How an AI visibility report is made

Above all, we survey what AI systems tell customers looking for a company in your industry; questions about the company by name are a separate, smaller part of the survey. We record every answer, and every number in the report comes from counting those answers. The method is the same in every report and at every level; what changes is the number of systems and models and the depth of the analysis.

Customers’ questions

From the company’s website we establish its industry, offering and market. We write the questions to fit the company’s profile, in customers’ language and without company names, the way someone looking for a supplier would ask. In the Full and Premium reports an analyst approves the question set. The set is assigned to the company and reused in the follow-up study and Monitoring, so results can be compared. For businesses that customers visit in person, we ask questions from four distances around their location.

AI systems and how we ask

Every question goes to the AI systems via API, with the same settings for every company. ChatGPT, Gemini and Claude get every question with and without web search, to separate what AI finds from what it remembers. We also query Google AI Overviews and AI Mode. AI can answer the same question differently, which is why in the Full and Premium reports we ask every question three times.

Questions about the company

Separately, we ask AI about your company by name, without search: what it does, where it operates, how to contact it. We compare the answers with your website and show every difference. Your website is the reference; you judge whether it reflects the current state of affairs.

The website as crawlers see it

We fetch the company’s pages the way each AI system’s crawler does, and check how many of five key facts from the website reach it. A fact can get stuck at one of four thresholds.

How we assess answers

In every answer we check whether your company’s name comes up, and determine how strongly AI recommends it: first choice, recommendation with reasons, position on a list, passing mention or no name. Every rating is based on a quote from the answer. The tasks in the action plan follow from issues found in the study, and each one is backed by evidence from the study.

AI Visibility Score (AVS)

The score from 0 to 100 falls into one of five bands: 0–14 Invisible to AI · 15–34 Rarely recommended · 35–54 Occasionally recommended · 55–74 Consistently recommended · 75–100 Category leader.

Evidence and archive

Every answer goes to the archive in full, with the model, the date and an evidence number. Quotes in the report carry an evidence number that leads to the answer in the archive, so every sentence of the report can be checked against its source. You get the archive together with the report. In the Full report the survey card also shows how much the answers varied between runs.

Later measurements

The follow-up study and Monitoring ask the same question set to the same systems with the same settings. Every task from the previous plan gets a verdict: Worked, No improvement or Not done. We describe changes below the variability threshold as no change.