A growing share of people now ask an assistant instead of searching, and get one answer naming two or three sources. Either you are one of them or you are not in the conversation. Ranking seventh does not help, because there is no seventh.
This is new enough that most of what is written about it is guesswork sold confidently. Here is what is actually known, what follows from it, and what to ignore.
How assistants pick who to name
Simplifying, but usefully: when an assistant needs current information it runs its own searches, retrieves a handful of pages, and writes an answer from what it finds. So two separate things decide whether you appear.
- Retrieval. Your page has to be found and fetched in the first place. That is conventional search visibility, which is why classic SEO has not stopped mattering.
- Extraction. Once fetched, your page has to contain a clean, quotable answer the model can lift with confidence. A page that ranks but buries its answer in marketing language loses to a page that states it plainly.
That second part is where almost all the available advantage sits right now, because so few sites are written for it.
What actually helps
Answer the question in the first two sentences
Put the direct answer immediately under the heading that asks it, then expand. Models extract the first clear statement after a relevant heading. A page that warms up for three paragraphs gets skipped in favour of one that does not.
Write headings as the questions people ask
“How much does a website cost?” beats “Our pricing philosophy”. It matches the query, it signals what follows, and it gives the retrieval step something unambiguous to match against.
State facts in a form that can be checked
Models prefer sources they can corroborate. Specific, verifiable statements — what you do, where you are, what something costs, how long it takes — are more quotable than adjectives. “Leading provider of innovative solutions” is unquotable by design.
Mark up your entities
Schema.org JSON-LD describing your organisation, your services and your FAQs tells a machine unambiguously what you are rather than making it infer. It is cheap, it is invisible to visitors, and most of your competitors have not done it.
Publish an llms.txt
A short plain-text file at the root of your site summarising what you do and pointing to your key pages. An emerging convention rather than a guaranteed one, which is the honest way to describe it — but it costs twenty minutes and ours is at /llms.txt if you want to copy the shape.
Be consistent everywhere else
If your own site, your listings, and the places people discuss you all say the same thing, a model finds agreement and confidence rises. If they contradict each other, it hedges and names somebody clearer.
Be present where people ask
Forums, community sites and question-and-answer threads are retrieved heavily. Honest, disclosed participation where your buyers already ask questions does double duty: those pages rank in search, and they feed the answers.
What does not help
- Hidden text aimed at models. Instructions buried in your page for an AI to obey. It does not work, and it is the kind of thing that gets a site classified as manipulative.
- Mass-generated pages. The same tactic that stopped working for search, now also marking you as low quality to the models.
- Paying to be “added to ChatGPT”. There is no such list and nobody can sell you a place on it. If someone offers this, you have learned something useful about them.
- Blocking the crawlers and hoping anyway. Worth checking your robots.txt: plenty of sites quietly block GPTBot, ClaudeBot and PerplexityBot by default and then wonder why they are never cited. That is a deliberate choice either way — just make it knowingly.
How to measure it
There is no console for this yet, so the practical method is manual and works fine: write down the ten questions a buyer would ask before choosing someone like you. Ask each one in ChatGPT, Claude, Perplexity and Google’s AI results. Record who gets named.
Repeat monthly. You will see movement, you will see which competitors are entrenched, and you will see which of your pages get quoted and which never do. It is unglamorous and it is more honest than any dashboard currently being sold for it.
This is part of every SEO engagement we run, and it is in the free first look: we test which assistants mention you today, and for which questions, before anyone talks about a budget.
The short version
Be findable the conventional way, then be the easiest source on the page to quote accurately. Everything above is a variation on those two sentences, and almost nobody is doing the second one yet.
Written by the people who would also do your work. If any of this applies to you, tell us what you are dealing with — the first look is free and you keep the findings.