Chorus · last updated
Research notice for YouTube commenters
If you wrote a public comment under a YouTube video about AI or technology, your comment may have been used in the research behind Chorus. We cannot tell every commenter personally, so we publish this notice instead, as Article 14(5)(b) of the GDPR asks.
Who is responsible
Baki Bektas, Berlin, Germany, an independent developer. Contact: [email protected] (to be confirmed by Baki).
Why
The research asks one question: can we predict which comments a creator will answer, using only numbers about each comment and never its text? The answer powers Replyworthy, a feature of the free Chorus extension that helps creators find comments worth a reply.
What was processed
- Which videos. 1,177 public videos from 384 channels, all in the curator's own AI playlist. The selection reflects the curator's interests. No viewing data of the curator is part of the research.
- What was fetched. The public comments and replies under those videos, through YouTube's official Data API, between January and 25 September 2026: the comment text, the author's public name and channel id, the time and whether the creator replied.
- What was computed. For each top-level comment, 46 numbers: its shape (length, question marks, emoji, timestamps, mentions), its timing, its video, the channel's reply history and whether the creator replied. No text, names or ids are in this table.
- Coding a sample. For a related study of what comments are about, a sample of comments was read by AI language models (Anthropic's Claude, used under the curator's own account) to code topic, stance and type of comment. [Scope and wording to be confirmed by Baki.]
Some comments reveal opinions on politics, health or religion. We do not look for these, do not code anyone by them, and publish nothing that could tie such a code to a person.
What is published, and what is not
- Published: aggregate results (for example how well the model predicts replies across all comments), the list of public video ids the research used, the method and the code.
- Never published: comment text, usernames, channel ids of commenters, comment ids, the per-comment table and per-comment predictions. They stay on one computer and are not shared.
The method is described in how the table was built and the published files are listed in the data overview.
Legal basis
Legitimate interest in scientific research (GDPR Article 6(1)(f)), with the safeguards of Article 89(1): only the data needed, pseudonymised ids, derived numbers instead of text, and nothing personal published. Where a comment reveals special categories of data, the basis is Article 9(2)(j) with section 27 of the German Federal Data Protection Act (BDSG), and Article 9(2)(e) for data the author made public.
We do not contact commenters one by one because we hold no contact details, and finding each author among many thousands would take a disproportionate effort (Article 14(5)(b)).
How long it is kept
- The fetched comments, and the tables built from them, are deleted or refreshed from YouTube within 30 days of fetching, as YouTube's API policies require. For the main fetch of 25 September 2026 that means by 25 October 2026. [Date to be confirmed by Baki.]
- The keys that link pseudonymised ids back to YouTube ids are deleted when the research ends.
- Published aggregate results contain no personal data and stay online.
- Anyone can rebuild the table from the public video ids with their own YouTube key; they are then responsible for their own copy.
Your rights and how to object
You can object to the use of your comments (Article 21), and ask for access to or deletion of what we hold about you (Articles 15 and 17). Write to [email protected] (to be confirmed by Baki) with the link to the video and your YouTube handle, so we can find your comments; we need nothing else. We will remove your comments from our copy and from any table we rebuild, and answer within one month.
You can also complain to a data protection authority. Ours is the Berliner Beauftragte für Datenschutz und Informationsfreiheit.
If you use the Chorus extension
This notice is about the research. What the extension does on your own computer is in the privacy policy. The source code is public.