Guides · long read, updated

How to get found by AI assistants: what a page needs before it can be quoted

People ask assistants instead of searching, and assistants answer from pages they can parse. This guide explains how they choose sources, the facts a page has to state before it can be quoted, the tricks that do not work, how to measure whether you are mentioned, and what to do first. The examples are this website and a study we ran on 272 clinic sites.

Updated 2026-09-04 · 2 min read · numbers only with a source

How assistants choose sources

An assistant answering “who builds Webflow sites for construction companies in Poland” does not rank ten links. It looks for pages that state the answer in text it can lift: what the company does, where, for whom, at what price, with a date and an author. Then it prefers pages other sites link to and pages that agree with each other.

01The question is parsed
What, where, for whom, at what price: the assistant looks for pages that state exactly those.
02Candidate pages are read
Not scanned for keywords; read for facts it can quote.
03Facts are lifted
A price, a location, a list of what is included, a comparison with a named alternative.
04Sources are weighed
Pages with a clear author and date, linked by others, consistent with each other.
05An answer is written with a source
The page that stated the fact gets the citation.

The facts a page needs

In August 2026 we measured 272 clinic websites in the US against four things an assistant needs before it can recommend a clinic by name: what it does, where, for whom, at what price, stated as text. Only a few published all four. The list is short; the work is writing the facts down.

Checklist0 of 6

What you do, in one sentence a machine can quote
Where you are and where you serve, as text, not only a map
For whom: the customer named, not “everyone”
What it costs: a price or a starting price, on the page
Who wrote the page and when it was updated
A facts file at the root of the site and structured data on each page

in our clinic study, 6.3% of 272 sites published the first four (HeadPills study of 272 US clinic websites, August 2026)

What does not work

Most of what worked for search engines ten years ago does not work for assistants, and some of it hurts.

Keyword stuffing: assistants read for facts, not for repetitions
Hiding text for machines: pages written for parsers read badly and convert worse
Slogans on the home page: there is nothing to quote
Blocking assistants in robots.txt while hoping to be mentioned
Promising a mention: nobody can, and anyone who does is guessing

How to measure it

There is no public counter of AI mentions. The measurement is asking: a fixed list of your buyers’ questions, run through the major assistants on a schedule, with the answers saved. Whether you are named, whether the facts are right, which page is cited. The first run is the baseline; the report shows the change each month.

Buyers’ questionsAsked monthlyAnswers savedNamed? Correct? Cited?Change vs baseline

the same questions, the same assistants, every month; a measurement, not a promise

What to do first

In order, because each step makes the next one possible.

01Write the facts down
Prices, areas, what is included, who it is for. If they are not written, nothing can quote them.
02Put them on the pages that answer questions
One question per page, the answer in the first paragraph.
03Publish a facts file and structured data
A plain-text file at the root and machine-readable labels on each page.
04Run the baseline
Ask the assistants your buyers’ questions and save the answers.
05Measure monthly and fix what is wrong
Wrong facts and missing pages first; only then more pages.

What we do of this

Our part

This website is built on the rule: every page opens with the answer, a facts file is at /llms.txt, structured data is on each page, and the prices are public. We do the same for clients under the name generative engine optimization, priced after a short call, with a monthly measurement and no promise of mentions.

Generative engine optimization Our research AI consulting and audit

Questions

Questions from readers

Will an AI assistant definitely mention my business if we do all this?

No, and nobody honest can say otherwise. Assistants change how they choose sources without notice. What can be done is removing every reason to skip you and measuring every month whether it worked.

Does a facts file like llms.txt help with AI assistants?

It removes a reason to skip you: a plain list of what the company does and where the facts are, written for assistants the way robots.txt is written for crawlers. It does not ensure a mention; we publish one on every site we build, including this one.

What is the first thing to fix for AI visibility?

Write the facts down: prices, areas, what is included, who it is for. Most sites we measure have nothing an assistant can quote, and no amount of structure fixes a missing fact.

Ask us what the assistants say about you today.

The audit shows it; the price covers changing it and measuring it monthly.

Talk to us See pricing