Last updated August 2026
Your brand ranks number one on Google. You appear at the top of every organic SERP for your category keywords. You have thousands of backlinks. None of that guarantees you show up in ChatGPT or Perplexity when a buyer asks which tool to use.
That is the core problem share of model solves. And the evidence is stark: according to an Ahrefs study of 15,000 queries (August 2025), only about 12% of URLs cited by AI assistants (ChatGPT, Gemini, Microsoft Copilot, and Perplexity combined) also appear in Google”s top 10 results for the same prompt. The figure drops to roughly 8% for the non-Perplexity assistants. AI engines and search engines draw from almost entirely different source pools.
If you are measuring your brand only through organic search rankings, you are measuring the wrong channel.
The share of model formula
The calculation is simple. The hard part is doing it at scale across engines.
Share of model = (your brand”s mentions across N prompt runs) / (total brand mentions across those same N runs) x 100
Run it per engine, per prompt cluster, per time period. Then track the delta week over week.
A concrete example: you run 100 prompts against buyer-intent queries in your category on ChatGPT. Your brand is mentioned in 28 of those 100 responses. Competitors are mentioned across those same responses a combined total of 120 times (including your 28). Your share of model on ChatGPT for that prompt cluster is 28 / (28 + 92) = 23.3%.
Why share of model is decoupled from organic search
Most marketing teams assume that what ranks in Google will appear in AI answers. The data says otherwise.
According to an Ahrefs study of 15,000 queries (published August 2025 by Louise Linehan and Xibeijia Guan), an average of only 12% of URLs cited by AI assistants also rank in Google”s top 10 for the same query. Perplexity is the outlier with roughly 29% overlap. ChatGPT and Microsoft Copilot sit around 8%.
The practical implication: 88% to 92% of what AI engines cite when answering a buyer”s question does not come from Google”s top results. AI engines weight training data, retrieval sources, third-party citations, reviews, editorial coverage, and structured content on your own site. They do not weight your Google ranking position.
This makes share of model a distinct metric that requires distinct measurement, not an extension of your existing SEO reporting.
The zero-sum framing
Share of model is zero-sum within a prompt. When AI engines answer a category question, there are a finite number of brands they can mention in a given response. Every time a competitor is mentioned and you are not, that is share you do not own.
This is why the metric compounds. A brand that builds from 12% to 30% share of model over six months has not just gained 18 points: it has taken those points from competitors, who now appear less often to buyers running the same queries. The AI channel operates more like earned media than search, but the competitive dynamics are as zero-sum as advertising share.
How to structure your share of model measurement
Step 1: Define your prompt set
Start with 30 to 50 prompts that reflect how real buyers in your category actually ask AI engines for help. Use:
- Evaluation queries (“what are the best tools for X”)
- Comparison queries (“compare X and Y”)
- Problem-first queries (“how do I solve X problem”)
- Recommendation queries (“which tool should I use for X if I have Y budget”)
Do not use navigational queries or brand-name lookups. Those do not measure competitive share; they measure whether you exist. See /glossary for definitions of prompt family and prompt cluster.
Step 2: Choose your engines
Track at minimum ChatGPT and Perplexity. Add Google AI Overviews, Gemini, and Microsoft Copilot for full coverage.
Engine selection matters for a specific reason: the citation pools are almost completely non-overlapping. According to Profound”s analysis of 100,000 prompts run across ChatGPT and Perplexity (published July 2025), only about 11% of cited domains appear on both platforms. A brand that dominates ChatGPT can have near-zero share on Perplexity for the same prompts. Tracking only one engine misses this.
Step 3: Run each prompt multiple times
Five runs per prompt is the minimum. For high-entropy prompts (broad category questions, open-ended comparisons), run 10 times. Average the results. The variance across runs tells you as much as the average: a prompt where you appear in 2 of 5 runs has very different implications than one where you appear in 5 of 5.
Step 4: Record mentions, not just presence
Presence tracking (did your brand appear: yes or no) is a blunt instrument. Record which brands are mentioned in each response, how many times each brand appears, what context the mention appears in (named as a leader, listed with caveats, recommended for a specific use case), and whether your domain is cited as a source.
This produces a mention share calculation, not just a presence flag. And it surfaces qualitative signal: being mentioned as “one option” is different from being the first recommendation.
Step 5: Calculate share per engine and aggregate
Calculate your share per engine on each prompt cluster. Then aggregate across engines using a weighted or unweighted average depending on which engines your buyers actually use.
| Engine | Your mentions | Total mentions | Your share |
|---|---|---|---|
| ChatGPT | 28 | 120 | 23.3% |
| Perplexity | 41 | 135 | 30.4% |
| Google AI Overviews | 19 | 110 | 17.3% |
| Gemini | 33 | 128 | 25.8% |
| Blended average | 24.2% |
The blended figure is your overall share of model. The per-engine breakdown tells you where the gaps are.
Step 6: Track week over week
A single share-of-model snapshot is an interesting data point. A weekly trendline on the same prompt cluster is an actionable signal. Direction of change matters more than absolute level. A brand moving from 15% to 28% over a quarter is improving. A brand holding at 30% for six weeks while a competitor moves from 18% to 35% has a problem even though its absolute number stayed flat.
Tools that measure share of model
Manual measurement at the scale you need (30 to 50 prompts, five runs each, across four or five engines) is around 600 to 1,250 prompt executions per week. Most teams automate this.
Profound is the specialist pick for enterprise teams that need the deepest citation attribution alongside share-of-model tracking. Its Prompt Volumes feature surfaces real buyer demand signals, not just a manually curated prompt list. Pricing starts at $99/mo for ChatGPT-only access (Starter) and $399/mo for full multi-engine coverage (Growth). The effective entry for a real AI visibility programme is the Growth tier.
Otterly.AI tracks prompt-level citation and share-of-AI-voice benchmarking across six platforms, with a public API for workflow automation. It holds G2 High Performer (Answer Engine Optimization, Winter 2026) and Gartner Cool Vendor 2025 status. Entry at $29/mo (Lite, monitoring only); competitive benchmarking requires Standard at $189/mo.
Semrush has added AI visibility tracking to its platform, pulling from its existing search keyword infrastructure to surface share-of-mention data. Useful for teams that want AI share of model inside the same tool they already use for traditional SEO metrics, though the AI-specific depth is shallower than dedicated platforms.
Scrunch AI covers up to nine platforms at the Enterprise tier and includes SOC 2 Type II compliance. Core plan ($250/mo) covers four engines and 125 prompts. Best suited for enterprise brands with security procurement requirements.
Temso is an all-in-one AI SEO platform that tracks share of voice, brand mentions, citations, and sentiment across eight engines from $89/mo. It is the most accessible option for teams that want share-of-model data and an execution plan in the same subscription. Setup takes around five minutes. It does not include traditional backlink or keyword rank tracking, so teams that need both AI and conventional SEO data will want a separate SEO tool alongside it.
The full comparison of platforms that measure share of model is at /rankings/ai-visibility-tools. The methodology used to score them is at /methodology.
What share of model does not tell you
Share of model answers “how often am I mentioned.” It does not answer:
- Whether the mention is positive, neutral, or negative (sentiment)
- Whether the facts stated about your brand are accurate (accuracy)
- Whether your website is cited as a source alongside the mention (citation rate)
- Whether buyers who see those mentions convert (downstream pipeline)
A complete AI visibility measurement programme tracks all four alongside share of model. Share of model is the headline number; the others are the diagnostic layer underneath it.
The case for starting now
Share of model compounds. Brands that build AI citation presence now will be harder to displace later: AI engines tend to reinforce sources they have already learned to trust. The brands that are invisible in AI answers today are not making a neutral choice; they are ceding ground in a channel where the gap is already widening.
The formula is simple. The execution requires consistency. Start with 30 prompts, five runs each, two engines, and a weekly tracking cadence. Expand from there.
Ready to see your current share of model across ChatGPT, Perplexity, Gemini, and five other engines? Temso runs the measurement and converts the gaps into a prioritised fix plan, starting at $89/mo with no credit card required for the trial.