Ask ChatGPT for the best payroll tool for a small team, or a good dentist in your city, and you get an answer: a short list of names, a sentence on each, sometimes a few links. No page two, no ten blue links. Either you're in that answer or you're not.
LLM SEO is the work of getting into it. LLM stands for large language model, the technology behind ChatGPT, Claude, Gemini, Perplexity and Copilot. You'll also see it called generative engine optimisation (GEO), answer engine optimisation (AEO) or AI search optimisation. They describe the same goal: being named, accurately and favourably, when someone asks an AI tool a question your business should be the answer to.
How an AI tool decides who to name
There are two ways a model knows about you, and they behave differently.
What it learned in training. Models are trained on a huge snapshot of the public web. If your category is discussed widely, and you're part of that discussion, the model has a view of you before anyone asks. That view changes slowly, only when a new model version is trained.
What it finds when it searches. Most assistants now search the web before they answer, especially for questions about products, services and anything recent. They read a handful of pages, then summarise. This is where things move quickly: change what those pages say, and the answer can change within weeks.
In both cases the model is mostly reading what other people say about you, not what you say about yourself. A forum thread where real customers compare options, a review platform, a trade publication, a reference page: these carry more weight than your homepage, because they look independent. We wrote about why in why a model trusts a 300-upvote thread more than your homepage.
How LLM SEO differs from traditional SEO
Traditional SEO and LLM SEO overlap, but they're not the same job.
| Traditional SEO | LLM SEO | |
|---|---|---|
| The result | A position in a list of links | A mention inside an answer |
| What you work on | Mostly your own site | Mostly other people's sites |
| Main signals | Relevance, links, technical health | Corroboration across sources AI trusts |
| How you measure it | Rankings and clicks | Share of answer, accuracy, sources cited |
| What a win looks like | Page one | Named, and described well |
The overlap is real. AI tools that search the web lean on pages that rank, so a site Google can't read is a site AI tools can't read either. Clear pages, fast loading and sensible structure still matter. But a perfect site with no independent mentions rarely gets recommended, because the model has nothing to corroborate it with.
What actually moves the answer
In our audits, the sources behind AI answers usually fall into a short list:
- Communities and forums. Reddit above all, plus niche and local forums where buyers ask each other for advice. Models quote these because they read as unsponsored opinion.
- Review platforms. Google, Trustpilot, G2, Capterra and their equivalents. Rating, volume and recency all show up in how a model describes you.
- Reference pages. Wikipedia first, where you genuinely qualify. Models treat it as a factual baseline.
- Press and trade publications. Especially roundups and comparisons in your category.
- Directories and listicles. "Best X for Y" pages that already rank are often the exact pages an assistant reads.
- Your own site. Still worth getting right: clear service pages, accurate facts, structured data. It confirms what others say rather than replacing it.
The work is earning a true, positive presence in the places on that list that matter for your category, in the open and by each platform's rules. That means disclosed accounts on Reddit, honest reviews from real customers, and Wikipedia edits that follow its conflict-of-interest rules.
Where to start
1. Find out what AI says about you now. Pick ten to twenty questions your buyers really ask. Put them to the main models and note who gets named, how you're described, and which sources each answer draws on. Our guide to baselining what AI says about you walks through it.
2. List the sources behind the answers. You'll usually find the same few threads, review pages and articles coming up again and again. That's your map.
3. Fix what's wrong before adding what's missing. Outdated prices, a product you discontinued, an old complaint that was resolved: inaccuracies are often quicker to correct than gaps are to fill, and they do more damage.
4. Earn presence where you're absent. Join the conversations that already shape the answer, ask happy customers for reviews on the platforms models read, and get included in the comparisons that rank.
5. Measure the same way every month. Same questions, same models, same settings. Our post on GEO KPIs covers what to track and what to ignore.
What this doesn't mean
A few honest limits, because this field attracts a lot of overpromising:
- Nobody can guarantee a mention. Models change, answers vary from one run to the next, and anyone promising "#1 on ChatGPT" is guessing. You can raise your odds and measure them; you can't fix the outcome.
- It isn't a trick you can run on a model. Hidden text, keyword stuffing aimed at AI, or fake accounts posting about you don't last, and they can damage how you're described for a long time.
- Your SEO work isn't wasted. A healthy, well-structured site is still part of the picture. LLM SEO adds to it rather than replacing it.
- It isn't instant. Answers built from live search can shift in weeks. What a model learned in training changes only when a new version comes out.
The short version
LLM SEO is about being the name an AI tool gives when your buyers ask for a recommendation. You get there mostly by what others say about you, in the communities, reviews and references models trust, backed by a site that's easy to read. Start by finding out what the answer says today.
If you'd like that done properly, our AI visibility audit puts your buyers' questions to five models, records every source behind each answer, and gives you a ranked plan. Or see how we run the full programme on our AI visibility page.