Guides · LLM Optimization
What is LLM Optimization (LLMO)?
LLM Optimization is making a site usable by large language models and the tools built on them, which in practice means publishing machine-readable versions of your content and letting the crawlers that feed those models fetch it.
The part that is actually distinct
GEO, AEO and AI Overviews are all about how your words are used once a machine has them. LLMO is mostly about whether it can get them at all, and in what form.
That makes it the least glamorous of the four and the one with the most concrete tasks, because access and format are things you either have or do not.
Who is allowed to read you
Model crawlers identify themselves: GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot and others. Your robots.txt either lets them in or does not, and many sites block them without anybody having decided to — a security product, a Cloudflare setting or a copied robots.txt did it.
Blocking them is a legitimate choice. Blocking them while wanting to be quoted in AI answers is not a choice, it is a contradiction, and it is worth knowing which of the two you have.
The asymmetry is worth understanding before you decide. A crawler that trains a model and a crawler that fetches a page to answer somebody's question right now are different things wearing similar clothes. Blocking the first costs you nothing today; blocking the second removes you from answers being written this minute.
llms.txt and its long form
A plain-text file at the root saying what the site is and where its important pages are, with a companion llms-full.txt carrying the text of them. Neither is a standard, nobody is obliged to read them, and no search engine promises anything for having them.
They are cheap and they are the sort of thing that becomes expected before it becomes required. The one thing that matters is that they stay true: they are hand-written and they quote facts that live elsewhere, so a price or a claim in them goes stale silently while being read aloud to strangers by a machine.
Plain-text versions of pages
A page served as markdown when the request asks for markdown saves a machine from downloading HTML and guessing which parts are content. More usefully, it removes the guessing: what gets quoted is what you meant rather than what its stripper happened to keep.
It has to be the same content at the same address, chosen by the header the client sent. Serving different content to machines on any other basis is cloaking, and the distinction is that here the client asked.
Does this apply to your site?
We check for this automatically, along with everything else on the same page of your site, and tell you the exact line. The first audit is on us.