What Is Share of Model? A Complete Guide
Share of Model is the percentage of relevant AI-generated answers — across tools like ChatGPT, Gemini, Perplexity, and Claude — in which your brand is mentioned, cited, or recommended. It’s calculated by running a representative set of category questions through AI models and dividing the responses that surface your brand by the total tested — the same idea as share of voice, except the landscape being measured is now a model’s output.
If you’ve spent time in SEO, this will feel familiar and unfamiliar at once: familiar because it’s still fundamentally a visibility metric, unfamiliar because there’s no results page to screenshot and no guarantee the same question asked twice produces the same answer. That unpredictability is why Share of Model needs its own mental model, not a repurposed SEO dashboard.
What Share of Model Actually Measures
Traditional visibility metrics were built around a stable, observable artifact: the search results page. You could track a keyword, watch a position, and know that ranking #1 meant a meaningful share of searchers would see your listing. Share of Model measures something structurally different — not a position on a page, but a probability of inclusion inside a generated answer.
When someone asks ChatGPT “what’s the best CRM for a small agency,” the model isn’t retrieving a ranked list from an index the way Google does. It’s synthesizing a response from training data, retrieved sources, and internal weighting of what it considers a credible answer. Your brand either gets pulled into that synthesis or it doesn’t — Share of Model is the aggregate rate at which it does, across a defined set of prompts that represent your category.
Think of it less like a ranking and more like a batting average: you’re not trying to win one query, you’re trying to understand what fraction of every reasonable way a prospective customer might ask about your category actually surfaces your brand.
Why This Matters Now, Specifically
Search behavior has genuinely split. A growing share of research, comparison, and recommendation queries — the exact queries that used to drive a click to a results page — now terminate inside an AI answer, with the user reading a synthesized response and never clicking through to any website. That’s visible in the referral traffic patterns agencies are already seeing across client accounts.
This creates a real measurement gap. A business can be losing category visibility — quietly, steadily — while its Google rankings look completely unchanged, because rankings only measure what happens on a results page, not what happens when a prospect skips it and asks an AI assistant instead. Share of Model exists to close that gap: it answers what rank tracking can’t — when someone asks an AI model to solve the problem your business solves, are you even in the conversation?
How Share of Model Is Calculated
The math itself is simple, even if the execution takes discipline. Share of Model is:
(Number of category prompts where your brand is mentioned) ÷ (Total number of category prompts tested) × 100
The work is in building the prompt set and running it consistently — a representative list of questions your real buyers actually ask an AI model when researching your space, not just your brand name but category-level questions where you’d want to be considered. For an agency, that might include prompts like “best SEO agency for a mid-size ecommerce brand” or “how do I choose an SEO consultant.”
A workable process looks like this:
- Build the prompt list. Pull from real customer language — sales call transcripts and the questions your team hears at discovery calls translate directly into prompt variants.
- Run each prompt across models. Test the same set in ChatGPT, Gemini, Perplexity, and Claude, since each draws on different data and weighting, and your score will vary by platform.
- Log every mention. Record whether your brand appeared, how it was described, whether it was cited with a link, and where it landed relative to competitors in the same response.
- Repeat on a cadence. Model outputs shift as models are updated and the web changes, so a single snapshot tells you less than a tracked trend.
Because model outputs aren’t deterministic, a credible Share of Model number comes from averaging multiple runs per prompt, not a single test. Treat the first result as a data point, not a verdict.
The Signals That Feed Share of Model
Share of Model isn’t one flat number — it’s built from several signals that each tell you something different about how a model treats your brand.
Mention rate
The most basic signal: does your brand name appear in the response at all, regardless of context. It’s the denominator most people start with, but it’s also the least informative signal on its own.
Citation presence
Some AI tools, particularly Perplexity and Gemini’s AI Overviews, show explicit source links. A direct citation is stronger and more traceable than a bare mention — it tells you which page the model considered credible enough to attribute.
Recommendation strength
There’s a meaningful difference between being one of eight options listed and being named the top recommendation. Track where your brand lands in the response — position within the answer behaves like position within a SERP.
Sentiment and framing
A mention isn’t automatically a good outcome. Models sometimes surface a brand alongside hedging language, outdated information, or an unflattering comparison — the sentence you appear in matters as much as the fact that you appear.
Accuracy
Models occasionally get basic facts about a brand wrong — pricing, service scope, even whether the company still exists. A frequent but inaccurate mention can do more damage than no mention at all.
A complete practice tracks all five signals together. A brand with a high mention rate but weak sentiment and frequent factual errors doesn’t have a visibility problem — it has a correction problem, and the fix is different.
How Share of Model Differs From Rankings
It’s tempting to treat Share of Model as “SEO rankings for AI,” but the differences run deeper than the tool used to measure it.
- No fixed positions. A results page has ten slots and a stable order. An AI answer has no fixed slot count — sometimes it names one brand, sometimes six, sometimes none.
- Non-determinism. Rank position is generally stable day to day. The same prompt run twice can produce different mentions, framing, even different competitors named.
- Aggregate, not single-query. A rank tracker gives a precise number per keyword. Share of Model is only meaningful as an aggregate — a single result tells you almost nothing on its own.
- Shaped by training data, not just live content. Rankings respond to what’s on your site right now. Model outputs are shaped by what was true when the model was trained, so outdated information can persist after you’ve fixed it.
- Attribution is fuzzier. You can point to the exact page that ranks #3. You often can’t point to the exact reason a model chose to mention or omit your brand.
None of this makes Share of Model less real — it makes it a different kind of metric, one that requires sampling and trend-tracking rather than single-point precision.
Where to Start Measuring Share of Model
You don’t need an enterprise platform to get a first, honest read on your Share of Model. Start narrow and manual, then scale once you know which prompts actually matter.
- Write ten to fifteen category prompts that mirror how a real buyer would ask — not “what is [your brand],” but the upstream questions a stranger would ask before they’ve heard of you.
- Run each prompt in at least three tools — ChatGPT, Gemini, and Perplexity cover the bulk of real usage and behave differently enough to be worth comparing.
- Score each response for mention, citation, position, and sentiment in a simple spreadsheet before reaching for a paid tracking tool.
- Compare against named competitors in the same responses — the number only matters in context of who else is winning the same prompts.
- Set a recheck cadence. Monthly is reasonable, since model updates and content changes take time to propagate into outputs.
This is close to how we approach it with clients at Salterra: before recommending any tooling spend, we run a manual prompt audit across a client’s real category questions to establish a baseline. It’s unglamorous work, but it’s the only way to know, in plain terms, whether an AI model would currently recommend that business to a stranger who asked. Treat the number as a diagnostic, not a vanity metric — a low Share of Model usually points back to weak entity signals or content that doesn’t clearly answer buyers’ questions, which is exactly what the rest of this course track helps you fix.
Frequently Asked Questions
Is Share of Model the same as Share of Voice?
Related but not identical. Share of Voice measured your brand's presence across paid media, PR, and social channels relative to competitors. Share of Model applies that same competitive-visibility logic to AI-generated answers, but the mechanics differ — models respond to entity signals and off-site authority rather than ad spend or media placements.
Can I improve my Share of Model without a big content overhaul?
Often, yes, at least initially. Fixing factual inconsistencies about your brand across the web, strengthening entity signals through consistent naming, and earning a handful of credible third-party mentions can move the needle before any large content investment. Bigger gains usually require deeper content, but the first improvements are often cleanup work.
Does Share of Model vary a lot between AI tools?
Yes, often significantly. ChatGPT, Gemini, Perplexity, and Claude draw on different training data, retrieval methods, and source weighting, so a brand can have strong presence in one tool and near-zero presence in another. That's why a credible measurement tests across multiple tools rather than relying on one.
How is Share of Model different from just checking if ChatGPT knows my brand?
Asking a model "have you heard of [my brand]" tests recall of your name, not whether it surfaces you unprompted when a real buyer asks a category question. Share of Model deliberately avoids brand-name prompts in its core measurement — it tests the upstream, non-branded questions where visibility actually competes.
Do small or local businesses need to track Share of Model?
If your customers use AI assistants to research purchases or compare local options — increasingly common even for local services — then yes, at a smaller scale. A handful of manually run, category-relevant prompts checked periodically is enough to establish whether AI models recommend you within your service area.
What's a realistic first goal for Share of Model?
Don't start by targeting a percentage — start by establishing your baseline and confirming the mentions you do get are accurate and positively framed. Many businesses discover their first priority isn't mention frequency at all, but correcting outdated or wrong information a model is already surfacing about them.
Terry has 30+ years in software and SEO. He’s the founder of Salterra Digital Services and SEO Spring Training, host of the Roundtable SEO Mastermind, and lead instructor at SEO University — teaching the exact tactics his team uses on client work.
Continue learning
Ready to master this?
This guide is one lesson from the Measuring AI Visibility Share of Model course. Get every lesson, framework and checklist — plus the full 38-course catalog — inside SEO University.
Practitioner-focused training across the full digital marketing stack — from technical SEO to conversion optimization and the AI search era. By Salterra Digital Services, since 2011.