Your next client will ask an AI about you.

Why is this happening now?

Because research has moved into conversation. The questions that used to be ten searches and a shortlist are now one exchange with an assistant, and the search data shows the shift plainly: the fastest-growing queries in our field are full questions, asked in plain language, exactly as you would ask a colleague. People are no longer hunting for pages. They are asking for answers.

The consequence for a firm like yours, or ours: buyers arrive pre-briefed. By the time someone emails us, an assistant has usually summarised who we are, what we claim and whether anyone corroborates it. The first meeting happened without us in the room. That is not a threat. It is a brief, and you are the one who wrote it, deliberately or not.

What do LLMs actually cite?

Watch what surfaces in assistant answers and a pattern emerges. Specific claims travel: a number, a name, a date, a stated position. Vague excellence does not. Questions answered in the words people actually ask them get quoted; clever headlines get skipped. Consistency compounds: when your facts match everywhere they appear, models treat you as a reliable source, and when they conflict, you dissolve into uncertainty.

Structure matters more than most firms think. Machine-readable facts, structured data, attribution that resolves to a real and checkable identity, a plain summary a model can ingest whole. None of it is exotic. It is housekeeping, done deliberately.

And honesty turns out to be a ranking factor. Models cross-reference. A claim that nothing else supports gets hedged or dropped. The published record you can defend is the only one worth building.

What should a serious organisation do?

Five things, none of them tricks. State your facts once, correctly, and repeat them identically everywhere. Publish judgement, not filler: a position someone could disagree with is citable; padding is not. Answer the real questions in their literal words, on pages a machine can parse. Put your authors’ names and credentials on what they write. And keep private things private: assume anything published will be read, summarised and repeated.

We run our own site this way, from structured data to a plain-text summary for models, and the practices are visible to anyone who looks. This is what we mean by being findable: not gaming the answer, but deserving the citation.

What should you not do?

Do not stuff pages for bots; you will bore the humans who still make the decision. Do not manufacture reviews or plant claims; cross-referencing punishes it. Do not chase each model’s quirks; they change monthly, and clarity survives every change. The reader at the end of every AI answer is still a person deciding whether to trust you. Write for them, structure for the machine, and the two goals stop competing.

Common questions

How do we appear in ChatGPT or Claude answers?

Be the clearest citable source on the questions you want to own: real claims, literal answers, consistent facts, machine-readable structure.

Is LLM optimisation different from SEO?

SEO earns a ranking; LLM visibility earns a citation. Both reward clarity and honesty; LLMs weight consistency and specificity harder.

Can we control what AI says about us?

No. You control the record it reads. Be the best evidence available to the answer. The most durable evidence is work a stranger can inspect without anyone’s permission, which is one reason we give our platforms away.