You have spent years building the brand. The logo is recognized. The reviews are strong. You rank on page one of Google for the terms that matter. By every traditional scoreboard, you are winning.
Then a customer opens ChatGPT, types “what's the best option for [what you sell],” and your name never comes up.
This is not a glitch. It is a common surprise when companies first measure how they appear inside AI-assistant answers, what we call AI Visibility. A brand can be strong everywhere a marketer normally looks, and still be close to invisible in a place buyers can now look first: the answer an AI assistant gives them.
The reason is simple to state and uncomfortable to absorb: AI assistants cite pages, not brands.
The assumption that quietly breaks
Most teams carry an unspoken equation in their heads: strong brand + good reviews + good SEO = the AI will recommend us.
It feels obviously true. It is not. And you can watch exactly where it breaks if you run the same brand through two different kinds of question.
Ask an AI assistant about your company by name, “is [your brand] any good?” and it will usually do fine. It knows you exist. It can describe you. It will say sensible things. That is brand recognition, and you have earned it.
Now ask the question a real prospect actually asks, the one where they don't yet know your name. “Best [category] for [use case].” “Compare the top options for [problem].” “Who should I look at for [need]?” This is the discovery question, and it is where new customers come from. On these, strong brands can come up short.
We saw this in a published audit of a large spa franchise with strong reviews. Asked about by name, the AI described it well. Asked the discovery question a real consumer types, “best day spa near me,” it scored far lower on AI Visibility. (You can read that full audit, including its method, on the Pondral blog.)
That gap, known by name vs. named on discovery queries, is the problem. So what closes it?
How an AI assistant actually decides who to name
It helps to picture what happens in the second between a person hitting enter and the answer appearing. An AI assistant does not pull a recommendation from a brand-reputation database. For anything current or local, it runs something closer to this:
1. It breaks the question down. “Best [category] for [use case]” becomes a set of sub-questions: what counts as good here, who are the options, what do they offer, what do others say.
2. It gathers candidate sources. It retrieves a shortlist of pages that look relevant. Your brand strength and SEO matter here: they help you make the shortlist.
3. It reads what it gathered. It tries to pull clear, specific facts out of each candidate page. Pages it can actually parse, with clear headings, direct answers, and machine-readable facts, give it something to use. Vague, salesy, template pages give it very little.
4. It weighs the sources. It judges them on things like specificity, accuracy, structure, and whether they truly answer the question.
5. It writes the answer and decides who to name and link. It names and links brands from the sources it judged clearest and most useful.
Notice where brand strength helps: step 2, making the shortlist. Notice where the decision is actually made: steps 3 through 5, where the assistant reads, weighs, and chooses. A strong brand gets you considered. Your pages are what the assistant reads when it decides who to name.
This is why a small local competitor with one excellent, specific page can be named ahead of a national brand with a thin, templated one. The assistant isn't reading your market share. It's looking for the page that best answers this question, and brand size does not settle that on its own.
What the AI Visibility score measures: the five factors, in plain English
So if it isn't brand size, what is it? At Pondral we score AI Visibility on five factors with published weights. You don't need the math to use them: each one maps to a question you can ask about your own brand. (The full, public methodology lives on the Pondral methodology page.)
1. Presence: do you get mentioned at all? (20%)
The most basic question. On a given prompt, across a given assistant, does your name show up, yes or no? Everything else is moot if the answer is no. Presence is binary and it is the foundation: you can't be prominent, well-described, linked, or winning the comparison if you were never in the answer to begin with.
2. Prominence: where do you land in the answer? (25%, the heaviest)
Being mentioned last, in a throwaway “others include” clause, is not the same as being the lead recommendation. Prominence measures how early and how centrally you appear. We weight it highest, more than Presence itself, because the lead name is the one the answer frames as the recommendation. A name buried at the bottom of a long answer is technically present and easy to skip.
3. Context: how does it talk about you? (20%)
Being named is not enough if the assistant damns you with faint praise, or gets the facts wrong. Context asks how well you are described: how you are portrayed (recommended outright, mentioned neutrally in passing, or actively warned against), whether the claims about you are accurate, and whether you are placed in the right category for the question. Presence cannot tell any of those apart. A separate AI grader reads the text around each mention, judges all three, and combines them into a five-point grade from recommended to criticised. A confident, hostile mention can be worse than no mention at all, and so can a glowing one built on wrong facts.
4. Citation Link: does it point a source link back to you? (20%)
Many AI answers cite their sources with links. Citation Link asks the pointed question: when the assistant backs up its answer, is your domain among the links it shows? This is the difference between being described about and being given a link a reader can follow. It also shows which source the assistant chose: it is effectively saying “here's where I got this,” and you want that to be you, not a directory, a review aggregator, or a competitor's page that happens to mention you.
5. Competitive Presence: of everyone named, how much of the conversation is yours? (15%)
AI answers rarely name just one option. Competitive Presence asks: among all the brands mentioned in the same answers, what proportion is you? It reframes visibility as relative, not absolute. You can be present and still be drowned out, named once while a competitor is named, described, and linked three times. This factor is how you see whether you're leading the conversation in your category or merely attending it.
Add those up: Presence 20, Prominence 25, Context 20, Citation Link 20, Competitive Presence 15. You get a single AI Visibility Score out of 100 for any given answer. Average it across the real questions your buyers ask, across the assistants they use, and you have a number you can track.
Why your SEO and brand work still matter (just not the way you think)
None of this means your brand and SEO investment was wasted. Read the five factors again and you'll see where it fits: a recognized brand and authoritative pages help with the first step behind Presence, getting into the candidate set the assistant considers. That is real and necessary.
But it is only the first step. Prominence, Context, Citation Link, and Competitive Presence carry 80 of the 100 points in the published weights, and they depend on what the answer says about you once you are in it. That answer is built from individual pages: how clearly each one answers the specific question, how machine-readable its facts are, how much genuinely useful, specific content it carries versus how much it simply asserts that you're great.
That's the reframe. SEO gets you retrieved. AI visibility decides whether you're chosen after that. They're related, but they are not the same game, and a top SEO position is no guarantee of either prominence or citation in an AI answer.
What this means for you, practically
You don't need to rebuild your site. You need to stop assuming brand strength is doing a job it isn't, and start treating your most important pages as answers rather than brochures. Three questions to take into your next content review:
Does this page answer a real question, in the first paragraph, the way a person would ask it? Not “Relax, Restore, Revive” but “what does this cost, who is it for, how is it different.”
Can a machine read the facts on it? Clear headings, structured data, real specifics (prices, locations, named differences). Assistants can't quote what they can't parse.
Is anything on this page genuinely, verifiably unique to you? Templated copy that could belong to any competitor gives the assistant no reason to pick you over them.
Many brands have never checked this.
The starting move is simply to measure. You can't work on a number you've never seen, and many brands, including strong ones, have never seen this one.
Pondral measures AI Visibility and shows what to work on: how brands appear in AI-assistant answers across ChatGPT, Claude, Gemini, Perplexity, and Grok. You can run a free check on 2 of those engines at pondral.com/demo to see where you stand.
-- Philipp Groubii, Founder, Pondral