Framework
GEO Measurement Framework: from prompts to pipeline
AI mentions are a leading indicator, not a business result. This framework defines every metric from the first prompt to closed-won revenue — with formulas, sources and cadences you can adopt as-is.
Key takeaways
- Measure two chains side by side: Spend → Leads → SQL → Pipeline → Revenue, and AI prompts → Mentions → Citations → Referral sessions → Leads → Revenue.
- Use a fixed, versioned prompt set and sample each prompt more than once per engine; report trends, not screenshots.
- North Star metrics: qualified pipeline generated, CAC, cost per SQL, AI-assisted pipeline and AI share of voice.
- Never use impressions, traffic or AI mentions alone as the North Star.
The two chains
One dashboard, two funnels.
Paid & organic growthSpendLeadsSQLPipelineRevenue
AI searchAI promptsMentionsCitationsReferral sessionsLeadsRevenue
Metric definitions
GEO metrics, defined.
| Metric | Formula | Source | Cadence |
|---|---|---|---|
| Prompt coverage | Prompts where the brand is mentioned at least once ÷ prompts tracked | Prompt tracker | Monthly |
| AI share of voice | Answers mentioning the brand ÷ answers sampled (per engine, per topic) | Prompt tracker | Weekly / monthly |
| Average mention position | Mean rank of the brand in list-style answers | Prompt tracker | Monthly |
| Citation rate | Answers linking to a brand-owned URL ÷ answers sampled | Prompt tracker | Weekly / monthly |
| Citation share | Citations to brand domains ÷ all citations in the prompt set | Prompt tracker | Monthly |
| Sentiment | Positive / neutral / negative framing of mentions | Prompt tracker + review | Monthly |
| AI referral sessions | Sessions in the AI Assistants channel | GA4 | Weekly |
| AI referral conversion rate | Leads from AI sessions ÷ AI sessions | GA4 + CRM | Monthly |
| AI-sourced pipeline | Opportunity value where first or last touch is an AI source | CRM | Monthly |
| AI-assisted pipeline | Opportunity value with any AI touch, incl. self-reported | CRM | Monthly |
Sampling method
Make the numbers trustworthy.
AI answers vary by phrasing, locale, personalization and model version. Control what you can and report variance honestly.
- Version the prompt set; change it deliberately and note changes in reports
- Group prompts by intent: problem, shortlist, comparison, pricing, implementation
- Fix locale and language per market; use logged-out sessions where possible
- Run each prompt at least three times per engine per sampling window
- Store full answers and cited URLs, not just yes/no mentions
- Report trends over at least three windows before drawing conclusions
Measurement layers
The full website measurement framework.
| Layer | Metrics |
|---|---|
| Awareness | AI share of voice, SERP visibility, branded search |
| Acquisition | Sessions, CTR, CPL, demo requests |
| Qualification | MQL, SQL, target-account rate |
| Pipeline | Opportunity value, pipeline sourced |
| Revenue | Closed-won, CAC, ROAS |
| GEO | Prompts won, citation rate, AI referrals, AI leads |
| Retention | Client expansion, renewal, NPS |
Related: How to track ChatGPT and AI referral traffic in GA4.
Next step
See where AI recommends your competitors instead of you.
Run the instant AI-readiness check in 30 seconds. Our strategists follow up with a prompt-level audit across ChatGPT, Google AI, Gemini and Perplexity within 48 hours.