How we measure your brand's visibility in AI

When we say your brand has 42% visibility in ChatGPT, that number follows a specific recipe. In this article we explain how we put it together: what we ask, whom, from where, how often, and how we turn thousands of answers into metrics you can compare.

The basic unit: a chat

Everything starts with a chat: one model's answer to one prompt, from one country, on one day. If you track 50 prompts on 3 models from Mexico, you get 150 chats a day, about 4,500 a month. That's the universe we calculate everything on.

The full formula is prompts × models × countries × days, and every factor matters:

• Prompts. These are the questions your customers ask. We mix questions with and without your brand, and with different intents: informational (“how does a digital bank account work?”), commercial (“best fee-free digital accounts”) and transactional (“open a digital account online”).
• Models. ChatGPT, Gemini, AI Mode, AI Overviews, Perplexity or Copilot. Each one uses different sources and answers differently, so we measure them separately.
• Countries. The same question asked from Mexico and from the US can recommend different brands. Each prompt runs from the country and in the language you set up.
• Days. Models don't always give the same answer: the same question can get a different response tomorrow. That's why we repeat every day and work with trends, not screenshots.

What we keep from each answer

For every chat we store the full answer and the sources the model used. Then we analyze it to detect which brands appear (by name, alias and domain), in what order, in what tone, and which sites were cited. When the model exposes them, we also record the searches it ran before answering, which we call fan-outs.

You can open any chat and read exactly what the model said. No number in Fedlum is a black box.

The four metrics

Visibility. The share of chats that mention your brand. If 6 out of 10 answers name you, your visibility is 60%. It's the first signal: does AI consider you an option?

Position. Your average place when you are mentioned, counting every brand detected in the answer (1 is first). Lower is better. It only counts chats where you appear, so always read it together with visibility.

Sentiment. The tone the model uses about you, on a 0–100 scale. We calculate it from the sentences around each mention, and you can see the sentences behind the number.

Share of Voice. Your slice of all mentions among the brands you track. If you get 4 mentions and your competitors get 12, your Share of Voice is 25%. Across all the brands you track, it adds up to 100%.

Alongside those four we calculate the win rate: the share of chats where you appear first. It's always less than or equal to visibility.

A quick example

Picture 10 chats about “best apps to send money to Mexico”. A brand appears in 6 of them: first in three and second in the other three.

• Visibility: 6 out of 10, so 60%.
• Position: (1 + 1 + 1 + 2 + 2 + 2) ÷ 6 = 1.5.
• Win rate: 3 out of 10, so 30%.

Sources and citations

Brand metrics tell you where you stand; source metrics explain why. For every domain and URL we measure how many times a model used it (retrievals), in what share of chats it appeared and how many times the answer cited it explicitly. Gap analysis crosses that with mentions and shows the sources that cite your competitors but not you, ranked by opportunity.

Changes and comparisons

Every metric is compared with the previous period of the same length: if you look at the last 7 days, the change is against the 7 days before. Filters by model, country, topic and tag apply the same way to every metric, and the numbers in the dashboard, exports, the API and the MCP server come from the same queries, so they always match.

What we don't do yet

Fedlum is in private beta. Daily collection is switched on project by project as we connect each model, and until then you can explore a demo project with sample data. Whenever a number comes from simulated data, the product will tell you.