LLM SEO: How to Get Your Business Recommended by AI Models
LLM SEO is the work of making your business the one a large language model mentions — when someone asks ChatGPT, Perplexity, Gemini, or Claude who to hire, what to buy, or which firm to trust. There are exactly two routes into those answers: the model's trained knowledge, and the live web search it runs before responding. Both can be influenced. Neither can be bought. Here's how each one works, the work that moves them, and how to measure any of it without kidding yourself.
Two doors into an AI answer
When a language model answers "who's a good SEO agency for tradesmen?", the answer is assembled from two sources, and the distinction matters because you work on them differently.
Door one: the model's trained knowledge. Everything the model absorbed about the web up to its training cutoff. If your business was described consistently across enough of the web — your site, directories, press, reviews, forum threads — the model "knows" you the way it knows any fact. This door moves slowly: it updates when a new model version is trained on a fresher snapshot of the web. Work done today pays out in a future model.
Door two: live retrieval. Most AI tools now search the web before answering anything current or local — ChatGPT search leans on Bing's index, Gemini on Google's, Perplexity runs its own crawler. The model reads the top results and writes its answer from them, usually with citations. This door moves at the speed of ordinary SEO: rank for the underlying query and you're in the source pool, today.
Almost every "LLM SEO secret" being sold is just one of these doors renamed. Keep the two-door picture in your head and the whole field stops being mysterious.
What a language model needs to know about you
Models deal in entities — distinct things with names, attributes, and relationships. Your job is to make your business an unambiguous entity: one exact name, one clear description of what you do and where, repeated consistently everywhere you appear. "RankLocal Agency, a local SEO agency in West Drayton serving London and the Home Counties" — same shape on the website, the Google Business Profile, every directory, every social bio.
Inconsistency is expensive here in a way it never quite was in classic SEO. A model that's seen three different names, two addresses, and four vague descriptions doesn't average them into a strong entity — it ends up with a blurry one it's less confident recommending. The boring hygiene work (same NAP everywhere, a proper About page, Organization and LocalBusiness schema, real author names on content) is entity-building, and entity-building is door one.
Third-party corroboration seals it. Models weight what others say about you — reviews, directory listings, press mentions, forum recommendations — above what you say about yourself, for the same reason Google always has. This is where GEO as a discipline earns its name: building the citation network that makes generative engines treat your business as an established fact.
Writing pages a model can lift an answer from
Door two — retrieval — rewards a specific shape of content. The model skims the pages it retrieved and extracts passages that answer the question; pages built from self-contained, factual, plainly worded passages get quoted, and pages that bury the answer under six paragraphs of throat-clearing don't.
Practically: lead every section with the answer, then elaborate. State facts a model can repeat safely — prices, timeframes, coverage areas, credentials — rather than adjectives it will discard. Use real headings that match how questions are asked. Add FAQ and article schema so the structure is machine-readable. And keep the site fast and crawlable for AI user agents, because a page the crawler can't fetch is a page that can't be cited — your robots.txt shouldn't be blocking GPTBot, PerplexityBot, or Google-Extended unless you've decided that on purpose.
An llms.txt file — a curated plain-text map of your site for AI crawlers — belongs in the same bucket: cheap, sensible, not magic. (You're reading a site that practises this; our own llms.txt lists every page on it.)
The unfashionable truth: classic SEO is the entry fee
Retrieval-augmented answers are built from search results. That means the LLM answer layer sits on top of the ranking layer, and no amount of AI-specific optimisation rescues a site that doesn't rank for anything. Technical health, genuinely useful pages, links and reputation — the whole traditional stack still decides whether you're in the pool of sources the model reads at all.
So the honest sequencing for a small business is: get the classic foundations right, layer the entity and passage work on top, then build the off-site footprint. If you want the map of how this AI-search work divides up — spoken answers versus generated citations — our GEO vs AEO explainer draws the lines, and the ChatGPT-specific playbook applies all of this to the one assistant your customers actually name.
Measuring AI visibility without kidding yourself
Three measurements are real. Referral traffic: visits from chatgpt.com, perplexity.ai, gemini.google.com and friends show up in your analytics as referrers — small numbers for local businesses today, but growing, and disproportionately high-intent. Assisted conversions: "found you when I asked ChatGPT" is worth adding to your how-did-you-hear question, because plenty of AI-influenced customers arrive by typing your name into Google afterwards. Direct sampling: ask the major assistants your money questions monthly — "best [trade] in [town]", "recommend an agency for X" — and log who gets named.
One warning on sampling: answers vary by phrasing, session, and day. A single "we're in!" screenshot is anecdote, not measurement — track the same question set over months and watch the trend instead. It's the AI equivalent of the geo-grid we run for map pack work: same questions, same cadence, honest picture.
Common questions about LLM SEO
Is LLM SEO different from normal SEO?
Less than the name suggests. When an AI tool searches the web to answer a question, it's reading pages a search index surfaced — so the classic work of being crawlable, fast, and clearly written still gates everything. What LLM SEO adds on top is entity clarity (models need to know exactly who you are and what you do), citable self-contained passages, and a wider footprint of consistent mentions across the web for models to learn from.
Can you pay to appear in ChatGPT or Perplexity answers?
Not in the organic answers. Perplexity has experimented with sponsored follow-up questions and OpenAI with shopping formats, but the recommendations themselves aren't for sale — they're assembled from the model's knowledge and live retrieval. Anyone selling guaranteed placement inside AI answers is selling something they don't control. What you can buy is the work that makes appearing more likely.
How long does it take to show up in LLM answers?
Retrieval-based answers can change in weeks: publish a genuinely useful, well-structured page, get it indexed, and AI search tools can start citing it as soon as it ranks for the underlying query. Being embedded in the model's trained knowledge is slower — that depends on your presence across the wider web at the time the next model version is trained, which is why consistent mentions and reviews now pay off quarters later.
Does llms.txt actually do anything?
It's a low-cost hedge, not a lever. The llms.txt convention gives AI crawlers a curated map of your site in plain text; adoption by the major AI companies remains partial and undeclared. It takes an hour to add, can only help, and we ship one on every site we build — but nobody should sell it to you as the thing that gets you into AI answers. The content it points at does that.
Daniel Gardner
Founder, RankLocal Agency
Local SEO specialist working with London tradesmen and small service businesses. Focuses on Google Business Profile optimisation, Astro web builds, and rank tracking for the map pack and organic results.
LinkedInFind out what the AI models currently say about you.
Our free SEO audit — the Visibility Snapshot — includes an AI search readiness check: whether the assistants can find you, what they say when asked, and which of the two doors needs work first. Written report, reply within 24 hours — usually same day.