Measuring Share of Answer: a working definition
Share of Answer is the percentage of relevant prompts in which a brand is cited inside an AI-generated answer. Here's how we measure it.
Founder 1 min read
Summarize with AI Open this article in your preferred assistant
In this article
Most agencies measure rankings. We measure citations.
Share of Answer (SoA) is the percentage of relevant prompts in which a brand is cited inside an AI-generated answer — across ChatGPT, Perplexity, Gemini, and Google AI Overviews.
It’s not a vanity metric. It’s the closest thing we have to “rank” in a world where the answer is the destination.
Why it matters
When a buyer asks ChatGPT “best GEO agency in Spain”, the answer they get does three things:
- Picks a small set of named brands (usually 3–7).
- Quotes or paraphrases content from those brands’ sites.
- Links — sometimes — to a source.
If your brand isn’t in step 1, none of the rest matters.
How we measure it
For every engagement, we run 50 prompts across four models — ChatGPT (GPT-4o + reasoning), Perplexity (default + Pro), Gemini (2.5 Pro), and Google AI Overviews — and log:
- Whether the brand was cited (binary).
- Position in the cited list (ordinal).
- Whether the citation linked to the brand’s own site (boolean).
- Which sources the model cited instead (competitor map).
We then compute SoA as cited / total, weighted by prompt commercial intent.
What good looks like
For a growth-stage B2B brand in a niche category, 30–45% SoA is a strong baseline after 90 days of work. Above 60% usually means you’ve cornered the entity graph for your category.
We re-measure monthly. The delta is the report.
Reference baseline
Source: Citable — 50-prompt sets, 24 engagements, 2025–2026
Typical Share of Answer baselines by surface (B2B SaaS, pre-engagement)
- Perplexity 18%
Most volatile; updates fastest
- Google AI Overviews 12%
25% of Google queries surface one
- ChatGPT 9%
Slowest to update — strongest signal
- Gemini 7%
Cited from Search Index + Knowledge Graph
Frequently asked
Questions buyers ask before booking
Why a fixed 50-prompt set rather than a dynamic one?
Consistency. Share of Answer is only a useful metric if you can compare it across months. Adding or removing prompts mid-engagement breaks attribution. Fix the set at engagement start, only update on real category change.
Why test in the live UI instead of the API?
API responses do not always match what real users see. Models receive system prompts, retrieval scaffolding, and personalization in the chat UI that the bare API call does not replicate. Buyers see the UI; we test the UI.
What is a good Share of Answer baseline?
Strongly category-dependent. For mature B2B SaaS categories, baselines often start at 5–15% (cited in 3–8 of 50 prompts) before any GEO work. Boutique categories with one or two dominant players can baseline higher. The number matters less than the trend after intervention.
How often should I re-measure?
Monthly during active engagement. Quarterly after a campaign ends, to track decay. Weekly for high-velocity categories where competitor activity is intense — but weekly measurement is operationally expensive without tooling like Profound or Peec AI.