Eli Pariser's demonstration, in March 2011, was one slide and it did more work than the book.

Two friends. Same search term. Completely different results. One received investment news; the other received coverage of an environmental disaster. Neither had asked to be shown a different internet. Neither could see that they had been.

That is the whole concept, and the load-bearing word is invisible. An echo chamber is something you build — you can see the walls because you put them there. A filter bubble is assembled for you, continuously, out of what you clicked last month, and it does not announce itself. Platforms, Pariser wrote, are prediction engines, creating a unique universe of information for each person.

Then the research arrived, and it complicated the story in a way worth stating plainly.

A large 2015 study, using one platform's own data, measured how much cross-cutting content — material from the other side — got removed at each stage. The algorithm cut it by 5% for conservatives and 8% for liberals.

User choice cut cross-cutting clicks by 17% and 6%.

The algorithm narrowed. On average, people narrowed more — though for liberals the algorithm's cut was the larger of the two. And separately, an analysis of personalised music recommendations found personalisation led listeners to broaden their tastes rather than narrow them — the same mechanism producing the opposite result in a different domain.

So the honest position is neither the filter bubble is a myth nor the algorithm is choosing your reality. It is: the mechanism is real, the magnitude is contested, and the larger filter appears to be the one operating between your eyes and your finger.

Which is less comfortable than the original thesis. A bubble built by a company is something that could be regulated. A bubble built by preference is something you would have to want to leave.