Share of Model: The New Share of Voice for AI Search
Share of model is how often an AI engine names your brand versus your competitors across a fixed set of buyer questions. It is the natural successor to share of voice, because in an answer that names only two or three brands and has no second page, the question is no longer how loud you are. It is whether the model reaches for you at all.
Why share of voice needed a successor
Share of voice measured your slice of attention: your ads against all ads, your rankings against the results page, your press against the category’s press. It assumed a crowded field where more presence meant more of the pie.
Generative engines break that assumption in two ways. The answer is short, often naming a handful of brands and stopping. And most of those answers are zero-click, so the exposure never appears in your analytics at all. Semrush measured roughly a 93% zero-click rate for Google’s AI Mode on informational queries.
That combination is why traffic is the wrong primary metric here and why a presence metric is needed instead. If the concept of optimizing for AI answers is new, start with what GEO is; share of model is how you keep score once you are doing it.
The two definitions worth keeping
Hold these steady, because changing the definition mid-programme destroys the trend.
Presence share is the percentage of relevant answers that name your brand at all.
presence share = answers naming you ÷ total answersRun 30 prompts and appear in 12, and your presence share is 40%.
Competitive share is your mentions as a percentage of all brand mentions across those same answers.
competitive share = your mentions ÷ all brand mentionsIf the engine names brands 50 times across your prompt set and 10 of those are you, your competitive share is 20%.
Presence share tells you how visible you are. Competitive share tells you how visible you are relative to rivals, which is the number that actually mirrors classic share of voice.
How to calculate it, step by step
- Fix a prompt set. Thirty to fifty real buyer questions covering category, comparison, problem, and recommendation intents. Keep the set stable so cycles compare fairly.
- Run each prompt several times. Answers vary between runs, so three to five runs per prompt is the minimum honest sample.
- Decide the counting rule before you start. Does an unlinked mention count? Does a mention in a list count the same as a recommendation? Write it down.
- Log every result. Brand mentioned yes or no, competitors named, whether a source was cited, and whether the description was accurate.
- Calculate both shares, then segment.
Segment it, or it will mislead you
An average across everything hides exactly where you win and lose. Break the number down at least three ways:
| Segment | Presence share | Competitive share |
|---|---|---|
| Category prompts (“best X tools”) | 45% | 22% |
| Comparison prompts (“X vs Y”) | 30% | 15% |
| Recommendation prompts (“who should I hire”) | 20% | 10% |
The figures above are an illustrative layout, not a claim about any brand.
Segment by engine as well. ChatGPT, Perplexity, Gemini, and Google’s AI surfaces retrieve from different indexes and select differently, so a single blended number can hide a total absence on one engine. ChatGPT’s search leans heavily on Bing, which is why a site missing from Bing can score well everywhere else and near zero there.
What the gaps tell you to do
The value is not the headline figure. It is the diagnosis underneath it.
- Low presence everywhere. The engines do not have enough clear, quotable material to reach for. That is a content and structure gap: see content chunking for AI.
- Decent presence, low competitive share. You appear, but rivals appear more often. Study where they are cited from, which is usually third-party sources rather than their own site.
- Strong on category prompts, weak on recommendation prompts. The engine knows your topic but does not yet trust you as the pick. That is an entity and corroboration problem: see entity SEO and digital PR for AI search.
- Uneven across engines. Check indexation and crawler access per engine before assuming it is a content problem.
- Present but described wrongly. A different problem entirely, with its own fix: when AI gets your brand wrong.
Setting a target without inventing one
There is no industry benchmark for share of model, and nobody can credibly give you one. The category is too new, the prompt sets are bespoke, and counting rules differ between every tool that reports it.
Two honest ways to set a target:
- Relative. Name your three real competitors and aim to close a specific gap against the leader in your own prompt set.
- Directional. Commit to a trend rather than a number, for example three consecutive cycles of improvement in recommendation prompts.
Anyone quoting an industry-average share of model is describing their own sample, not your market.
How to track it over time
Run the full set monthly, log every result, and chart presence and competitive share per engine. Watch direction, not decimals: this is sampling a moving target, so a three-cycle rise matters more than any single reading.
Pair it with the other AI visibility metrics that actually matter, use measuring brand visibility in ChatGPT as your collection method, and see AI citation tracking tools if you want to automate the gathering.
Common mistakes
- Changing the prompt set between cycles, which makes the trend meaningless.
- Single runs per prompt, on a system whose answers vary.
- Blending engines into one number that hides a total absence on one of them.
- Counting any mention as a win, including inaccurate or negative ones.
- Reporting decimals from a sample of thirty answers, which implies precision the method cannot support.
Where share of model fits
Share of model is a framework, not an official platform metric. That is fine, provided you are disciplined: one clear definition, one stable prompt set, one consistent counting rule, tracked long enough to trust the direction. A well-kept directional number is worth far more than a fabricated exact one.
Want your share of model measured against your real competitors and turned into a plan to raise it? Request an AI visibility audit and we will benchmark where you stand across the major engines and show you which gap to close first.
Frequently asked questions
What is share of model?
Share of model is the percentage of relevant AI answers that name your brand, measured across a fixed prompt set and compared with the brands the engine names for the same prompts. It is the generative-search equivalent of share of voice.
How is share of model different from share of voice?
Share of voice measures your presence across ads, media, or search rankings, where many brands can appear at once. Share of model measures how often an AI engine names you in a synthesized answer that may mention only two or three brands in total, with no second page.
How do I calculate share of model?
Run a fixed set of buyer prompts several times each, count how many answers name your brand, and divide by the total answers to get presence share. For competitive share, divide your mentions by all brand mentions across those same answers.
What is a good share of model?
There is no published benchmark, and any vendor quoting one is inventing it. Judge yourself against the competitors named in your own prompt set and against your own trend. Direction over three cycles is far more meaningful than any single figure.
Is share of model an official metric?
No. It is a practical framework, not a number any platform publishes. That makes your method the important part: one clear definition, one stable prompt set, one consistent counting rule.
How often should I measure share of model?
Monthly for the full prompt set, with weekly spot checks on your highest-value prompts. Re-measure after major model updates, since answers can shift noticeably when a new version ships.