Methodology

PrivacyRadar captures public privacy-policy pages, hashes the cleaned text, and only then asks a model what the company discloses. If the hash matches the last successful capture, no model runs.

Model output is untrusted. A claim becomes public only after a quote is found in the captured snapshot and an operator publishes a revision. Cosmetic date-stamp diffs stay off the public feed.

We report disclosed practices, not proven behavior. “We have not found evidence” is not the same as “this never happens.” Failed or blocked fetches are not empty policies.

Corrections supersede a publication. Prior revisions remain readable. This is not legal advice.