Method

How Seedli measures AI visibility

Every number Seedli reports can be traced to a measured answer from a named model on a dated day. This page documents how that measurement works, what Seedli refuses to do with it, and where the data lives.

What Seedli measures

Seedli measures what AI models tell buyers at every stage of the decision journey: who gets considered, which criteria decide, who gets chosen, who customers are told to stay with and recommend. The full product surface, stage by stage, is on the features page.

The method, in citable form

Seedli asks AI models the questions buyers ask, in the buyer’s own phrasing, and records what comes back. As of Q3 2026 that means ChatGPT, Gemini, and Claude.

The same questions are asked repeatedly, on a schedule, so a brand’s share of answers is a measured rate rather than a screenshot. Answers are parsed for the brands named, the criteria applied, and the elimination language used, then reported over rolling time windows. No figure is reported until it rests on a minimum count of answers; a share built on too few responses is displayed as insufficient data, not as a number.

The same method runs the public category decodes at The Sunday Shortlist, where the measurement windows and answer counts are printed with every finding. That archive is the method working in public.

What Seedli does not do

Seedli does not sell placement. No one can make an AI model recommend a brand, and Seedli does not claim to. What the instrument sells is measurement: where a brand stands in the answers, and the map of which evidence gaps drive that standing.

A brief is a template for a claim, not a licence to make it. Seedli’s content briefs are built from measured gaps, but the facts that close a gap must come from the customer’s own record. Where a claim cannot be verified, the guidance is to not publish it.

Measurement about a company is never published without that company. Public research reports category-level patterns. What the instrument found about any specific company stays between Seedli and that company, including the fact of having been measured.

Where the data lives

Customer data is stored in an EU database (AWS, Stockholm) and deleted within 30 days of account termination. Market descriptions and brand names, never account data, are transmitted to the model providers for analysis. The complete sub-processor list, transfer safeguards, and breach-notification terms are public in the Data Processing Agreement.

Behind the product

Seedli is built and operated by Seedli Labs ApS, CVR 46386981, Denmark. The method comes from The Recommendation Gap, the book that defined recommendation design, written by Seedli’s founder, Flemming Rubak.