What is Share of Model — and why it's the new SOV

TL;DR: Share of Model (SoM) is the percentage of AI-generated answers in your category where your brand is mentioned. It replaces Share of Voice for the answer era. Measure it across ChatGPT, Claude, Gemini and Perplexity. Track it weekly. Grow it with the same three levers every time: be known, be easy to cite, be vouched for.

Why Share of Voice is breaking

For thirty years, marketers tracked Share of Voice (SOV): the percentage of paid impressions or organic search results occupied by your brand against the category. The implicit assumption was that attention to a brand correlated with consideration of that brand. Buy enough ads, rank for enough keywords, get enough press, and you would be in the consideration set.

That assumption is being unwound by a generation of buyers who skip search engines for AI assistants. When a marketing operations lead opens ChatGPT and asks "best AI visibility platform for a B2B SaaS company", the answer is two or three brands, in prose, with no blue links. The model has already made the shortlist. Share of Voice is no longer a leading indicator of consideration — it is a backward measurement of attention in a channel buyers are abandoning.

What is Share of Model?

Share of Model (SoM) is the percentage of generative-AI answers about your category where your brand is named. If you ask fifty high-intent buyer questions across six models, and your brand appears in twenty of those answers, your SoM is 40%.

The definition is simple, but the measurement isn't. Three things make SoM harder than it looks:

• Models disagree. ChatGPT might mention you in 50% of answers while Perplexity mentions you in 20%. A naïve average of all responses lets a chatty model dominate your score. The right approach is to compute per-model SoM and then average across models — so each model is one equal vote in your score.
• Queries matter. If you only test "what is the best CRM" you'll get vendor-neutral answers and underestimate yourself. You need real buyer queries across the full journey: problem-aware ("how do I reduce churn?"), solution-aware ("is HubSpot or Salesforce better for agencies?"), and decision-stage ("HubSpot vs Salesforce for a 5-person agency").
• The category is yours to define. SoM is only meaningful inside a specific competitive set. "AI visibility" is one category; "Generative engine optimization for B2B SaaS" is another. The narrower the category, the higher your SoM — but the more credible the measurement.

How to measure SoM without fooling yourself

A defensible SoM measurement has five inputs:

1. A curated list of 30–50 buyer queries, spread across journey stages and grounded in real search-intent data.
2. Direct API calls to OpenAI, Anthropic, Google, Perplexity, Tavily and Exa. Not scraped chat sessions. Not third-party proxies. The query you measure must be the query a real buyer would type.
3. Multiple temperature samples per query. Generative models are non-deterministic. Run each query three to five times and aggregate, or you'll attribute random variance to your content changes.
4. Named-entity recognition tuned for brand mentions — both exact match and known aliases. "Acme", "Acme Inc", "acme.com" all count.
5. Per-model normalization. Compute mentions ÷ queries per model, then average across models.

The output is one number, a per-model breakdown, a competitor leaderboard, and — if you've done it right — a list of the specific queries where you're missing and which competitor is eating your lunch.

How to grow your Share of Model

Once you have a baseline, the levers fall into three groups, in order of how quickly they move the number:

1. Be known

Models can only mention you if they've seen you. Your site, your product pages, and your category-defining content all need to be findable, crawlable and indexed by the major AI training pipelines. Concretely: ship structured-data markup (Organization, SoftwareApplication, FAQPage), submit a sitemap, get listed in AI-friendly directories, and earn coverage on the third-party domains the models trust as sources. The flywheel takes 6–12 weeks to register because training and retrieval indexes update on a lag.

2. Be easy to cite

When a model considers mentioning a brand, it looks for the combination of name + crisp factual claim + ideally a source. The easier it is to lift a one-sentence definition of what you do, what you cost, who you serve, the more often you get cited. This is why comparison pages, glossary entries and FAQ sections out-perform long-form thought leadership for AEO — they're built for extraction.

3. Be vouched for

The third lever is the slowest and the most durable. Models calibrate their confidence by counting how many trustworthy sources name you. A single G2 page is a ripple; a Forrester report, a podcast appearance, a peer review on a top-100 marketing blog — those are weight. Pitching the right third-party sources, in the right order, is the most strategic move you can make against a well-funded competitor.

What good looks like, week-by-week

A team that's serious about Share of Model runs a tight operating loop:

• Weekly: re-run the SoM audit. Check your score and per-model deltas. Flag any new competitor showing up.
• Bi-weekly: ship one piece of content from the action plan. A comparison page, a JSON-LD upgrade on a key page, or a pitch to a third-party source.
• Monthly: rebalance the query set. Buyers' language drifts; your audit should drift with it.
• Quarterly: review the model mix. New models enter the buyer's vocabulary (Claude, Perplexity, then domain-specific assistants). Add them when their share of buyer attention crosses ~5%.

The bottom line

Share of Voice was a proxy for being in the consideration set in a world where attention was the gating step. Share of Model is the measurement when generative AI is the consideration set. The brands that win the next decade will be the ones that already know their SoM, ship to grow it weekly, and treat AI visibility as a real channel — not an afterthought of SEO.