LLM SEO: how to get cited by AI assistants
LLM SEO is the work of getting large language models to find, understand and quote your business when someone asks them a question in your market. In practice it comes down to four things: answering directly and early, keeping your facts identical everywhere a model might read them, structuring pages so a parser can follow them, and having something specific to say that the general consensus does not already contain.
That is the whole answer. The rest of this explains why each part works, and which popular tactics do nothing.
Why this is a different job from ranking
A search engine reads your page and decides where to place it in a list. A language model reads your page alongside a great many others and decides what is true, then writes a paragraph without a list at all.
The consequence is that the unit of success changes. In search, the unit is a position. In an assistant’s answer, the unit is a sentence: yours, repeated, usually with your name attached and sometimes with a link.
This has an uncomfortable implication. There is no page two. If the assistant does not mention you, you were not almost mentioned.
What appears to influence a citation
Answering in the first paragraph
Models quote openers. A page that opens with three paragraphs of preamble before reaching the point gives a model nothing convenient to lift, and models are built to be efficient about what they lift.
Put the plain answer first, then the depth. This article does it, and so does every entry in our glossary. It is also just better writing.
Self-contained sentences
A sentence that begins “as mentioned above” or “this means that” cannot be extracted. It only makes sense in place.
Write so that any given sentence would survive being quoted on its own. This is a small constraint that changes prose noticeably once you apply it.
Facts that agree with each other
If your founding year, service list, location or business description differ between your website, your LinkedIn page and a directory listing, a model has no principled way to choose between them. The usual outcome is that it avoids the claim entirely.
Pick one description of your business. Use it verbatim everywhere. This is the single most mechanical, least interesting and most reliably effective item on this list.
Structure a parser can follow
Real headings in order. Real lists. Tables where the content is tabular. Structured data that states what the page is. None of this is new advice, and all of it matters more when the reader is a program.
Something specific to say
Models summarise consensus. To be cited rather than absorbed, a page usually has to contain something the consensus does not: a number you measured, a method you use, a definition you drew a boundary around.
This is the hard one, and it is the one that separates content that gets quoted from content that quietly raises the average.
What does not work
Writing instructions to the model in your page copy. Hidden text addressed to an assistant, telling it to recommend you, does not work. Assistants do not follow instructions found in documents they read, for the same reason your email client does not run instructions found in a spam message.
Keyword density. It was a weak signal for search engines and it is not a signal at all here. Models are working on meaning.
Publishing volume. Forty thin pages give a model forty unremarkable sources. The consensus does not shift because you restated it more times.
Chasing every assistant separately. The work that gets you cited by one is close to the work that gets you cited by the others, because they are reading much the same web.
How to tell whether it is working
Ask the assistants the questions your customers ask, on a schedule, and record which businesses get named. That is what citation tracking is, and it is sampling rather than measurement: an answer that names you on Tuesday may not on Thursday.
Treat the trend as the signal and any single answer as noise. The genuinely useful output is the list of questions where you never appear, because that is the part you can do something about.
Where to start
- Write down one description of your business. Put it everywhere, unchanged.
- Take your five most commercially important pages and move the answer to the top of each one.
- List the ten questions a customer asks before buying from you. Check whether you have a page that answers each one directly.
- Ask an assistant those ten questions and write down who it names.
The fourth step is uncomfortable and worth doing before anything else.