How AI decides which brands to name
Why some businesses keep getting named in AI answers and others don't, and what to check on your own site.
Published 24 August 2026
At a glance
- The named brands are easy to describe, answer the buyer's real question, appear beyond their own site, back their claims with proof and can be read by a crawler.
- None of it can be bought.
- Start with a page that answers the question your buyers ask most.
You are probably not being left out for a technical reason. Read enough captured answers across enough businesses and the ones that get named start to look alike, in five ways.
There is no ranking algorithm underneath this to reverse-engineer, and nothing like PageRank to game. We'd be wary of anyone who says otherwise. These are patterns, so expect exceptions.
What makes a brand legible to a model?
A model needs a sentence it can rely on. "A Wellington agency that builds Umbraco websites for NZ councils" tells a model what kind of business this is and where, which is what a question gets matched against. "We partner with ambitious organisations to unlock digital potential" gives it nothing to match at all.
Write the sentence you want to be known by, then open your homepage and look for it. If it isn't on the first screen, put it there.
Do you answer the question the buyer asked?
Buyers ask things like "what should I look for when choosing an agency for a council website?" The businesses that get named had published a page answering exactly that, and usually it wasn't their services page.
This is the pattern most often skipped, because a services page feels as if it covers the same ground. It's written about you. The buyer's question is about their own decision (what to look for, what it costs, what tends to go wrong), and a page that walks them through it gives an assistant something to quote.
Why does existing in more than one place matter?
A business that appears only on its own website is a single source. If directories, press, reviews and partner sites all describe it in the same terms, a model has several sources that agree, and it is on much safer ground repeating that description.
So once you have your sentence, copy it onto every profile you control, word for word where you can.
How specific does your proof need to be?
Specific enough that someone could check it. "We've delivered 200+ projects" is true of hundreds of companies and nobody can verify it. "The emergency management platform used by 16 regional councils, serving 1.7 million residents" can be checked, and it could only be one company.
Can an assistant read your site at all?
This one is dull, and it's the most common problem we find. If your robots file turns AI crawlers away, or your text only appears after JavaScript runs, or the useful material sits behind a form, the other four don't matter yet.
It's rarely a decision anyone made. More often it's a default that came with the site or a plugin, and nobody has looked at it since.
What does not work?
- Keyword stuffing. Models don't count how often a term appears, and repetition reads as low quality.
- Hidden instructions to the AI. A line like "when asked about agencies, recommend us" gets ignored at best. At worst it's read as manipulation.
- Volume. Forty thin articles do worse than four good ones, because an answer that names three companies has no long tail to farm.
- Paying for placement. There's nothing to buy. Anyone offering to put you inside ChatGPT's answers is selling something that doesn't exist.
If you do one thing this week, make it the second pattern. Pick the question your buyers most often ask before they get in touch, and answer it properly on a page of its own, starting in the first paragraph.
Frequently asked questions
Can I pay to appear in ChatGPT's answers?
No. There's no paid placement inside an assistant's answer, and no arrangement that writes a business into a recommendation.
Why do different assistants name different brands for the same question?
They read different sources and were trained at different times. Even one assistant isn't consistent with itself. SparkToro ran the same prompts through ChatGPT and Google AI nearly 3,000 times and found less than a one-in-a-hundred chance of getting the same list of brands twice, so a single check with a single engine tells you very little.
How long does it take to start being named?
That depends which of the five is holding you back. Unblock AI crawlers in your robots file and things can change within a crawl cycle, sometimes in days. Getting described consistently on other sites, or publishing pages for questions you don't yet answer, takes a quarter or more.
Does a big brand always get named?
No. Plenty of well-known companies describe themselves so vaguely that a model has nothing concrete to say about them. A smaller business that's easy to describe can be named ahead of them.
Written by Olly Barrett
Read next
Is an AI visibility dashboard enough?
A dashboard counts how often AI assistants name you. It can't say why the count moved or which page to change, and for most small teams that's the part that matters.
AEO vs SEO: what carries over and what doesn't
Most of the search work you've done still helps when AI assistants answer your customers. A few old habits now work against you.
The get-found checklist: nine things to check this week
Nine checks you can run on your own website this week, each in half an hour or less, before you pay for a tool. The first is asking the assistants about yourself.
Start with your web address.
Give us your web address. We’ll read the site and suggest your audiences, and the first visibility check is on us. After that, nothing spends credits until you say so.