How to measure AI visibility in ChatGPT, Perplexity and Gemini
Build a repeatable sample of shopping questions, track mentions and citations, and separate AI visibility from referral visits and sales.
In this article6 sections
AI visibility is a measurement of where your brand appears in a defined set of answers. It is not automatically a count of shoppers reached. A useful report makes the sample, collection method and limits visible.
Define what you want to learn
Start with a business question: are shoppers finding your brand when comparing materials, choosing a product for a use case or evaluating alternatives? Build questions from customer support, site search, product knowledge and relevant search research.
Use three groups: category discovery, comparisons and product-specific questions. Keep branded questions separate; asking an engine about your own brand is a different test from asking it to recommend options in a category.
Record the test conditions
For each observation, record the engine and product surface, date, market, language, exact question and any relevant context. Save the answer and cited URLs. Distinguish a consumer-interface test from an API response; do not assume they represent the same experience.
Include ChatGPT, Perplexity and Gemini where relevant. Treat Google AI Overviews and AI Mode as separate surfaces. Add Claude or other engines when there is a reason tied to the audience, not just to increase a logo count.
Use clear definitions
| Measure | Definition for your sample |
|---|---|
| Mention rate | Eligible answers naming the brand ÷ eligible answers collected |
| Citation rate | Eligible answers linking to the brand’s site ÷ eligible answers collected |
| Accuracy | Reviewed answers with correct product facts, using a written rubric |
| Referral visits | Visits identifiable as coming from an AI source in your analytics |
| Referral conversions | Conversions attributed under your stated analytics rules |
Define “eligible” before collection. Record failed runs and refusals separately so a change in collection reliability does not look like a visibility change. If several brands appear in an answer, each may count as mentioned; that is not a share-of-voice calculation.
Repeat the sample
One answer can be misleading. Repeat questions on an agreed schedule, retain raw observations and avoid changing the sample halfway through a comparison. If you add questions, report them as a new cohort.
An illustrative sample of 40 eligible answers with 10 brand mentions has a 25% mention rate. That means 25% of this sample mentioned the brand. It does not mean a quarter of all shoppers saw it.
Turn findings into work
Classify each gap before assigning a fix. Missing specifications may call for product-information updates. Conflicting descriptions may need a review of site and retailer copy. Repeated citations to relevant third-party guides may inform an outreach investigation.
A citation is evidence of what appeared in an observed answer, not proof that copying the cited page will reproduce the recommendation. Keep the hypothesis explicit and evaluate it after implementation.
Report business outcomes separately
Place identifiable referral visits and conversions beside visibility metrics, not inside them. A shopper may discover a brand in an AI answer and purchase through another channel; another visitor may click without buying. Neither can be resolved by multiplying a visibility score by average order value.
Google documents that its AI features are included in overall Web search reporting in Search Console. Do not label all Google organic traffic as AI traffic. Google measurement guidance
For a structured starting point, explore our AI visibility audit and ecommerce AEO service.
