How we measure, no black boxes
If you're going to make decisions with our data, you have the right to know exactly where it comes from.
1. Real questions
We start from questions a customer who doesn't know you yet would ask: recommendations, comparisons, alternatives to a competitor and specific problems. Most don't mention your brand, so we can measure whether AI suggests you on its own. You can edit them and add your own.
2. Querying the official APIs
We send each question to ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google) and Perplexity through their official APIs, with web search enabled and your market's location whenever the model allows it. We don't use fake accounts or automate their apps.
3. Repetitions
Assistants don't always give the same answer. We repeat each question 1 to 3 times depending on the plan and work with percentages, never with a single answer.
4. Analyzing each answer
We detect whether your brand (and its variants) appears, its position against the competitors mentioned, whether your site is linked, which sources are cited and the tone used to describe you. Tone and new brands are classified with an AI model and can be checked by reading the full answer.
5. Margin of error
Each percentage shows how many answers back it up and its margin of error at 95% confidence. We only alert you to a rise or drop when it exceeds that margin.
Score = 50% visibility + 20% position + 15% citations of your site + 15% tone
| Visibility | Percentage of answers that mention your brand. |
| Position | Average of 1/position when you appear (first place is worth 1; third, 0.33), weighted by your visibility. |
| Citations | Percentage of answers that link to your domain as a source. |
| Tone | Positive = 1, neutral = 0.5, negative = 0, weighted by your visibility. |
Limitations worth knowing
- Consumer apps can personalize answers based on the user's history. The APIs return a non-personalized version, which is the most stable reference.
- “Gemini” measures Gemini with Google Search. Google's AI Overviews are measured separately.
- Brand detection depends on the variants you provide. Add acronyms, your legal name and product names.
- With few questions or repetitions the margin of error is larger. You'll always see it next to the figure.
- The score in your dashboard is a stable level: it combines your latest measurements (up to 4) to reduce noise. If a measurement rises or falls by more than the margin of error, the average restarts from that level and you see it right away.
What we never do
- Write or buy reviews.
- Generate traffic with bots.
- Create fake accounts on forums or social networks.
- Publish content on your site without your approval.