What Is Share of Model?
Also known as SoM, AI share of voice, share of AI voice, LLM share of voice
Share of model is a metric that measures how often a brand appears in AI-generated answers for a defined set of relevant prompts, expressed as a percentage. It adapts the idea of share of voice to large language models, showing how visible a brand is in ChatGPT, Gemini, Claude, Perplexity and similar tools versus competitors.
Share of Model explained
To calculate share of model, you pick a set of prompts that represent how buyers ask about your category, run them across the AI tools you care about, and count the responses that mention your brand. Some teams also calculate it as your mentions divided by all brand mentions in those responses, which shows your slice of the conversation against competitors.
It matters because AI answers often present a short list of options. Being in that list, consistently, is the AI equivalent of ranking on page one. Share of model gives marketing and leadership a single number to track over time and compare with competitors, which is easier to report than dozens of individual prompt results.
Treat the number carefully. Results depend heavily on which prompts you choose, the tools you test, and run-to-run variation in answers. Keep the prompt set fixed, run each prompt more than once, and segment by tool and topic. A rising share of model should line up with better mentions, citations and accuracy, not just one lucky month.
Share of Model formula
Share of model = Responses that mention your brand ÷ Total responses for tracked prompts × 100
- Responses that mention your brand
- AI answers in your tracked set that name your brand or product, counted once per response.
- Total responses for tracked prompts
- All AI answers collected, across every prompt, tool and repeat run in the measurement period.
Worked example: You track 50 prompts in four AI tools and run each twice, giving 400 responses. Your brand appears in 88 of them. Share of model = 88 ÷ 400 × 100 = 22%. Your main competitor appears in 140, which is 35%, so you know where the gap is.
Share of Model example
A CRM vendor tracks 40 prompts like "best CRM for small B2B teams" across ChatGPT, Gemini and Perplexity, three runs each, for 360 responses. In Q1 it appears in 54, a 15% share of model, while the category leader sits at 60%. After publishing comparison pages and earning coverage in three roundups, Q3 tracking shows 90 mentions, a 25% share, with gains concentrated in Perplexity, which retrieves the web for most answers.
How to calculate share of model
Pick representative prompts
Choose 30 to 100 prompts that reflect real buyer questions, including category, problem, comparison and alternatives queries.
Choose tools and runs
Decide which AI tools to test and run each prompt several times per period to smooth out answer variation.
Log every brand mention
Record which brands each response names, along with their position, accuracy and any cited sources.
Calculate the percentage
Divide responses mentioning your brand by total responses and multiply by 100. Do the same for each competitor.
Segment and trend
Break results down by tool, topic and funnel stage, and compare each period against the same fixed prompt set.
Common mistakes
- Changing the prompt set between periods, which makes share of model numbers impossible to compare.
- Choosing prompts that already favor your brand, such as ones that include your product name.
- Running each prompt only once, so random variation looks like real movement.
- Reporting share of model without checking whether the mentions describe your brand accurately or positively.
Share of Model FAQ
There is no universal benchmark. It depends on how crowded your category is and which prompts you track. A more useful target is relative: close the gap with the main competitors in your space and grow steadily over consecutive periods on the same prompt set.
Share of voice measures your brand's portion of advertising, media coverage or social conversation. Share of model measures your portion of mentions inside AI-generated answers. They often move together, since broad coverage feeds what models learn, but they are tracked separately.
Enough to represent your buyers' real questions across topics. Many teams start with 30 to 100 prompts and run each a few times per period. More prompts reduce the effect of any single unusual answer, but keep the set manageable so you can review the results.
For AI tools that search the web before answering, it can shift within weeks as new pages are indexed and cited. For answers based mainly on training data, changes usually wait for new model versions. Expect a mix of fast and slow movement.