For twenty years, SaaS marketers reported one number above all others: rank position. Where do we sit on Google for our money keywords? That number is losing its grip. When a buyer asks ChatGPT, Gemini or Perplexity to recommend a sales-automation platform, there is no page one. There is a single synthesised answer, a handful of citations, and a shortlist of brands the model chose to mention. Either you are in that answer or you are invisible.

A new metric has emerged to describe this reality, and it is rapidly becoming the headline number in AI-search reporting for SaaS brands.

Share of model is the percentage of AI-generated answers, across a defined set of prompts and engines, in which a brand is mentioned, cited or recommended — and how favourably. It replaces rank position as the core visibility metric for AI search, where answers are synthesised rather than listed. Because AI answers vary between identical runs, share of model is measured by sampling: running a fixed prompt panel repeatedly and averaging the results.

Where the term comes from

Unlike “bounce rate” or “domain authority”, share of model does not have a single agreed definition — and it is worth being honest about its origins rather than pretending otherwise.

The clearest verifiable origin is Jellyfish, the digital agency housed within The Brandtech Group, which launched a trademarked Share of Model™ platform in 2024. As AdExchanger reported, the platform surveys major large language models — ChatGPT, Claude, Llama and Gemini — daily with questions about brands, products and categories, then analyses the responses for mention patterns and sentiment. Jellyfish’s Jack Smyth framed the goal neatly: the point is not to extract accurate information from the model, but to help a marketer understand what the model might be saying to customers.

Around the same idea, AI-visibility platforms such as Profound, Peec AI and Otterly.AI, along with SEO incumbents like Semrush, popularised closely related terms: “AI share of voice”, “share of answer” and “AI visibility score”. The marketing press has picked up “share of model” as a convenient umbrella label for all of them. Whatever the label, the underlying question is the same: of all the times an AI engine answers a buying question in your category, how often are you part of the answer?

Why rank position doesn’t transfer to AI answers

It is tempting to treat AI search as a new SERP and ask “where do we rank in ChatGPT?” The question doesn’t map, for three structural reasons.

• Answers are synthesised, not listed. A classic results page shows ten or more links and lets the user choose. An AI engine composes one answer and makes the choices itself. Profound’s analysis of roughly 700,000 US ChatGPT conversations (October–December 2025) found the assistant attaches only around six unique citations per cited conversation — a brutally short shortlist compared with a results page.
• Cited sources are few and unevenly distributed. The same Profound dataset found Wikipedia alone appearing in about 18% of conversations that carry citations, with the overall distribution of cited domains highly unequal. We unpack what that concentration means for SaaS content strategy in our post on AI citation source concentration.
• Most AI answers end without a click. When the engine has already summarised the comparison, many users never visit anyone’s website. Visibility inside the answer is the impression; the click is optional. That is the dynamic we covered in zero-click AI search and brand visibility — and it is why a metric based on answer inclusion, not link position, is the right unit of account.

There is a fourth reason, and it changes how measurement itself works: AI answers are non-deterministic. Ask the same engine the same question five times and you will get five overlapping but different answers, sometimes with different brands in them. Rank tracking assumes a stable object to measure. Share of model assumes a distribution, and samples it.

Classic SEO metrics and their AI-search equivalents

Classic SEO metric	AI-search equivalent	What changes
Rank position	Share of model / brand mention rate	No ordered list to rank in; you measure the percentage of sampled answers that include the brand at all
Click-through rate (CTR)	Citation rate	Success is being one of the few sources the answer links, not winning a click from a list of ten
Keyword rankings report	Prompt-panel coverage	You track a curated set of buyer questions (prompts), not keyword strings, across multiple engines
SERP feature ownership (snippets)	Recommendation rate	The strongest outcome is the model naming you as its pick, not occupying a visual slot
Monthly rank check	Repeated sampled runs	One-off checks are meaningless under non-determinism; you need N repetitions and an average
Backlink profile	Citation-source footprint	What matters is presence on the third-party pages engines actually cite, not raw link volume

How share of model is measured in practice

Every serious measurement approach — vendor or in-house — shares the same skeleton.

• A fixed prompt panel. A curated set of questions real buyers ask in your category (“best AI sales agent for B2B”, “alternatives to X for outbound”, “does AI cold calling work”). The panel stays constant so results are comparable over time.
• Multiple engines. ChatGPT, Gemini, Perplexity, Copilot and Google’s AI answers behave very differently, cite different sources, and must be tracked separately.
• Repetition and sampling. Because identical prompts return varying answers, each prompt is run multiple times per period and results are averaged. Vendors typically re-run panels daily.
• Graded outcomes. Good measurement distinguishes at least three tiers: a mention (the brand is named in the answer), a citation (the brand’s own page is linked as a source), and a recommendation (the model presents the brand as its suggested choice). A citation of your domain and a mention sourced from someone else’s comparison page are very different assets.
• Competitive denominator. Share of model is a share: your mentions divided by all brand mentions in the sampled answers for your category.

Tools that track it (verified as of August 2026)

• Profound — enterprise AI-visibility platform tracking mentions, citations and sentiment across ChatGPT, Perplexity, Claude, Gemini, Grok, Microsoft Copilot, DeepSeek and Google AI Overviews.
• Peec AI — tracks visibility, position and sentiment across ChatGPT, Perplexity and Gemini against custom prompt panels.
• Otterly.AI — monitors brand mentions and citations across ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Copilot, Perplexity and Claude, reporting share of voice as your percentage of citations versus competitors.
• Semrush AI Visibility Toolkit — brings AI share of voice into the classic SEO suite, with a Brand Performance report weighting both mention frequency and prominence.
• Jellyfish Share of Model™ — the agency-side original, surveying LLMs daily for brand perception and recommendation patterns.

Whether these numbers translate into pipeline is a separate question — one you answer in your analytics, which is why we recommend pairing any share-of-model tracking with proper AI referral measurement in GA4.

The limitations nobody should gloss over

Share of model is a genuinely useful metric, but it is young, and its failure modes are real.

• Volatility between identical runs. SE Ranking ran the same 10,000 queries through Google’s AI Mode three times on the same day (20 June 2025) and found the average overlap of exact cited URLs across the three runs was just 9.2%, with over a fifth of queries sharing no URLs at all between runs. Any single-run share-of-model number is closer to a coin flip than a measurement. We dig into the repeated-prompt research — and what sample sizes actually stabilise the metric — in our post on AI answer volatility.
• Prompt-set bias. The metric is only as honest as the panel. A vendor (or an internal team keen to show progress) can pick prompts a brand already wins. Two companies “measuring share of model” with different panels are measuring different things.
• No standard definition. Some tools count any mention; some count only citations; some weight by position or sentiment; Jellyfish’s approach measures brand perception inside the model, not just answer inclusion. A “34% share of model” from one vendor and a “12%” from another are not comparable, and neither is auditable from the outside.
• Personalisation and context. Logged-in AI assistants adapt to user history and location. Your tracking tool’s clean-room answers may not match what a real buyer with six months of chat history sees.
• It measures visibility, not truth. A high share of model built on third-party pages that describe you inaccurately is a liability wearing a KPI’s clothes.

None of this makes the metric worthless. It makes trend lines on a fixed methodology valuable, and cross-vendor comparisons or one-off snapshots close to meaningless.

Run it yourself: a starter method with no tooling budget

You can produce a defensible share-of-model baseline with a spreadsheet and a few hours a week. This mirrors how disciplined teams (ours included) run internal measurement.

• 1. Build a panel of 15–25 prompts. Pull them from sales-call questions, support tickets and your keyword research, phrased the way a buyer would ask an assistant. Include head questions (“best X for Y”), comparison questions and problem questions. Freeze the list.
• 2. Pick 3 engines. For most SaaS categories: ChatGPT, Gemini and Perplexity. Use fresh sessions (no login, or a clean account) so personalisation doesn’t contaminate results.
• 3. Run each prompt 3–5 times per engine, per period. Weekly or fortnightly is enough for a trend. Yes, this is tedious; that repetition is the difference between a measurement and an anecdote.
• 4. Log four things per answer: was your brand mentioned (yes/no); was your domain cited as a source (yes/no); were you recommended or merely listed; and which competitors appeared. Log every cited URL — over time this tells you which third-party pages are feeding the engines.
• 5. Compute the shares. Mention rate = answers mentioning you ÷ total answers. Citation rate = answers citing your domain ÷ total answers. Share of model = your mentions ÷ all brand mentions across the sampled answers. Report all three; they move independently.
• 6. Cross-check against outcomes. Watch AI-referral sessions and assistant user-agents in your analytics, and tag inbound leads that say “ChatGPT recommended you”. The metric earns its place when it correlates with pipeline.

What should you do to move the number? That is answer-engine optimisation, and it is a discipline of its own — our AEO guide for SaaS covers the tactics, from citable-source strategy to structured answers, so we won’t repeat them here.

Visibility is only half the funnel

A rising share of model produces a distinctive kind of demand: fewer clicks, but higher-intent enquiries from buyers who arrive pre-sold by an AI’s recommendation. Those enquiries are unforgiving of slow follow-up — the buyer asked an assistant precisely because they wanted an instant answer. That is where Zian sits: our AI sales agents respond to inbound enquiries in seconds across phone, SMS, email and WhatsApp, with SmartReach AI™ handling outreach timing and PrecisionPitch AI™ matching the message to the enquiry. If you are investing in AI-search visibility, pair it with speed-to-lead automation so the demand you win in the answer box doesn’t die in your inbox. Apply For Partnership

Frequently asked questions

What is share of model in AI search?

Share of model is the percentage of AI-generated answers — across a fixed set of prompts and engines — in which your brand appears, relative to all brand appearances in those answers. It is the AI-search successor to rank position: because engines like ChatGPT synthesise one answer instead of listing results, the metric tracks inclusion in answers rather than position on a page.

Who coined the term “share of model”?

The clearest documented origin is Jellyfish, part of The Brandtech Group, which launched a trademarked Share of Model™ platform in 2024 to track how large language models perceive and recommend brands. The broader idea has been popularised in parallel by AI-visibility platforms under names like “AI share of voice” and “share of answer”, so no single individual can honestly be credited with coining the concept.

How is share of model different from AI share of voice?

In practice the terms overlap heavily and are often used interchangeably. “AI share of voice” usually follows the classic formula — your AI mentions divided by total AI mentions across all brands in your category, as Semrush defines it — while “share of model” is sometimes used more broadly to include how a model perceives a brand, not just how often it mentions it. There is no standards body; always check how a given tool defines its number.

Why do share of model numbers change between identical runs?

AI engines are probabilistic: the same prompt can surface different answers, brands and citations on consecutive runs. When SE Ranking ran 10,000 identical queries through Google’s AI Mode three times in one day, the average overlap of exact cited URLs across runs was just 9.2%. This is why credible share-of-model measurement always samples — running each prompt multiple times and averaging — rather than trusting any single answer.

Can I measure share of model without paying for a tool?

Yes. Freeze a panel of 15–25 buyer-style prompts, run each one 3–5 times per engine on a weekly or fortnightly cycle using clean sessions, and log mentions, citations of your domain, recommendations and competitor appearances in a spreadsheet. Compute mention rate, citation rate and share of mentions separately. The method is tedious but produces a defensible trend line — which matters more than any single snapshot.

Does a higher share of model actually produce revenue?

Visibility in AI answers tends to produce fewer clicks but higher-intent enquiries, because buyers arrive after the assistant has already shortlisted or recommended you. To connect the metric to revenue, track AI-referral sessions in your analytics, tag leads that mention an AI recommendation, and make sure inbound enquiries get an immediate response — AI-referred buyers expect instant answers, so speed-to-lead determines how much of that demand converts.