Practice
Measuring AI visibility: prompts, mentions, sources
1The short answer
AI visibility is measured by regularly running a fixed catalogue of business-relevant questions across several AI systems. The analysis covers mention rate, the context and tone of each mention, the competitors named and the sources cited – tracked as a trend over time, not as a single screenshot.
Why a screenshot proves nothing
AI answers are not static. The same question can be answered differently depending on wording, conversation history, location and timing. A single screenshot showing a brand being named therefore says very little. Measurement only becomes meaningful through repetition and structure.
Step 1: Define the prompt catalogue
The foundation is a catalogue of questions that potential customers actually ask. We typically divide it into four groups:
- Category questions: “Which providers are there for …?”
- Comparison questions: “Which is better, A or B?” or “Which provider is suitable for …?”
- Problem questions: “How do I solve …?” – without naming a brand
- Brand questions: “What is [your brand]?”, “Is [your brand] reputable?”
Step 2: The right metrics
| Metric | What it tells you |
|---|---|
| Mention rate | In how many relevant answers the brand appears |
| Share of Voice | How often the brand is named relative to competitors |
| Citations | Whether and how often your own website is linked as a source |
| Context | Whether the brand is recommended, merely mentioned or portrayed critically |
| Accuracy | Whether services, locations and positioning are represented correctly |
| Source landscape | Which third-party pages shape the answers on your topic |
Step 3: Analyse by platform and topic
ChatGPT, Gemini, Perplexity and Google AI Overviews draw on different sources. A brand can be strong in one system and barely visible in another. Visibility also often varies widely between topics. Breaking the analysis down by platform and topic shows where action will pay off most.
Step 4: Connect with web analytics
Many AI systems link to their sources. Visits from ChatGPT, Perplexity or Gemini can therefore be analysed as a separate channel in web analytics. This shows whether mentions also lead to visits and enquiries. It is also worth looking at the “AI Performance” report (beta) in Bing Webmaster Tools.
What reliable measurement looks like
- A disclosed prompt catalogue and measurement period
- Several queries per prompt instead of one-off results
- Comparison with competitors instead of isolated figures
- No guaranteed placements
How we deliver this measurement as an ongoing service is described on the AI Visibility Monitoring page.
Author
Jan Hugo · Managing Director, Search, Tracking & AI
Technical foundations, data models, tracking architecture and AI-assisted analysis.
