Share of model: definition and why it matters for AI visibility

Ask a buyer in 2026 how they shortlist a supplier and a growing share will tell you they asked an assistant first. Something quiet happens to your brand in those conversations: it gets named, or it does not, and you never see the exchange. No click, no referrer, no line in your analytics. Share of model is the metric that puts a number on that invisible moment.

It is the natural heir to a metric every marketer already knows, share of voice, moved onto ground where the rules are different. For the discipline that sits behind all of this, see our agence GEO pillar and the GEO (generative engine optimization) definition.

Share of model: the one-sentence definition

Share of model is a brand's share of voice inside the answers produced by AI engines such as ChatGPT, Perplexity, Claude or Google's AI features. On a fixed set of questions, it measures how often a brand is named or recommended, relative to all the brands named across those answers.

Picture the moment it describes. A prospect asks an assistant which providers to consider for a given need. The engine does not return twenty links; it writes a paragraph and slips in a handful of names. Your share of model is how often your name lands in that paragraph, across all the questions that matter to your business. The higher it is, the more of the mental space the assistant hands back to the reader is yours.

The term is recent because the thing it measures is recent. While search returned lists of links, we talked about position and traffic. With generated answers, the playing field moves to the mention inside the text. Andreessen Horowitz described the same shift in terms of "reference rates," the frequency with which a brand or a page is cited as a source by the models, rather than click-through rates (a16z, 2025). Share of model aggregates those appearances into a single share-of-attention figure.

Share of model vs share of voice vs share of search

Share of voice is your slice of paid and earned presence across media. Share of search is your slice of branded search demand. Share of model is your slice of the brand names an AI writes into its answers. The first two assume a page of competing results; the third assumes a single answer that names only a few brands.

The three are cousins, and confusing them leads to real mistakes. Share of voice and share of search both describe a world where several options sit side by side and the user scans them. Share of model describes a world where the user often reads one answer and stops. That single difference rewrites the mechanics of competition, so it is worth laying the three out plainly.

The consequence is blunt. On a results page, being fourth is still visible. In an AI answer that names only three players, being fourth means being absent. Share of model therefore rewards the concentration of quality far more than the spread of volume. Note too that a naming is not the same as a citation: most namings carry no link at all, which is exactly the distinction our companion pages on the AI citation and the AI brand mention unpack in depth.

Why share of model matters now

When an AI answer appears, far fewer people click the classic links beneath it. Your presence then plays out in whether the answer names you, not in the rank of a link nobody clicks. Tracking position alone becomes a blind spot.

The behaviour change is measured, not assumed. A July 2025 study by the Pew Research Center found that when an AI summary appears in Google, only 8 percent of users then click a link, against 15 percent when no summary is shown, and roughly 26 percent end their session straight after reading it. Clicks thin out, so value migrates to the mention. You can be first on Google and invisible in the answer sitting above it.

Share of model puts a name and a number on that blind spot, and the stakes concentrate on precise questions rather than broad ones. Across the 4887 French keywords Cicero Studio has analyzed, 34% draw fewer than 100 searches a month (Cicero Studio internal data). Those are not scraps. They are the specific, high-intent phrasings a buyer types the moment before choosing, and they are exactly the questions an assistant answers in two sentences naming two brands. Winning one of those two spots is not a traffic play; it is being in the room when the decision is made.

I watched this play out on SME accounts through spring 2026. On a dozen typical commercial questions put to ChatGPT and Perplexity, most answers named two to four brands, almost always the same ones. Companies that ranked perfectly well on Google simply never appeared: their share of model was near zero while their SEO was fine. That gap is precisely what this metric surfaces, and nothing on a rank tracker will show it to you.

How to measure share of model

Define a list of real customer questions for your market, put each one to the main assistants, then count how often your brand appears relative to all the brands named. Repeat on a schedule so you read a trend, not a single frozen number.

The method needs four steps and no sophisticated tooling to begin.

1. List the real questions. Not your dream keywords, but what a prospect actually types when they look for your kind of solution. Fifteen to thirty questions are enough to start.
2. Ask the assistants. Put each question to ChatGPT and Perplexity, then check what Google's AI features return. For each answer, record every brand named.
3. Compute the share. Divide the number of answers that name you by the total number of answers, or your namings by all brand namings. That is your share of model on this basket of questions.
4. Repeat over time. Run it again each month. Because answers vary with phrasing and model, it is the trend across several readings that counts, never a single one.

A worked example makes it concrete. Say you run 20 questions across two assistants, so 40 answers in total. Your brand is named in 6 of them: your answer-level share is 6 ÷ 40 = 15%. Now suppose those 40 answers contain 100 brand namings in all and 9 of them are yours: your mention-level share is 9 ÷ 100 = 9%. Report both, because they answer different questions. The first tells you how often you show up at all; the second tells you how loud you are when the field is crowded.

Alexis Dollé

CEO & Founder of Cicero Studio

I measure share of model the slow way, by hand, one real business question at a time. Across the 1210 SEO and GEO audits Cicero Studio has produced, the recurring gap is almost never a shortage of content; it is content a machine cannot cleanly lift or trust. The habit I would pass on is simple: never trust a share of model from a single reading. Watch the trend, or you are reading noise.

LinkedIn →

How to improve your share of model

Improving your share of model means becoming a source the model wants to reuse. The founding research on the subject, the "GEO: Generative Engine Optimization" paper from teams including researchers at Princeton and the Allen Institute for AI, tested across a benchmark of real queries what makes content more citable. Adding cited statistics and reformulating with the vocabulary of the query lifted visibility in generative answers by up to 40 percent, while keyword stuffing did nothing at all (Aggarwal et al., 2024).

In practice, a few signals carry most of the weight.

• A clear answer at the top of each section. A model lifts a passage that answers in one or two sentences before it develops far more readily than a buried one.
• Named, dated sources. A claim backed by an identified regulator or study gives the model a proof it can reuse. Unsourced, the claim is ignored, or handed to a competitor who did source it.
• First-hand figures. Your numbers, your field results, your comparisons: the things a model cannot invent and a rival cannot copy.
• Clean structured data and a coherent brand entity. Structured data and a consistent identity across the site help the engine understand who you are, which feeds the entity SEO that memory-based namings depend on. Google's own Article structured-data documentation sets out what the markup is for and how engines read it.

This work has a name: GEO, generative engine optimization. It is the discipline that turns adequate content into a citable source, and a weak share of model into steady presence. If your aim is to be recommended by a specific assistant, our guides on the AI Overviews and the E-E-A-T signals underneath go a level deeper.

What share of model does not tell you

The metric earns its keep only if you know where it stops. Treated as a trophy rather than a compass, it misleads.

One honest note to close on. This page reflects what I observe in the field and the public research available when it was last updated, on 18 July 2026. The engines do not publish how they weigh a name, and the memory side in particular is opaque to everyone outside the labs, myself included. I would rather mark the edge of what I know than sell a certainty that does not exist.

Going further

We document the method in the open, because that is the best proof we have. Each resource below takes one angle further. Pick whichever matches the question you are actually holding.

Sources

Every factual claim above is tied to a named, dated source below. The academic work covers what earns reuse, the research institution covers the change in click behaviour, and the vendor documentation covers how answers and structured data behave.