Foundations
How assistants choose who to name
Three stages, and you can influence all three. Most sites only ever think about the first.
When an assistant answers a question that needs current information, roughly three things happen. Understanding them tells you exactly where your website can fall out of the process.
Stage 1 — Retrieval
The assistant turns the question into one or more searches and pulls back a set of candidate pages, usually via a search index. This stage is mostly classic SEO: if you are nowhere in the index for the topic, you are not a candidate.
Where sites fail here: blocked crawlers, no sitemap, thin content, or pages that only exist inside a JavaScript app.
Stage 2 — Grounding
The retrieved pages are read and the relevant passages are extracted. This is where most of the damage happens, and it is the stage almost nobody optimises for.
The model is not reading your page the way a person does. It is looking for a passage that answers the question cleanly enough to reuse. A page that buries the answer in paragraph nine, or splits it across a tab component that never rendered, has been retrieved and then discarded.
Where sites fail here: the answer exists but is not liftable. No question heading, no direct statement, no list, key facts in an image.
Take any page. Can you point at one paragraph that answers a specific question with no other context? If not, that page cannot be grounded on.
Stage 3 — The citation decision
The model has an answer and now decides whether to attach a name to it. This is a risk judgement. Naming a business is a claim, and models are trained to be cautious about claims they cannot support.
Signals that reduce the risk:
- A clear publisher entity — Organization schema, an about page, a real address
- A named author for editorial content
- A date, so freshness can be judged
- Agreement with other sources — your name, address and description matching what appears elsewhere
- Specifics. "From $49 per month, no per-seat fee" is citable. "Affordable pricing" is not.
Where sites fail here: everything is technically fine but nothing is verifiable. Anonymous, undated, vague.
What this means practically
Work the stages in order, because they gate each other.
- Make sure you are retrievable at all — robots, sitemap, server-rendered HTML.
- Make one passage per page cleanly liftable — question heading, direct answer, specifics.
- Make yourself safe to name — entity schema, author, date, contact.
That order is also the order the free checker reports fixes in.
Common questions
Do AI assistants run JavaScript?
Assume not. Some crawlers execute some JavaScript, but many do not, and the ones that do may not wait for slow client-side rendering. If your content only exists after a script runs, treat it as invisible.
Does being cited once mean I will keep being cited?
No. Retrieval happens fresh for most queries, so your position can change as the index changes and as competitors improve their pages.
Take this to your assistant
Paste it into ChatGPT, Copilot, Claude or Gemini and apply it to your own website.
Nothing is sent anywhere. The text is copied to your clipboard.