Definition: what is share of model?
Share of model transfers the logic of classic share of voice into generative systems: if 8 out of 100 topically relevant prompts mention your brand, your SoM is 8 percent. The value is collected per model and then weighted across the model landscape — a mention in a heavily used system counts for more than one in a niche system.
SoM matters because generative answers act as recommendations: they condense the market to a handful of named providers before any click happens. A brand that is systematically absent from the relevant prompt space simply does not take part in this pre-selection.
Model weighting
Weights follow actual usage in the target market and need regular recalibration — the shifts are fast. As a working model for the EU market (own calibration, 2026): ChatGPT ~52%, Google Gemini ~22%, Claude ~11%, Perplexity ~10%, Bing Copilot ~5%. Other markets differ considerably — ChatGPT dominates more strongly in Türkiye, Perplexity is more strongly represented in the US.
Important: these weights are market assumptions, not constants. They should be checked quarterly against current usage data and disclosed in reporting.
SoM within the KPI system
SoM answers the competitive question (“what share of model attention does my brand hold?”), citation rate the source question (“how often are we formally cited?”) and PVI the quality question (“how prominently and positively do we appear in the answer?”). Only the three together form a steerable picture; individually they invite misreading.
Practice and limits
- Prompt-set quality decides: SoM is only as good as the prompt set — it must reflect real user questions of the target market, not wish-list keywords.
- Plan for variance: generative answers vary between runs; robust values need several runs and time series instead of single measurements.
- Regionalise: collect and weight SoM per market — a global blend hides the differences that matter.
Related terms
SoM belongs to the KPI family of generative visibility with PVI, citation rate and brand mention density. How SoM fits into the reporting set of the zero-click era is covered in Zero-click, ROI & KPIs.
SoM is the market share of the generative era
Generative answers condense markets to a few named providers. Share of model measures how often your brand belongs to that pre-selection — weighted across the systems your target market actually uses.
FAQ on share of model
How is share of model calculated? ▾
Per model: the share of answers in the prompt set in which the brand appears as a top source or mention. The per-model values are then weighted by market usage and combined into one total. Several runs per prompt absorb answer variance.
How does SoM differ from classic share of voice? ▾
Share of voice measures presence in media or advertising reach. SoM measures presence in generative answers — the pre-selection AI systems present to users. The logic is related; the data source and the levers are different.
How often should model weights be recalibrated? ▾
Quarterly. System usage shares shift quickly, and the weights are market assumptions, not constants. Every reporting period should disclose the weights used.
What is a good SoM value? ▾
There is no universal target — what matters is the comparison with direct competitors in the same prompt set and the trend over time. A SoM near zero in your core prompt space is the unambiguous alarm signal.