Share of Model: The New Brand Visibility Metric

Search rankings gave marketers a comforting fiction: if your brand sat above the competitor, you owned the moment. Share of model is arriving to ruin that simplicity.

When someone asks ChatGPT which airline is best for a trip to Tokyo, or asks Gemini which cannabis brand has the clearest product information, there is no blue-link leaderboard to inspect. There is an answer. Your brand is either in it, described correctly, and presented as a credible choice, or it is missing while somebody else gets the recommendation.

That change is already moving from theory into media planning. Japan Airlines is working with Jellyfish to understand how advertising affects its visibility and sentiment across AI platforms. The interesting part isn't the airline. It's the measurement problem every serious brand is about to inherit.

The search result is becoming an answer

Traditional search exposed the competition. You could see ten blue links, paid placements, map packs, review stars, and the obvious gaps. AI assistants compress that messy market into a few sentences. They choose which brands to mention, which attributes to attach to each one, and which options feel safe enough to recommend.

That compression changes the job. A brand doesn't only need to be discoverable. It needs to be legible to the model and useful in the exact situations where people ask for help.

The distinction matters because AI systems don't treat every mention equally. A brand can appear often but be framed as expensive, unreliable, or irrelevant. Another can appear less frequently but own the phrase that matters, such as "best for first-time buyers" or "strongest customer support." Counting mentions alone is the easy version of the metric. The strategic version includes position, context, accuracy, and sentiment.

This is where the AI search measurement crisis becomes a brand problem, not just an analytics problem. If an assistant sends no click, your dashboard may never record the discovery. The user may still remember the recommendation, search the brand later, and convert through a channel that receives all the credit.

What share of model should measure

The phrase sounds like a direct replacement for share of voice. It isn't. Share of voice mostly asks how much attention a brand buys or earns inside a defined media environment. Share of model asks how an AI system represents the brand when people pose relevant questions.

A useful score needs at least four layers:

• Mention rate: How often does the model include the brand in a consistent set of category prompts?
• Competitive position: Is the brand the first recommendation, one option among many, or a footnote?
• Narrative accuracy: Are the model's claims about products, pricing, locations, and policies correct?
• Commercial usefulness: Does the answer place the brand in a situation where a real customer could act?

The prompt set matters more than the dashboard. A brand that tests only "What is Brand X?" will get a flattering but useless report. The real questions sound like customers: "Which airline has the best schedule for a short Tokyo trip?" "What should a small dispensary fix before running paid search?" "Which analytics platform is easiest for a lean marketing team?"

Run those prompts across ChatGPT, Gemini, Perplexity, and the systems that matter to the category. Repeat them on a schedule. Save the full responses, not just the position of a brand name. Model outputs drift, and a single screenshot is not a trend.

There is a warning here for anyone tempted to turn this into another vanity score. A model can mention you because it has stale information. It can omit you because the prompt is ambiguous. It can recommend you because of a source that disappears tomorrow. The number is a signal, not a verdict.

The data is messier than a rank tracker

Rank tracking gave marketers the illusion of mechanical precision. Position 3 was worse than position 1, and the change was easy to chart. Model visibility is probabilistic. Ask the same question twice and you may receive a different answer. Change one adjective and the competitive set can shift.

That does not make the metric useless. It makes the methodology part of the strategy.

Teams need a fixed prompt library tied to real buying situations. They need a record of model, date, location, language, account state, and prompt version. They need to separate changes in the model from changes in the brand's public evidence. If a recommendation improves, the team should be able to trace that movement to something concrete, such as better product documentation, stronger third-party reviews, clearer location pages, or a correction published on the brand site.

This is also why generic generative engine optimization advice is wearing thin. The GEO blind spot is treating every AI answer as a ranking surface while ignoring the evidence models use to construct an answer. A brand cannot prompt-engineer its way out of a weak reputation, contradictory product data, or thin first-party information.

The best work looks less like keyword stuffing and more like public-company hygiene. Make the facts easy to verify. Make the offer easy to describe. Make the customer experience match the claims other people repeat about you.

The agency brief is about to change

A search brief traditionally asks for keywords, landing pages, backlinks, and a target position. An AI visibility brief will ask different questions.

Which customer decisions are we trying to influence? Which descriptions do we want the model to get right? Which competitors are appearing in those answers? What evidence would make the recommendation more likely? What happens when the model is wrong about us?

That last question deserves more attention. A bad AI answer is not always a visibility failure. Sometimes the brand appears prominently with an outdated price, an invented policy, or a product it stopped selling last year. More visibility can create more risk if the underlying information is weak.

This connects to the problem of hallucinating marketing agents. The same organizations that want AI systems to recommend, plan, and buy also need to make sure the systems are working from current facts. Brand visibility and brand governance are now tied together.

The practical response is not to hand the whole problem to an SEO vendor. It is to bring brand, content, analytics, customer experience, and legal into the same review. A model's answer is a public expression of what the web seems to believe about the company. Marketing owns part of that expression, but it doesn't own all of the evidence.

A better weekly review

Most teams don't need another platform before they have a disciplined review. Start with a small panel of prompts that represent the category's highest-value decisions. Run them weekly across the two or three models customers actually use. Store the full text and score the results with the same rubric every time.

A useful review asks:

• Did the brand appear when it should have?
• Was it described accurately?
• Was it recommended for the right reason?
• Did a competitor own a better phrase or use case?
• What public evidence could change the answer?

Then connect those findings to business signals without pretending they are perfectly attributable. Compare shifts in model visibility with branded search, direct traffic, assisted conversions, customer survey language, and sales-team notes. If the model starts describing a product as "easy for small teams" and sales calls begin using the same phrase, that is meaningful evidence even if no analytics tool can draw a clean line between the two.

Don't pay someone to deliver a monthly screenshot deck that says your brand was mentioned 42 percent of the time. Demand the prompt set, the raw responses, the scoring rules, the change log, and the recommendation for what to fix next.

The metric is only as good as the brand

Share of model will become a useful phrase because it names a real shift. People are asking machines to narrow choices, and machines are shaping the shortlist before a conventional search ever happens.

But the metric can also become the next layer of marketing theater. Brands will publish colorful dashboards, agencies will promise artificial visibility, and teams will optimize for being mentioned without asking whether the mention helps anyone choose.

The durable advantage won't belong to the company that appears in the most answers. It will belong to the company whose facts are clear, whose reputation travels, and whose recommendation still makes sense when the prompt gets specific.

Search rankings measured where you appeared. Share of model will force marketers to confront what they mean when they say the brand is visible.