Do online filter bubbles narrow what we see, or do we narrow ourselves?

The idea of a "filter bubble", popularised in a well-known talk by Eli Pariser, is that personalised feeds quietly show us more of what we already agree with and less of what we do not, and that this could weaken the shared facts a democracy needs.

It is a powerful idea, and also a contested one. Some studies find that people's own choices of whom to follow and what to click matter at least as much as the algorithm; others find that recommendation systems can push attention in particular directions. The evidence differs by platform, country and time.

We are asking readers to weigh in with evidence, not only impressions. A good answer would say what you have seen yourself, cite what is known where possible, distinguish what the platform does from what users do, and suggest one change, to design or to behaviour, that could be tested.

Watch the talk first if you like, then tell us what you think it gets right and wrong.