Methodology
Version 1.1. Every score on this site carries the version that produced it, so a change in method never passes for a change in a publisher.
What we read
Four public sources, once a week:
- sellers.json from each exchange we follow, which lists the publishers it represents. This is how a publisher enters our index.
- ads.txt on each publisher’s site, which lists who may sell its ads, directly or as a reseller.
- /.well-known/adagents.json on each site, which names the AI seller agents the publisher has authorized.
- The title and description of the home page, and nothing else on it. We skip the page when robots.txt refuses it.
What we never do
- Log in, fill a form or follow links into a site
- Collect anything about a site’s readers
- Estimate a figure we could not read
- Accept payment to enter or rank in a package
The two scores
Relevance: is the site about the topic?
Each topic has precise terms and generic ones, in English, French, German, Spanish and Italian. We look for them as whole words in the title, the description and the domain name.
A precise term counts two points and a generic one counts one. One point scores 30, two score 60, three score 80, four or more score 100. A generic word alone is never enough.
Supply path: how directly does it sell?
No ads.txt scores 0, because nobody can verify who may sell the site. A file with no direct seller scores 30.
A file with at least one direct seller starts at 60 and gains up to 40 points with the share of its lines that are direct.
From scores to a package
The package score is 60% relevance and 40% supply path. A publisher enters a topic’s package at 70 or more.
Confidence is High when we could read the title, the description and the ads.txt, Medium when one is missing, and Low otherwise.
A package with one edition is marked Baseline. We show a trend only once several editions exist.
Known limits
- Supply path is not traffic quality. We do not measure audience, viewability or fraud.
- A seller list names the selling company, which is sometimes a parent group rather than the site itself.
- Matching words in a title is a blunt reading of a site’s subject. Some publishers are missed, and a few are misfiled.
- We know what is published, not what is available to buy or at what price.
Ad Server figures
A site counts as live only if it served at least one ad in the last 30 days. Test campaigns are excluded.
Impressions and clicks are raw counts, not yet filtered for bots.
Versions
| Version | First used | What changed |
|---|---|---|
| v1.1 | 10 Oct 2026 | Precise and generic terms, five languages, whole-word matching, supply path no longer penalized by the number of resellers, Travel topic added. |
| v1.0 | 10 Oct 2026 | First method: English terms only, equal weight for every term. |