What is share of model and how is it measured?

Share of model is the metric that measures how often a brand is mentioned or cited by the answers of ChatGPT, Perplexity, Gemini and Google's AI Overviews, compared to competitors in the same market. The base formula is simple: a brand's mentions divided by total mentions of all relevant brands across a set of questions, multiplied by a hundred. Tools like Semrush refine the calculation by also weighting the position of the mention within the answer, not just its presence; others, like Profound, build a fixed panel of buyer-intent questions (typically 100-200 queries) run on a weekly cadence across multiple engines. The point that separates this metric from GEO: GEO is the discipline that optimizes content to be citable, share of model is the number that tells you whether that optimization is working. It is not a single tool to buy, it is data to collect and govern over time, exactly like organic ranking.

Why the measure is statistical, not a fixed number

The point that almost no article explains well is that generative model answers vary from one run to the next even for the exact same question: token sampling changes, internal routing in mixture-of-experts models changes, even server load at that moment changes. Querying a model once and taking that answer as truth produces a useless data point: the same question asked twice can return a different brand at the top of the answer. That is why a serious methodology requires a stable sample of prompts, repeated on a regular cadence (weekly is the standard for tools like Otterly.ai, which publishes a 0-100 visibility score on recurring scans) and aggregated across multiple runs, not a single query. Anyone treating share of model as a one-time test rather than a time series is looking at statistical noise and mistaking it for a signal.

Mention and citation are not the same thing

The two metrics need to be kept separate. A mention is the brand named in the answer's text, even without a link. A citation is the URL the model chooses as a verifiable source, the one that appears as a clickable reference. They are often misaligned: Profound analyzed 6.8 million citations across 1.6 million responses and found that only 11% of the domains cited by ChatGPT overlap with those cited by Perplexity, and that 80% of AI-cited sources do not appear on the first page of Google's organic results. A brand can be mentioned often and cited rarely, or the other way around: a detailed technical piece can attract citations even when the brand itself stays under-named in the text. Distinguishing the two metrics avoids optimizing for the wrong thing.

An enterprise example

A B2B company selling data-management software builds a panel of 150 realistic questions from its industry ("what's the best data governance platform for a banking group", "alternatives to [product] for data quality"), not questions about its own brand. It queries that panel on ChatGPT, Perplexity and Gemini every week, for months, tracking mentions and citations toward its own content separately from three direct competitors. When a model's grounding shifts its preferred source, or a competitor publishes content that starts getting cited, the trend over time shows it before it turns into lost commercial opportunities; a single hand-run query would never have revealed that.

Why it matters for decision makers

Anyone watching only classic organic ranking measures a channel a growing share of users no longer passes through: the answer comes straight from the generative engine. Share of model gives a number comparable over time, provided it is built on a stable, repeated sample rather than a one-off query. The edge is not buying yet another tool, it is treating this as a data collection and governance problem: the right question panel, repeated with discipline, mentions and citations read together as complementary signals.