed as a fact: identical questions change their verdict 19% of the time in our data. The honest method asks the same questions repeatedly across engines, scores each answer as recommended, mentioned or absent, and quotes proportions with a confidence range. Anything else is anecdote.
If retrieval is your worry, start by seeing it: the free ten-question audit shows what each engine currently retrieves and says about you, question by question. Nothing worth fixing means exactly that, and we say so. Pricing is public: pilot from £500, monthly from £700.
Updated July 2026