Measuring Share of Model
How to build a prompt panel and measure citation share weekly across the five engines that matter, with the exact scoring formula we use.
1. What Share of Model actually is
Share of Model (SoM) is the percentage of buyer-intent prompts, across a defined engine mix, in which a brand is cited or named. Think of it as share of voice, but for machines that no longer show ten blue links.
2. Building the prompt panel
A panel of 40 well-chosen prompts beats a panel of 400 lazy ones. Categorize prompts across three tiers:
- 01Category prompts ("best CRM for early-stage SaaS"), high volume, low intent-to-buy but shape category perception.
- 02Comparison prompts ("Stripe vs Adyen for marketplaces"), mid volume, high intent.
- 03Solution-aware prompts ("how do I reduce Stripe chargebacks under $50"), low volume, buying intent.
Aim for roughly 40/30/30 across the three tiers. Revisit the panel every quarter, categories drift fast.
3. Which engines to score
For B2B: Google AI Overviews + ChatGPT Search + Perplexity. Add Gemini and Copilot if your buyers skew enterprise. For consumer: same three, but weight ChatGPT and Perplexity higher, that is where product research is drifting.
4. Scoring: named vs cited vs recommended
| Outcome | Weight | What it means |
|---|---|---|
| Recommended | 3 | Brand is the primary answer, first in list or explicitly endorsed |
| Cited | 2 | Brand link appears in the citation panel |
| Named | 1 | Brand mentioned inline without a citation link |
| Absent | 0 | No mention |
Weighted SoM reflects quality of presence, not just presence. A brand cited once but recommended weekly is winning.
5. Cadence and drift
Weekly is the sweet spot. Daily is noise. Monthly misses the reranker updates that move share by 10-20 points in a single week.
6. Turning scores into decisions
- Low SoM + high classic rank → GEO content shape problem. Restructure top-10 pages first.
- Low SoM + low classic rank → coverage / authority gap. Prioritize topical map expansion and digital PR.
- SoM strong in Perplexity but weak in ChatGPT → structured content works, freshness signals lag. Publish cadence needs to rise.
- ✓Start with 40 prompts across 3 intent tiers, not 400.
- ✓Weight recommendations higher than raw citations.
- ✓Weekly cadence catches reranker shifts you'd miss monthly.
- ✓Cross-engine gaps tell you which lever to pull next.