AI visibility is the new market share

TL;DR. AI visibility is the share of category questions where an AI engine names your brand. In the brands Zaraftis tracks, it moves up six to ten weeks before new business does, closely enough that you can use it as an early warning your funnel software cannot give you. Run a set of 100 to 300 questions every week and track it the way you track market share.

Market share has been the most useful number in marketing for sixty years for one reason: it correlates with everything that matters and is hard to fake. If your category has ten dollars of revenue and you have three of them, you have 30%, and that 30% keeps showing up downstream in your pricing power, your hiring, your moat, and your acquisition multiples. The number is a summary statistic for “how much of the market thinks of you as the answer.”

The problem with market share in 2026 is that the market has stopped thinking and started asking. The thing it asks is an AI engine that gives one confident answer to one question. When a buyer is deciding which two or three companies to look at, the share of those answers that name you is the share that decides whether you make the list. Call it AI visibility.

What is AI visibility?

AI visibility is the percentage of questions in your category where an engine names your brand in its answer. It comes in two forms. The first is how often you show up at all: the share of answers that name you. The second, share of voice, is how often you show up compared to your rivals: your mentions divided by every brand’s mentions in the same answers. The first tells you whether you are present. The second tells you how you stack up against the brands named alongside you.

Written as a formula:

visibility(brand, prompt_set) =
   count(answers that mention brand) /
   count(answers)

share_of_voice(brand) =
   count(brand mentions) /
   count(all brand mentions in the same answers)

Being named in an answer is not the same as being cited as a source for it. A mention is the engine writing your name into the answer it shows the buyer: “the main options are Acme, Beta, and Gamma.” A citation is the engine pulling a page from your own website into the sources it read to write that answer. The two go together less often than you would think, because most of the time the engine names you based on what other sites say about you, not your own pages. Both are worth tracking, but they are different things. This post is about the mention, because the mention is what the buyer actually reads.

The thing that makes AI visibility worth more than a share-of-voice number you might already track in PR or social is the set of questions you measure against. That set is the whole game, and getting it right is most of the work. A weak set gives you a number that feels good and means nothing. A good set gives you a number that tells you what your sales team will hear next quarter. (For why the question, not the keyword, is now the thing you measure, see The end of keyword rankings.)

How do you choose the questions to measure?

The questions you measure against should be the ones your buyers actually type into AI engines, roughly weighted by how often they come up. Three places to find them, in order of how useful they are.

One: ask your sales team. In five minutes they will tell you the questions prospects open with. “We’re comparing you against [X] and [Y], can you tell us how you differ on [Z]?” That same question, with the names taken out, is what buyers now ask AI engines too, often before they ever get on a call. Add the comparison questions that lead to demos, the “best [X] for [situation]” questions in your category, and the “alternative to [competitor]” ones. You can write thirty to sixty of these in an afternoon for any clearly defined B2B category.

Two: borrow from your search data. The questions people have always typed into Google are still the questions they ask. The trick is to rewrite them out of keyword shorthand (“notion alternative product roadmap”) into a real question (“What are the best alternatives to Notion for managing a product roadmap?”). How you phrase it matters more than people think. The same question, asked two ways, can get two different answers.

Three: write down the questions you and your team ask. Marketers are buyers too. Watch how you research a tool when you are about to buy one. The way you phrase things is the way your buyers phrase things, and it is more honest than any survey, which only ever tells you what you wish people asked.

Start with 100 to 300 questions. That is enough to mean something, small enough to run every day without burning through credits, and broad enough to catch the places where you are quietly missing.

What does good AI visibility look like?

These are rough bands, drawn from what Zaraftis has seen across about 200 brands over the last six months.

For a category leader, a healthy share of voice sits in the 35 to 50% range. You will not hit 100, because AI engines deliberately give varied answers, and there are always two or three competitors named alongside you. But you should be the first name on most questions, and you should show up in nearly all of them.

For a strong challenger, a healthy share of voice is 15 to 30%. Buyers see you as a real option, you get named alongside the leader more often than not, and you show up in at least 60% of the questions you track.

For a brand that is in the market but not in the conversation, share of voice is usually 2 to 8%. You exist, you get named now and then, but you are not one of the names the engine reaches for first. Most B2B brands are here and do not know it.

For an invisible brand, share of voice is under 1%. Zaraftis sees this constantly, especially for brands with decent SEO and a real product that have done none of the work to be found by AI. They show up when someone searches Google for them and vanish the moment a buyer asks an AI engine instead. The CEO has no idea, because no dashboard they look at measures the gap.

Why does AI visibility predict new business?

The reason a marketing leader should care is that, across the brands Zaraftis tracks, AI visibility is one of the earliest signs of new business we have been able to measure. Among customers who shared their sales data, visibility went up six to ten weeks before new business did, and the two moved together closely, at a correlation of 0.6 to 0.7. That is about as tight as the link between branded search volume and new business, except visibility moves earlier.

This is not a number that exists in isolation, and it is not the first time an attention metric has been shown to lead demand. The idea predates AI by decades. Les Binet and Peter Field, working from the IPA’s databank of roughly a thousand campaigns, found that a brand whose share of voice runs ahead of its share of market tends to grow, at a rate of about half a market-share point a year for every ten points of excess share of voice. Mark Ritson’s later work on share of search found the same thing: how often people search for your brand by name, a pure measure of attention, runs ahead of market share. AI visibility is the next link in that chain. The place people pay attention changed from television to Google to the AI engine. The rule that attention comes before demand did not.

How this happens is simple. A buyer asks an AI engine a question, the engine names two or three brands, and the buyer remembers one of those names and looks it up directly a few days or weeks later. That later visit shows up in your analytics as direct or branded search traffic. By the time the lead reaches your sales team, the AI answer that started it has left no trace. You see the new business. You do not see what caused it.

AI visibility is the missing trace. It is the early form of the branded interest that turns into new business later. If you are a CMO trying to explain why branded search is up while your paid spend is flat, the cause is often a rise in AI visibility from a couple of months earlier.

The objection: “Mentions are not conversions”

Here is the pushback that comes up almost every time Zaraftis shows someone this number. “Sure, AI names us. So what? They didn’t click. They didn’t fill out a form. We can’t tie a sale back to a mention.”

Part of that is fair, and worth taking seriously instead of brushing off. AI visibility is genuinely hard to tie to revenue, for two reasons. People who hear about you in an AI answer rarely click straight through; they look you up later, and that visit gets filed as direct or branded search, so the trail is lost. And a launch, a funding round, or a wave of interest in your whole category can lift your AI mentions and your sales at the same time, which means the two moving together does not prove one caused the other. Anyone who promises you a clean line from an AI mention to a closed deal is overselling.

But “hard to tie back” is not the same as “not happening.” This is the same argument people made against measuring share of voice in advertising forty years ago, and it was wrong then for the reason it is wrong now: an effect your software cannot see is still an effect. AI visibility is a brand number, not a click number. You measure it because the brands that win it tend to win later, and you check it the way you check any early signal, by watching whether it moves ahead of branded search, direct traffic, and what your sales team says about where leads are coming from. When those line up, you have your answer.

One honest caveat, because there is an easy version of this story that overstates it. Traffic that arrives from an AI answer is not always ready to buy. For expensive B2B software, the people who show up after an AI conversation tend to buy, because the engine has already narrowed their list for them. For cheaper, faster purchases the picture is murkier, and at least one large study of online stores found that traffic from AI converted worse than ordinary search traffic. AI visibility is a sign of brand strength, not a sales channel you can squeeze. Read it as the first and it holds up. Treat it as the second and it will let you down.

How do you start without overcomplicating it?

You do not need a big framework to start. The smallest version that works:

1. Pick your 100 to 300 questions.
2. Run them across the major AI engines every week. Daily is better, but weekly is enough to begin.
3. Track three numbers: how often you show up, your share of voice against rivals, and whether the mentions are positive or negative.
4. Put the trend lines in your monthly marketing review.
5. When a number moves, look at which questions moved and why.

Any tool that watches AI answers can give you the raw numbers, Zaraftis included. The hard part is sticking to it. You need a set of questions you trust, you need to run them on a real schedule, and you need to make the number something your team is actually held to. Until AI visibility is on someone’s goals for the quarter, it will not get the attention it needs.

What this means for strategy

If AI visibility works like market share, the question for any brand becomes: what would it take to raise ours by 10 points over the next two quarters? That question has real answers, and they are the same things that get you cited as a source. You write stronger reference content on the sites engines trust. You clean up your structured data. You build a proper comparison page for every competitor that matters. You make sure your brand is described the same way everywhere on the open web. None of that is exotic, and all of it shows up in a number you can watch, which market share never did. (The specific steps are in the AI-readability checklist; why they work is in How AI engines decide which brands to cite.)

Here is the part that surprises strategy teams. Market share, in most categories, is a number you get once a year, in a private report whose method you cannot fully check. AI visibility is a number you can get every week, measured a way you control, on a dashboard you own. It is the same idea finance leaders already trust, in a form you can actually act on.

If your CEO has not asked about your AI visibility yet, they are about to. The only question is whether you will have an answer.

Frequently asked questions about AI visibility

Q: What is the difference between AI visibility and share of voice?

A: AI visibility, on its own, is how often AI answers in your category name your brand at all. Share of voice is the same data measured against your rivals: your mentions divided by every brand’s mentions in those answers. The first tells you whether you show up. The second tells you how you compare to the brands named next to you. Both are measured against a fixed set of questions, which is what makes the number mean something instead of being for show.

Q: Is being mentioned the same as being cited?

A: No. A mention is the engine naming your brand in its answer. A citation is the engine using a page from your own website as a source for that answer. They often differ, because most mentions come from what other sites say about you, not your own pages. Visibility tracks mentions. Citation share is a separate number that tracks whether your own pages are the ones the engine reads from.

Q: How do I work out AI visibility?

A: Visibility is the number of answers that name your brand divided by the total answers across your questions. Share of voice is your brand’s mentions divided by all brand mentions in the same answers. Run a fixed set of 100 to 300 buyer questions across the major AI engines every week and track both.

Q: What is a good AI visibility number?

A: It depends where you stand. As rough bands from the Zaraftis data, measured as share of voice: a category leader sits at 35 to 50%; a strong challenger at 15 to 30%; a brand that is in the market but not in the conversation at 2 to 8%; an invisible brand under 1%. Most B2B brands are in the 2 to 8% range and do not know it.

Q: Does AI visibility actually drive new business?

A: Among the Zaraftis customers who shared their sales data, visibility went up six to ten weeks before new business did, with the two moving together at a 0.6 to 0.7 correlation. The link is indirect: a buyer asks an AI engine, remembers the brands named, and looks one up later, so the visit shows up as direct or branded search and the original AI answer leaves no trace. It is an early sign of brand strength, not a last-click sales number, and the link can be muddied by anything that lifts both mentions and demand at once.

Q: How many questions do I need?

A: Start with 100 to 300 buyer questions. That is enough to mean something, small enough to run every day without burning credits, and broad enough to surface the questions where you are quietly missing. Pull them from your sales team, from your search data rewritten as real questions, and from the questions your own team asks when picking tools.

Q: How often should I measure it?

A: Weekly at least, daily if you can. AI answers shift as engines update what they read and how they rank it, and a drop can show up overnight. Put the trend lines in your monthly marketing review and tie them to someone’s goals for the quarter, or the number will not get the attention it needs.

How we know

The numbers in this article come from the Zaraftis platform, which runs buyer questions against ChatGPT, Gemini, Perplexity, Claude, Google AI Overviews, AI Mode, Copilot, and Grok every week. The visibility bands (35 to 50% leader, 15 to 30% challenger, 2 to 8% in-market, under 1% invisible) are rough patterns drawn from about 200 brands tracked over the six months ending May 2026. The new-business figures (six to ten week lead, 0.6 to 0.7 correlation) come from the smaller group of those customers who shared their sales data; the link is real and points in one direction, but it is not a controlled study. The point that share of voice runs ahead of market share comes from the published work of Les Binet and Peter Field on the IPA databank, and from Mark Ritson’s work on share of search.