SEO Galaxy is now GetMentioned. The platform stays the same. Want a demo? Book your slot here →
Share of Model tracks a brand's share of all brand mentions in AI answers. Why the formula is uncontested but the denominator is not, and why two vendors can report different values for the same brand on the same day.
David Hahn · August 23, 2026 · 10 min read
The short version
Share of Model measures how present a brand is in the answers of language models. It counts your brand against all brands named across a defined set of AI answers.
The abbreviation is SoM. The metric answers one question. How large is your share of the recommendations a model gives out in a category?
It says nothing about revenue. A whole customer journey sits between a mention and a purchase.
The calculation is short. Brand mentions divided by all brand mentions, multiplied by one hundred.
An example with invented figures. A prompt set holds 250 questions, and the answers return 800 brand mentions.
144 of them name your own brand. Divided by 800 that gives 0.18. The Share of Model reads 18 per cent.
Three quantities decide the result. The number of prompts, the choice of prompts, and the models queried.
Mention frequency counts in absolute terms. It climbs whenever the prompt set grows, even with nothing shifting against competitors.
Share of Model puts the same mentions into a ratio. It holds steady when every brand gains at the same rate.
Citations are a different event, the source link under an answer. A brand can be named while its website never appears there.
The term was coined by an agency and carries a trademark. Jellyfish launched the Share of Model platform on 5 December 2024.
The announcement calls it an interface for analysing how large language models perceive brands and products. That is a product description with no formula in it.
An earlier piece from the same agency is signed by Jack Smyth and dated 20 March 2024. Its body text says "AI Brand Awareness" throughout.
Share of Model appears there only in the meta title and the URL path. Tom Roach also works for Jellyfish and popularised the term, and his original posts return HTTP 403.
Authorship therefore stays open. No second vendor we checked uses the word, and the others measure something similar under names of their own.
Share of Model carries an old marketing idea onto a new surface. Beside its two predecessors, the point where the transfer stops becomes visible.
| Metric | Denominator | Data source | Moved by |
|---|---|---|---|
| Share of Voice | category media spend | third parties, external | buying media |
| Share of Search | category search demand | Google Trends, public | demand for the brand |
| Share of Model | a self-chosen prompt set | queries to AI platforms | wording of the prompt set |
Nielsen defines Share of Voice as a brand's media spend as a percentage of all media spend in the category, market and channel at that point in time.
The denominator is identical for every competitor. Third parties gather it, and media buying moves it.
Share of Search is the brand's share of all category search queries. Les Binet presented it at the IPA in 2020.
The data source is Google Trends, public and free back to 2004. For the car market the IPA describes a lead of up to a year over market share.
Here the analogy ends. For AI systems no external population exists that anyone could look up.
The denominator is the prompt set. Change the questions and the result changes with them.
Media spend can be bought, which moves the Share of Voice denominator. A prompt set can only be defined.
Every vendor follows the same pattern. Prompts go to the AI models repeatedly, the answers are scanned for brand names, the hits are counted. The differences sit in the detail.
Profound discloses the most. Its index methodology draws prompts from over 400 million licensed real conversations, and measurement covers ChatGPT and the US market only.
Evertune states its sampling. Every prompt runs one hundred times per model across eleven models and more. The metric is called Share of Answer there.
Peec.ai publishes the formula and names the source of the questions. Customers add their own prompts, and the metric is called Share of Voice there.
Rankscale asks for a representative set of prompts. How that set is arrived at does not appear in the documentation.
Sampling details exist in part. Formulas exist too. A published prompt set exists at none of the vendors we checked.
No figure is reproducible on that basis. Two vendors can report entirely different percentages for one brand on the same day, and neither is calculating wrongly.
Model dependency adds to that. A ChatGPT-only value describes something other than an average across eleven models.
I read Share of Model as a useful number and as a poor basis for a contract, for as long as nobody shows me their prompt set.
The number holds as a time series inside one vendor. Three conditions make it readable:
Change one of the three and the series starts over. How such a set gets built and documented sits in prompt tracking.
Comparing two vendor values does not hold. Two numbers side by side are two different lists of questions.
Two layers decide whether a brand turns up in an answer. They can be steered to different degrees.
Training data is the body of text a model learned from. It is frozen at the time of training.
Retrieval at runtime is the second layer. A system fetches current search results and builds the answer from them.
Google states that its generative features are rooted in the core search ranking systems. OpenAI documents only that its search bot surfaces websites in ChatGPT. Perplexity documents domain filters.
The mechanics support it. A system assembling an answer from several sources finds more to go on the more independent pages report the same thing.
Ahrefs has measured that relationship. The analysis of brand mentions and AI visibility covers 75,000 brands and uses Spearman correlations.
The values run from 0.656 for AI Overviews to 0.709 for AI Mode, with ChatGPT at 0.664.
Author Louise Linehan states in the same piece that correlation is not causation. No controlled experiment weighs one extra placed mention against one extra AI mention.
This article therefore stays with correlates with. Anyone promising a guaranteed Share of Model has no evidence behind it.
Google warns in its own optimisation guide for generative search against chasing inauthentic mentions. The section calls the practice less helpful than it seems.
Meant are mentions without an editorial reason. A forum post that drops a brand name and adds nothing else is one.
An editorially justified placement sits in an article that works without the brand. It fits the publication by topic and hands the reader a fact.
A comment under somebody else's article reading "we offer that too" is an inauthentic mention.
The better version is a trade piece on price ranges that lists your own range as one line among several and dates the period it covers.
Share of Model adds to the numbers already in use. It replaces none of them.
| Question | Metric that answers it |
|---|---|
| How visible is the domain in Google? | Visibility index |
| How often do AI answers name the brand? | Mention frequency |
| How large is your share of all mentions? | Share of Model |
| Is your own website cited as a source? | Citations |
The visibility index measures rankings in the search engines. It leaves AI Overviews out, and an AI metric closes that gap.
What the current state means for a brand is set out in LLM visibility. The measures that follow sit in AI visibility.
In our experience most misunderstandings arise at four points.
Each has a sound reading. Compare against your own previous value at the same vendor, record every change to the prompt set with a date, and check a drop against the number of brands measured.
What can be steered is the material an answer gets built from. That material is content on the web in which the brand appears alongside its topics.
The answer itself is not bookable. AI platforms work probabilistically, and one question can draw two different answers across two runs.
For marketing teams three things follow. Your own facts have to be findable. The brand has to appear in articles that describe a category, and those appearances have to recur.
GetMentioned is a link building agency with its own placement marketplace. Full service in link building, not full-service SEO. With us you book brand mentions and backlinks at publishers whose traffic, SEO data and price you see before booking.
Tracking your mentions in individual answer engines is not part of our product. We supply the placements an answer can draw on.
Share of Model answers a question worth asking. How large is your share of the recommendations language models give out in a category?
The formula is uncontested. The denominator is not, because no vendor shows its prompt set.
As a time series at one vendor the number carries. As a comparison between two vendors it does not.
Share of Model is a brand's share of all brand mentions in AI answers. The calculation is your own mentions divided by all mentions, times one hundred.
A prompt set goes repeatedly to one or more language models, and brand mentions in the answers get counted. Your share of that total gives the value in per cent.
The term comes out of the Jellyfish orbit. Jack Smyth published on it in March 2024 and the platform of that name launched in December 2024. No verbatim first definition appears in the reachable texts.
Both measure a share within a category. Share of Voice refers to media spend with a denominator gathered externally, while Share of Model refers to a self-chosen prompt set.
Every vendor uses its own prompt set, models and market, and none publishes its questions. Two values for one brand are two separate measurements.
No benchmark across industries has been established. What carries meaning is the movement of your own value at one vendor, with prompt set and model selection unchanged.
The data shows a correlation between mentions on the web and visibility in AI answers, around 0.65 to 0.71. A causal proof from a controlled experiment is missing.
About the author
Managing Director, GetMentioned
David has been building link acquisition and digital PR processes since 2016, first as an agency under SEO Galaxy, today as a platform with GetMentioned. He has scaled his own projects from zero to seven-figure monthly traffic and delivered thousands of campaigns for clients. Here he writes about what works in practice, and about what only costs budget.
Compare publishers, SEO data and prices in one place and book the placements that fit.
The same question returns different answers at the same moment. What a measurement setup that survives this looks like, and why numbers from two tools cannot be compared.
August 15, 2026 · 15 min read
AI answers recommend providers before anyone clicks a ranking. How to measure AI visibility, which signals carry it, and where the external evidence models lean on comes from.
August 10, 2026 · 12 min read
A brand surfaces either from model weights or from a web search at runtime. A half-hour test separates the two, and that decides whether the work takes days or months.
August 16, 2026 · 12 min read
