Skip to content
Try for free

Methodology

How VISIBILIS measures

How VISIBILIS asks, which models answer, how the answers are evaluated and how the AI Visibility Index is built.

How the queries run

How we ask

VISIBILIS puts the questions to the providers’ application programming interface (API), each time with web search switched on. The models research the web as they do in their apps. Every question runs as a new conversation, with no history.

Three models

We measure in ChatGPT, Claude and Gemini. All three are included in every offer. If a model is unavailable, we repeat the affected queries. If a provider replaces a model, we only switch to a model from the same maker.

Two variants per question

Every question is asked twice. The first variant is the question as a customer would ask it (“unprompted”). The second also asks for sources (“on request”). That shows whether a model cites your website unprompted or only when asked for evidence.

Question set and evaluation

The question set

The questions are built from your details: website, industry, up to six focus areas and, for local businesses, the region. They cover what customers ask before a decision, from first orientation through comparing providers to buying. The questions are worded so that the models name providers. You review the set and approve it before we measure.

How we evaluate

Every answer is stored word for word. We evaluate whether your brand is mentioned, in which position, whether your website appears as a source, which other providers are named and which websites the models cite. For the visibility analysis we also check the question set and the evaluation by hand.

Go to the visibility analysis

The AI Visibility Index

The AI Visibility Index sums up the result in one number from 0 to 100. An unprompted answer can therefore score 110 points, an answer on request 90. Both variants occur equally often, so the index tops out at 100. Questions that contain your brand name do not count. We take the average per model; the index is the average of the three models.

  • Mention

    50

    points if the answer mentions your brand.

  • Position

    20

    points for first place, 10 for second or third.

  • Source

    40

    points if your website is cited unprompted, 20 when asked for sources.

Why 24 questions are enough for the analysis

24 questions in two variants put to three models produce 144 answers. With six focus areas, that is four questions and 24 answers per focus area. The analysis is meant to describe where you stand and derive actions, not to show a trend. For that, breadth of topics matters. If you want to follow changes over months or measure individual topics in more depth, use monitoring with 15 to 250 questions.

  • Questions

    24

  • Variants per question

    2

  • Models

    3

  • Answers

    144

Limits of the measurement

Measure first, then decide.

Try for freeVisibility analysis

A free sample with two questions, or go straight to the fixed-price visibility analysis.