Naming Ten Brands Is Not Ten Mentions

TL;DR. An answer that names 10 brands is still 1 answer, so the credit inside it has to add up to one. We weight by position, exclude branded and direct questions, and read unclaimed answers first. If the tracked brands fill the whole chart, the tool erased the opportunity you were paying it to find.

The first version of our market share view was wrong in a way that felt completely reasonable while we were building it. It counted names. Every time a brand appeared in an answer, that brand got a point, and the points became a chart. Then we looked at which answers were producing the most points and realized we had built a machine for celebrating listicles.

Why is an answer that names ten brands still just one answer?

Because the answer is the unit of buyer demand, not the brand name. A response naming 10 brands still resolved 1 question 1 time, so the credit it carries has to add up to one no matter how many names appear inside it.

Watch what happens if you skip that rule. An assistant answers "what tools should I look at" with a tidy list of ten vendors, and every one of them banks a full unit of category evidence. Another assistant answers "which one should I use" with a single sentence naming one company, and that company banks the same single unit. The listicle now counts for ten times as much as the recommendation. That is exactly backwards. From the buyer's side, the list resolved nothing and sent them off to compare, while the recommendation resolved the whole question.

There is a second problem that is worse than the arithmetic. A tool that rewards appearing in long lists will teach its users to chase long lists. You would end up pushing your brand into every "top fifteen tools for X" page on the internet, which is the cheapest content on the internet to get into, and your chart would climb while your pipeline did nothing. Any measurement that is easy to inflate stops being a measurement and becomes a target. We hold the total credit per answer to one so that the only way to gain share is to take it from somebody.

The full view lives on our page about AI market share, and the underlying concept has a glossary entry under share of model.

Why is being named first worth more than being named tenth?

Because position carries information about preference. Readers rarely reach the bottom of a list of 10 options, and neither do the systems that summarize one. So the first name in an answer earns more credit than the last, and the gap between them is not decorative.

Anyone who has worked in search already believes this, they just have not transferred the belief. Nobody argued that the tenth blue link was as valuable as the first. The same intuition applies with more force in AI answers, because an answer is read as a recommendation rather than as a set of options. When an assistant writes "the main ones are A, B and C, though A is the most common choice for teams your size", A did not just appear. A won.

We are not going to publish the exact shape of the curve we use, and we would be suspicious of any vendor who does. Publishing it turns the measurement into a specification for gaming it, and the moment that happens the number stops describing your market and starts describing who read the documentation most carefully. What matters to you as an operator is the direction: moving from the bottom of a list to the top of the same list, in answers you already appear in, is a real gain that shows up in your share without you appearing in a single new answer.

That has a practical consequence for where you spend effort. Founders instinctively chase new coverage, because absence feels like the urgent problem. Very often the cheaper win is the set of answers where you are already named last, because the assistant has already accepted that you belong in the category and is simply not convinced you are the first choice. That is a positioning problem and a corroboration problem, and both are more tractable than being introduced from nothing.

Why do we exclude branded and direct questions?

Because a question with your name in it tells you nothing about category demand. If the buyer already typed your brand, the assistant naming you is not a discovery event. Branded and direct questions are excluded from the share calculation entirely.

This is the single largest source of inflated numbers we see in AI visibility reporting, and it is usually not deliberate. A team sets up a prompt list, and the prompts that come to mind first are the ones about themselves. Is our product any good. How much does our product cost. Our product versus the obvious competitor. Run that list and your share looks tremendous, because you have measured a set of questions that could not have been answered without naming you. Nothing about the market has been observed.

Branded questions are still worth running. They tell you whether assistants describe you accurately, whether they know your current pricing, and whether they repeat a competitor's framing of you back to a buyer who was asking about you specifically. That is reputation work and it matters. It is simply a different measurement, and blending it into share of answers corrupts the one number that was supposed to describe the outside world.

The question types we work with in head-to-head comparison are Discovery, Comparison, Recommendation, Pricing, Reviews and Use Cases, and each of them exists in both a branded and an unbranded form. The share calculation only ever sees the unbranded form. If you want to know whether your category is aware of you, you have to ask questions that do not contain your name.

What are unclaimed answers, and why do we look at them first?

Unclaimed answers are the share of category answers that nobody you track won: no brand was named, or the named brands were neither you nor any tracked competitor. It is the first number we read, because it is the only one describing open market.

Every other figure on the chart tells you how you did against people already in the room. Unclaimed tells you how much of the room is empty. For a young category, or a young company in an old category, it is routinely the largest single slice, and founders find that deflating for about a minute before they realize what it means. An unclaimed answer is a buyer question that no incumbent has locked down. There is nobody to displace. You are not fighting for a mention, you are the first credible candidate for one.

It also changes what you do next. If your share is small and unclaimed is small, you have a competitive problem and you need corroboration, comparison content and reasons for an assistant to prefer you over an established name. If your share is small and unclaimed is large, you have a coverage problem, and the fix is publishing answers to questions nobody has answered well. Those are completely different quarters of work, and the two situations look identical if the only number you have is your own share.

Why should you distrust a share chart where the tracked brands fill the whole board?

Because a category is not a closed set. If the brands you happen to track account for the entire chart, the tool has normalized away everything outside its own list, and the part it erased is exactly the opportunity you were paying it to find.

This is a design decision rather than a bug, and it is a tempting one. Normalizing to the tracked set produces a clean chart where the slices meet neatly at the edge, and clean charts get screenshotted into board decks. The cost is that your share can then only move when a tracked competitor's share moves. A new entrant taking answers away from everybody is invisible. A category where the assistants increasingly refuse to name anyone is invisible. Your own worst outcome, which is answers going to companies you have never heard of, is the one thing the chart is structurally incapable of showing you.

Our view keeps unclaimed as a first-class slice, and it makes the picture uglier. A founder who opens the market share tab expecting a neat competitive split gets a large grey wedge instead, labelled with the share nobody won. We keep it there because the grey wedge is the roadmap. Everything else on that chart is a description of a fight already in progress.

Which views actually answer a strategy question?

Four: by brand, by AI assistant, over time across your last 6 scans, and by question type. Each answers a different question, and reading the wrong one is how teams end up fixing something that was never broken.

• By brand. Who is winning the category right now, and how much of it is still unclaimed. This is the scoreboard and it is the least actionable of the four.
• By AI assistant. Where you are strong and where you are absent. Assistants disagree with each other far more than most teams expect, and a brand can be a default recommendation in one and unknown in another.
• Over time. Your last 6 scans, so a single odd run does not read as a trend. One scan is a snapshot, two is a rumor, six is a direction.
• By question type. Whether you lose at discovery, at comparison, at recommendation, on pricing, on reviews or on use cases. This is where the actual work assignment comes from.

The by-assistant split is the one that changes plans most often. Coverage is not uniform: ChatGPT, Perplexity and Gemini are included on every plan including the free credits, and Claude and Grok are Business-only, which means most teams start with 3 and expand to 5. When those views diverge sharply, the cause is almost always a difference in which sources each assistant leans on, and that is a sourcing problem you can act on rather than a mystery about model behavior.

How is share of AI answers different from mention rate?

Mention rate asks how often you are named across the questions you track. Share of AI answers asks how much of the category's answered demand you captured relative to everyone else. One is a fact about you; the other is a fact about the room.

They move independently and that is the point of keeping both. You can raise your mention rate and lose share, because two competitors rose faster than you did. You can hold mention rate flat and gain share, because a competitor lost ground or an unclaimed slice closed in your favor. Reporting one of those numbers and describing it as the other is the most common form of dishonesty in this category, and it is usually self-inflicted rather than malicious.

Both numbers also rest on something more basic: whether the matcher can recognize a brand when an assistant abbreviates it. If your aliases are registered and your competitors' aliases are not, every relative number you read is flattery. We wrote about how badly that can go in your brand name has an alias problem, where a plain recommendation of us scored zero because of one missing string.

Where should you start reading your own share?

Start with unclaimed, then question type, then position. Those three, in that order, tell you whether the work ahead is coverage, positioning or displacement, and they will disagree with whatever you assumed before you looked.

The habit worth building is to stop asking whether the number went up. Ask which answers changed and why. Share is a summary of a set of specific responses to specific questions, and every one of them is readable. When our own share moves, the first thing we do is open the answers, not the chart, because the chart cannot tell you that an assistant switched from listing you fourth to recommending you outright, and that is the change that actually matters.

If you have never seen your category measured this way, run free discovery first so the question set reflects what your buyers actually ask rather than what your team assumes they ask. Then read the grey wedge before you read your own slice.

Frequently asked questions

Why is an answer that names ten brands still just one answer?

Because the answer is the unit of buyer demand, not the brand name. A response naming 10 brands still resolved 1 question 1 time, so in DataEase AI the credit an answer carries always adds up to one no matter how many names appear inside it. Otherwise a vendor listicle would count for ten times as much as a direct recommendation.

Why is being named first worth more than being named tenth?

Because position carries information about preference. Readers rarely reach the bottom of a list of 10 options, and neither do the downstream systems that summarize one. DataEase AI gives the first name in an answer more credit than the last, so moving up inside answers you already appear in is a real gain rather than a cosmetic one.

Why does DataEase AI exclude branded and direct questions from share of AI answers?

Because a question with your name in it tells you nothing about category demand. If the buyer already typed your brand, the assistant naming you is not a discovery event. Branded and direct questions are excluded from the share calculation entirely, which keeps the number a measure of the market rather than a measure of your prompt list.

What are unclaimed answers?

Unclaimed answers are the share of category answers that nobody you track won: no brand was named, or the brands named were neither you nor any of your tracked competitors. DataEase AI reports it as a first-class slice because it is the only number on the chart that describes open market rather than the fight already in progress.

How is share of AI answers different from mention rate?

Mention rate asks how often you are named across the questions you track. Share of AI answers asks how much of the category's answered demand you captured relative to everyone else. You can raise mention rate and still lose share if competitors rose faster, which is why DataEase AI reports both rather than collapsing them into one figure.