The Share-of-Model framework

If you can’t measure it, you can’t improve it. And until recently, nobody was measuring AI visibility at the precision marketing teams expect from every other channel.

We’ve been building and iterating a framework called Share of Model — designed to give brand and performance teams a clean, monthly, defensible view of how visible they are inside the major AI assistants. Here’s how it works.

The three components of Share of Model

Share of Model isn’t a single number. It’s three:

1. Citation Rate

For a given prompt, what’s the probability that your brand appears in the model’s answer at all? We test this across the four major models (ChatGPT, Gemini, Perplexity, Claude), with a fresh session each time, in three locales, on a defined schedule.

If a buyer in Munich asks Gemini for “the best AI marketing agency in Europe”, does your brand surface? Citation Rate is the percentage of relevant prompts where the answer is yes.

2. Citation Quality

Surfacing isn’t enough — how you’re framed inside the answer matters enormously. We score citations on:

• Position — first mention, listed in the top three, or buried at the end
• Sentiment — neutral, positive, qualified, or negative framing
• Context accuracy — does the model describe what you actually do, or hallucinate something close
• Linked vs. unlinked — did the model cite a source URL, or just name the brand

A high citation rate with low citation quality is a brand-risk signal. A low rate with high quality is an authority opportunity.

3. Share of Model

The composite metric. Across the priority prompt set, summed across models and locales, weighted by quality — what percentage of available “answer real estate” do you own, vs. your top three competitors?

It’s the equivalent of share of voice for the AI era — and it correlates with what most CMOs actually care about: whether you’re the default recommendation in your category.

How we measure it

The mechanics matter. Without rigour, you get noise.

• Prompt set: 50–200 buyer-aligned prompts, scoped per client. We don’t measure against fluff.
• Sampling: Each prompt run 5+ times per model per cycle to control for variance.
• Models: ChatGPT (default + with web), Gemini (default + 1.5 Pro), Perplexity, Claude.
• Locales: Default plus your priority markets — sessions configured per region.
• Cadence: Weekly internal monitoring; monthly client reporting.
• Storage: Every response logged, hashed, and version-controlled so we can replay any data point.

The output is a dashboard, a monthly report with movers and drivers, and a list of three to five concrete actions to take next month.

What changes the number

Once you start measuring it, the next question is: what actually moves it?

Roughly in order of impact (but the mix varies by category):

1. Authority citations — getting referenced on the surfaces models trust (industry publications, Wikipedia, technical docs, podcast transcripts).
2. Structured data on owned properties — proper schema across services, products, and articles.
3. Recency of owned content — fresh, dated, comprehensive content on the prompts you care about.
4. Entity clarity — Wikidata cleanup, brand-name disambiguation, founder/team profiles aligned across surfaces.
5. Off-site presence — guest writing, expert quotes, podcast appearances.
6. Page-level GEO optimisation — content engineered for the way models extract answers, not just for SERP ranking.

Notice what’s not on the list: backlinks (still useful indirectly, but not directly causal), keyword density (irrelevant), and on-page rankings (a lagging indicator).

What it means for your team

If you’re a CMO, Share of Model is the metric that lets you have a serious conversation about AI search at the board level. If you’re a head of growth, it’s the metric that gives you a real view into which channels are influencing the answers your buyers are getting.

If you’re a brand team, it’s the early-warning system for category disruption: when a competitor’s Share of Model creeps up while yours plateaus, you’re already losing the conversation — even if your performance metrics don’t show it yet.

We run this for our clients monthly. It’s not magic; it’s just measurement done properly. If you want to see what your numbers look like, we run a free baseline assessment for serious prospects.