The decision specifically mentions algorithmic recommandation as being speech, ergo the recommandation itself is the responsibility of the platform.
Where is the algorithmic recommandation that differs per user on HN?
Where is the algorithmic recommandation that differs per user on HN?