How aeoda measures AI visibility

What goes into the score

The unit of measurement is a question that a buyer might ask. We generate category questions and problem questions without the brand name in them. Questions that contain the brand name are kept apart because they would inflate the score.

Each question is asked to each engine one or more times. For every engine the share of runs that mention the brand is recorded. The question value is the average of the engines that returned data.

The formula

Score = 100 × the weighted average of question values. Live-web engines count with weight 1 and model-knowledge engines with weight 0.5. An engine score uses the same formula with one engine.

Other metrics use the same approach: share of voice is your mentions divided by all brand mentions, average position is your rank among the brands named in an answer, and sentiment ranges from -1 to +1.

The confidence range

AI answers vary between runs, so one measurement is a sample, not a fact. We estimate the standard error across questions, which captures both question and run variance, and report a 90% confidence range around the score.

With fewer than five questions the result is marked as a preview with a wide range. More questions and more runs narrow the range.

When a change counts

A rise or fall is shown only when it is larger than the combined uncertainty of both measurements. Otherwise we write that the change is within the noise. Emails never say a score went up or down without this check.

Engine labels

ChatGPT and Claude answer with live web search and are labeled live web. Gemini is measured without live search and is labeled model knowledge, because it reflects what the model itself knows.

Google AI Overviews are not measured.

Limits you should know

We query engines through their official interfaces. Answers can differ from what people see in the consumer apps, so results are API observations rather than a census of every user conversation.

Results depend on the question set, the market and the language. We publish the method so that you can judge it, and we keep adjusting it as we measure.

Mentions are not sales. The score tells you whether AI includes a brand in its answers, not whether those answers convert.