Share of Model: The AI-Era Successor to Share of Voice

What is share of model in AI search?

Share of model is the percentage of relevant AI answers in which a brand appears within a defined category, competitor set, prompt universe, model set, and measurement period. The name describes the unit being measured: the model response, rather than the page, impression, keyword ranking, or media mention.

A useful share-of-model measurement starts with a fixed question universe. For a project-management platform, that universe might include category prompts, comparison prompts, has prompts, and use-case prompts that a buyer could ask ChatGPT, Perplexity, Gemini, Copilot, or Google AI Overviews. Each prompt is run under a documented method, and each response is coded for brand presence, position, recommendation context, and citation status.

The metric belongs to Generative Engine Optimization, often called GEO. GEO is the work of making authoritative information available where generative systems can use it in answers. Share of model is the scorecard for whether that work changes the answers people receive.

Share of model should be treated as a measurement framework, not as a claim that one company owns a model or controls its output. AI answers can change as models, retrieval systems, source indexes, locations, and prompt wording change. A credible measurement reports its scope rather than presenting one percentage as a universal truth.

OpenAI introduced ChatGPT search as a way to provide timely answers with links to relevant web sources, while Google documents AI has as part of Search. Those product surfaces make the answer itself a meaningful place to measure visibility, rather than treating a conventional rankings report as the complete picture. OpenAI's announcement and Google Search Central's AI has guidance describe these answer-oriented search experiences.

1. Define the category and intended buyer questions before collecting responses.
2. Choose the AI surfaces included in the score and record their versions or access dates.
3. Code every eligible answer with the same presence and citation rules.
4. Keep prompt wording, locale, and measurement window attached to the result.

How does share of model differ from traditional share of voice?

Traditional share of voice measures a brand's relative visibility across a defined media or search environment. It often uses advertising spend, media mentions, organic keyword visibility, estimated impressions, or social discussion. Share of model measures a different unit: the proportion of AI answers that include a brand for relevant questions.

That distinction matters because an AI answer compresses a research journey into a short response. A buyer may ask for the best software for a specific workflow and receive a small set of named options, supporting sources, and a direct recommendation. There is no page two in an AI answer. A brand can hold strong search rankings yet appear rarely in the answers that frame the category for that buyer.

Search visibility still matters. Search engines and answer systems need discoverable, understandable, current information to surface a brand. But a ranking report cannot tell you whether a model names the brand in a category answer, positions it against competitors, or links to a source that supports the recommendation.

Share of voice can also include channels with no direct relationship to an AI answer, such as paid media or social mentions. That can be appropriate for a communications plan. It is not a substitute for measuring AI-search presence. The two metrics answer different questions and should sit beside each other rather than being forced into one score.

Traditional share of voice asks how visible a brand is in an addressable media or search market. Share of model asks how often a brand is present when a model answers the questions that define purchase consideration. Citation Share adds a stricter layer by asking how often the answer cites the brand or its authoritative content.

1. Share of voice tracks relative visibility across media, search, or conversation.
2. Share of model tracks brand presence inside a defined AI-answer sample.
3. Citation Share tracks cited presence inside relevant AI answers.
4. A complete reporting view can keep all three measures separate.

Which signals belong in a share-of-model score?

A basic share-of-model score should count a response when the brand is named in a relevant answer. The measurement rule must state what counts as a name, including accepted product names, parent brands, and unambiguous variants. It must also define exclusions, such as a brand listed only in an irrelevant sidebar or a response that says the product is unavailable for the requested use case.

The most useful scorecards preserve more detail than one presence flag. Record whether the brand is mentioned, cited, recommended, compared, or placed in a shortlist. These fields let a team see whether it is merely appearing in answers or appearing with evidence that supports a buyer's next step.

Citation status needs its own field. A model can mention a brand without linking to its site, and it can cite a brand's page without making a recommendation. Combining those events creates a number that is hard to interpret. Omnicite uses Citation Share as the percentage of relevant AI answers in a category that cite a brand.

Answer Presence is another useful companion measure. It indicates breadth across the question universe, while Citation Count per day measures volume and Share of Voice measures relative competitor visibility. An optional composite score can be useful for internal reporting, but its component weights should be visible. A composite should never hide a decline in cited presence behind a rise in casual mentions.

The model set also belongs in the data. An answer from ChatGPT search, Perplexity, Gemini, Copilot, or Google AI Overviews is a distinct observation surface. A combined score can be helpful for executive reporting, but each model-level result should remain available for diagnosis.

1. Brand mention: the brand is named in an eligible answer.
2. Brand citation: the answer cites the brand or its content.
3. Recommendation context: the answer presents the brand as suitable for the stated need.
4. Answer position: the brand appears in a shortlist, comparison, or supporting detail.
5. Model surface: the answer source is retained for each observation.

How do you calculate share of model?

Calculate share of model by dividing the number of eligible AI answers that mention your brand by the total number of eligible AI answers collected, then multiplying by 100. The formula is: Share of Model = brand-present answers divided by eligible answers times 100.

An eligible answer is one produced for a prompt that belongs in the agreed measurement universe and meets the collection rules. If a prompt fails, returns an error, or produces a response outside the category because the prompt was malformed, record it separately. Do not silently remove observations after seeing which brands appear.

The denominator determines whether the score means anything. A category score should use questions that reflect category demand. A competitor comparison score should use comparison questions. A local service score should define geography in the prompt. Mixing these intent types without labels can turn a useful benchmark into a vague average.

Use a fixed reporting window and rerun the same prompt set at a documented cadence. This makes changes interpretable. If you add prompts, change country settings, revise brand-matching rules, or add a model, preserve a before-and-after note rather than treating the new series as directly comparable to the old one.

The calculation is simple. The discipline is in the sample design, answer coding, and audit trail. That is why a share-of-model report should retain the prompt, model, run date, response record, presence decision, citation decision, and source links for every counted observation.

1. Step 1: Define the relevant prompt universe and label each prompt by intent.
2. Step 2: Run every prompt across the selected AI surfaces within the stated window.
3. Step 3: Mark each eligible response as brand present or brand absent.
4. Step 4: Divide brand-present responses by all eligible responses.
5. Step 5: Report the result with the model set, prompt count, date range, and coding rules.

What does a worked share-of-model example look like?

A worked example makes the calculation concrete. Suppose a B2B software brand measures 50 eligible AI answers across ChatGPT, Perplexity, Gemini, Copilot, and Google AI Overviews during one monthly reporting window. The prompts cover category selection, comparison, integration needs, and a defined buyer use case.

The brand appears in 20 of those 50 answers. Its share of model is therefore 40 percent: 20 divided by 50, multiplied by 100. This is a methodology example, not a market benchmark or a client result. It shows the calculation only.

The same data can reveal a more useful story. If the brand is cited in 12 of the 50 answers, its Citation Share is 24 percent. If it is named in 20 answers but cited in only 12, the reporting team can investigate which answer types produce mentions without direct citations and whether authoritative supporting content is missing.

A competitor view should use the same denominator. If Competitor A appears in 28 of the same 50 answers, its share of model is 56 percent. The gap is 16 percentage points. That gap is a measurement result for this exact prompt set and window, not a prediction of revenue, rankings, or future citations.

Do not manufacture precision. A score based on 50 observations can be useful for direction and diagnosis, but it should be reported with the prompt count and scope. Expanding the question universe can improve coverage when it reflects real buyer questions rather than padding the denominator.

1. Eligible answers collected: 50.
2. Answers that name the measured brand: 20.
3. Share of model: 20 divided by 50 times 100 = 40 percent.
4. Answers that cite the measured brand: 12.
5. Citation Share: 12 divided by 50 times 100 = 24 percent.

Why should citation status remain separate from brand presence?

Citation status should remain separate because a mention and a citation perform different jobs in an AI answer. A mention shows that the brand entered the answer. A citation shows that the answer attached a source to a claim, recommendation, or supporting detail. Both matter, but they should not be collapsed into one metric.

A brand may be named from broad category knowledge while a competitor receives the cited source link. That pattern can signal a content coverage issue, an authority issue, or a model-specific retrieval difference. The response record gives the team something concrete to inspect instead of guessing from a percentage.

The reverse can also happen. A source may be cited for a narrow factual claim while the brand is not featured as a recommended option. That can be a content success without being a category-positioning success. Separating the fields prevents a team from calling either result a complete win.

AI answer systems commonly expose sources as part of their experience. OpenAI's ChatGPT search announcement describes responses that include links to web sources, and Google Search Central explains that AI has are part of Search's presentation layer. These sources support tracking citations as a distinct observable answer property, not treating them as a proxy for every form of brand presence.

A useful report pairs the two measures: share of model for presence and Citation Share for attributable sourcing. Then it breaks both scores down by question type and AI surface. That structure helps teams decide whether to improve category coverage, comparison content, source quality, freshness, or measurement scope.

1. Use share of model to measure whether a brand enters the answer.
2. Use Citation Share to measure whether the answer cites the brand or its authoritative content.
3. Review gaps by prompt intent rather than relying only on an overall percentage.
4. Inspect source links and response context before deciding what content to publish.

How should teams use share of model without misleading themselves?

Teams should use share of model as a repeatable competitive measurement, with a clear scope and a preserved audit trail. It is most useful when the prompt universe mirrors questions that buyers actually ask and when every competitor is measured against the same answer sample.

Start with a narrow, defensible category. A B2B SaaS team might measure category prompts and direct comparison prompts for a named workflow. A multi-location service business might measure service-plus-city questions. The prompt set should reflect the commercial question, not a pile of loosely related keywords.

Keep model results separate before creating a combined view. One model may cite documentation heavily, while another may produce concise shortlists with fewer visible links. A combined number is useful for leadership only when the underlying model-level results remain available and the weighting method is stated.

Pair measurement with editorial action. Where share of model is low on a category question, examine whether the site has clear, current, authoritative coverage of that question. Where citations are low, inspect whether the content provides direct evidence, comparison detail, and source-backed claims that an answer system can use. This is Citation Engineering: building quality, coverage, and freshness at a scale that helps AI systems trust the source.

Do not promise a specific citation count or ranking from the score. The result is a benchmark for a defined observation set. Its value comes from consistent measurement and better content decisions over time, not from pretending that AI outputs are fixed.

1. Publish the prompt universe and inclusion rules with the report.
2. Preserve the response evidence behind every counted answer.
3. Compare competitors using the same questions, models, and dates.
4. Use low-presence and low-citation gaps to prioritize editorial work.
5. Avoid turning a scoped score into a promise about future AI answers.

Sources

Source: OpenAI

ChatGPT search provides timely answers with links to relevant web sources, supporting answer-level and citation-level observation. OpenAI, 2024-10-31

Source: Google Search Central

Google documents AI has as Search experiences and provides guidance for site owners on appearing in those features. Google Search Central, 2025-05-20

Source: Google Search Central

Google Search's guidance describes how pages and content may be eligible for Search features, supporting the distinction between underlying web visibility and answer presentation. Google Search Central, 2025-05-20