Check a link

Paste an article URL and see which of its outbound links carry affiliate attribution, which network, and which merchant. The list of things this page will never do was published before the form was built, and it is the list the form has to satisfy.

What it does

Classify each outbound link against a signature bundle shipped as data, never as code. The extension does this entirely in your browser; this page asks our API, which fetches the article server-side and never follows the links on it.
Return one of four confidence tiers with the evidence that decided it — the host pattern, the path constraint, the parameter, or the publisher’s own rel="sponsored" attribute — so the answer can be checked rather than trusted.
Say plainly when it does not know. A cloaked link whose destination we have not observed is reported as unresolved, not guessed.
Name the merchant and, where a rate is public, show it with the URL it was read from and the date it was read.

What it never does

Modify the link. It will never strip, replace, inject, rewrite, redirect or fire an affiliate link. Not as a feature, not as an option, not behind a flag. Extensions that overwrite a creator’s tag with their own are the practice this project exists to document.
Send us the URL in the clear. Where a lookup cannot be answered on your device, it will send the first five characters of a hash and receive every match, filtering locally. The server cannot learn which link you asked about.
Require an account. No email, no identifier, no history, no profile.
Follow the link on your behalf. Redirect resolution happens on our own infrastructure, without cookies, and never from your browser — so no affiliate cookie is ever set by us and no creator loses a commission to us.

Why the constraints shipped before the feature

A checker that answers quickly and quietly logs what it was asked is easy to build and hard to detect. Publishing the constraints first makes them checkable now that the form exists: the list above is the list this page has to satisfy, and it was written before there was anything to excuse.

The classifier itself is open source and its rules are data rather than code. Until this page works, the same answers can be obtained by running it directly.

Elsewhere

Publisher report cards carry the same classification applied to a whole corpus, and the methodology records what the classifier misses and how often.