How to measure share of answer across AI assistants
A practical method for tracking whether your brand is named and cited by ChatGPT, Perplexity, Gemini, Copilot and AI Overviews — and what the numbers actually mean.
- PUBLISHED
- Updated
- FILED UNDER
- Measurement
- READING TIME
- 8 minutes
- WORDS BY
- Shopify GEO SEO Agency
The first question every merchant asks about GEO is how you would even know if it worked. It is a fair question, and the answer is less exotic than it sounds: you build a rank tracker, except the unit of measurement is a mention rather than a position.
Step one: fix the prompt set
Write 100 to 300 prompts phrased the way a shopper talks to an assistant — full questions with constraints, not keywords. "What's the best waterproof hiking boot for wide feet under $200?" rather than "waterproof hiking boots." Draw them from support tickets, on-site search logs and review questions.
Then freeze the set. Changing prompts between runs destroys comparability, which is the single most common mistake in the tools currently selling this as a feature.
Step two: control the run conditions
- Same model and version per assistant, recorded with every result — answers shift when a model updates, and you need to know which change is yours
- Fresh session, no memory, no personalisation, no prior context
- Consistent locale, since answers vary meaningfully by market
- Fixed schedule — weekly is enough; daily mostly measures sampling noise
- Several runs per prompt, because these systems are stochastic and a single run is a coin flip, not a measurement
Step three: record three things
- Named: did the answer mention your brand at all? This is share of answer.
- Cited: did the answer link your domain? This is citation share, and it is always lower.
- Displaced by: which domains were cited instead? This is your source gap, and it is the most actionable of the three.
What good looks like
There is no universal target or timeline for share of answer. Establish a baseline for your category and test conditions, then track repeat observations over time. Report raw samples alongside a rolling view, distinguish named mentions from linked citations, and record model changes that might explain movement.
Report it next to your conventional metrics, never instead of them. The failure mode we see most often is a brand whose answer visibility climbs while non-branded organic revenue quietly falls, because the technical work that would have supported both got deprioritised for content that reads well in a slide.
Put this into practice with Shopify SEO and Shopify GEO.