Share of Model and Citation Share: How to Measure Both

Share of model is the percentage of brand mentions in a category that belong to your brand across a fixed set of prompts run on AI assistants, and citation share is the share of source citations that point to your pages. To measure both, run the same prompt set on several assistants on a schedule, count the mentions and cited sources, and track the trend over time.

Updated November 10, 2026. Definitions come from the sources named below, several of which are vendors that sell measurement tools, and results vary by assistant and by session.

In short:

• Share of model counts how often a brand is mentioned. Citation share counts which sources are cited.

• A single run on a single assistant is not a measurement. Use a fixed prompt set, several assistants and a monthly schedule.

• Treat the number as a trend, because answers differ by model and change between runs.

What is share of model?

Share of model is the proportion of brand mentions that a company receives in AI-generated answers, relative to all brand mentions in the same category. The term was introduced by Jack Smyth of the marketing agency Jellyfish, and INSEAD Knowledge later published an article by authors including Jellyfish staff that describes measuring it by prompting language models at scale. It is presented as the AI-era counterpart of share of voice, which measured a brand's presence in advertising or media.

The common formula is simple: mentions of your brand divided by all brand mentions in the category answers, multiplied by 100. The work is in the inputs, which are the prompts, the assistants and the way mentions are counted.

What is citation share?

Citation share is the share of the sources cited in AI answers that belong to a given domain or page. Profound, an AI visibility platform, defines it in its reporting documentation as a share averaged per AI model, which makes it comparable across models. The same documentation defines citation coverage as the fraction of answers that cite a domain or page at least once.

The two metrics answer different questions. A brand can be mentioned often and cited rarely, which means assistants repeat its name but draw their facts from other sites. A brand can also be cited often for a narrow topic and mentioned rarely in broad category answers.

How do share of model and citation share differ?

Share of model measures mentions, while citation share measures sources. The table summarizes the difference and the main limit of each.

MetricWhat it countsQuestion it answersMain limit

Share of modelBrand names in answers, compared with competitorsHow visible is the brand in category answers?Depends on the prompt set and varies by assistant Citation shareCited pages and domains in answersWhich sources shape what assistants say?Not every assistant shows citations in every answer AccuracyFacts in the answer checked against the company's recordsIs the description correct?Needs a person to check each answer

How do you measure share of model and citation share?

You measure both by following six steps and repeating them on a fixed schedule. The method below is Virgo PR's working approach, based on the definitions above.

• Write the prompt set. Collect 25 to 50 questions that buyers, reporters and investors ask about the category, and group them by topic. Keep branded prompts, which name the company, separate from unbranded category prompts.

• Choose the assistants. Use at least three that the audience uses, such as ChatGPT, Gemini, Perplexity or Claude, and record which version and settings were used.

• Run every prompt on every assistant at the same time each month. Save the full answer and the list of cited sources.

• Count the results. For each answer, record which brands were named and which pages were cited.

• Calculate the numbers. Divide your mentions by all brand mentions for share of model, and your cited pages by all cited pages for citation share, for each assistant and overall.

• Check accuracy. Compare facts about your company in each answer with your own records, and list the errors with their sources.

What does a worked example look like?

A worked example shows how the arithmetic fits together. The numbers below are a hypothetical illustration, not data from a real company.

Suppose a software company runs 40 unbranded prompts on three assistants, which produces 120 answers. Its brand is named in 36 of them, so its inclusion rate is 30 percent. Across all 120 answers, competitors are named a combined 144 times. Total brand mentions are the company's 36 plus the competitors' 144, which is 180. Share of model is 36 divided by 180, or 20 percent.

For citation share, suppose the same 120 answers cite 300 pages in total and 30 of them are on the company's own domain. Citation share is 10 percent. The gap between 20 percent and 10 percent suggests that assistants name the company more often than they use its pages as sources, which points to work on its own pages and on third-party coverage.

Why do the numbers differ by assistant?

The numbers differ because each assistant uses different sources and answers can change between runs. An INSEAD analysis of Italy's laundry detergent market, run on Jellyfish's platform, found one brand with nearly 24 percent share of model on Meta's Llama and under 1 percent on Google's Gemini, as summarized by the Young Urban Project. That gap is a reason to report results by assistant and not only as a single blended figure.

Three habits keep the numbers honest. Run each prompt more than once if the sample is small. Report the range, not only the average. And compare each month with the previous month using the same prompts, because a changed prompt set changes the result.

Should you measure by hand or use a tool?

A company can start by hand with a small prompt set, but a tool is more practical at scale. Vendors such as Profound, Semrush and Jellyfish sell platforms that run prompts and tally the results. Those vendors have an interest in the metric, so a team should ask how each tool chooses prompts, which assistants it covers and how it counts a mention. Manual spot checks are useful for checking accuracy and for understanding why an answer says what it says.

What should a monthly report include?

A monthly report should show the prompt set, the assistants, the two shares, the accuracy errors and the changes since last month. It should also list the sources that assistants cited most often for each topic, because those are the places where coverage or corrections will matter. Clicks are an incomplete gauge, since AI answers often resolve a question without a visit. Pair the shares with impressions, AI-assistant referral sessions and the accuracy list.

How do you improve the numbers?

You improve the numbers by changing the sources that assistants draw on. Page-level work, covered in the guide to AEO for PR, makes a company's own pages easy to extract. Earned coverage is a separate workstream, covered in GEO for PR, where Muck Rack's May 2026 study found that earned sources made up most cited links in its sample. When an assistant states something wrong, the correction steps are in the guide to reputation in AI answers. More studies on how assistants describe public companies are in the Virgo PR research library.

The decision to act on: build the prompt set this week, run the first baseline, and set a date for the second run.

Frequently asked questions

What is share of model?

It is the percentage of brand mentions in a category that belong to one brand in AI-generated answers, measured across a fixed set of prompts and assistants.

What is citation share in AI search?

It is the share of cited sources in AI answers that belong to a given domain or page. Profound defines it as an average per AI model.

How do you track brand visibility in AI search?

Run a fixed prompt set on several assistants each month, record mentions and cited sources, calculate the two shares, check the facts and compare with the previous month.

How many prompts do you need?

Start with 25 to 50 questions that reflect real buyer and reporter questions. A larger set gives steadier numbers, and the set should stay the same from month to month.

Is share of model the same as share of voice?

No. Share of voice measures presence in advertising or media. Share of model measures presence in AI-generated answers.

Can anyone guarantee a share of model?

No. Assistants change, and their selection rules are not public. Use the metrics to prioritize work and track progress.