How does a recommendation algorithm create a filter bubble?

By repeatedly surfacing content close to what a user already engaged with, a personalisation algorithm narrows the user's exposure radius over many cycles. The slice of content they ever see shrinks geometrically, and content diversity collapses toward zero.

Filter bubble formation

The algorithm learns what you click — then shrinks your world.

Each dot is a piece of content in opinion space. A recommender keeps surfacing what sits closest to your last engagement. Watch your exposure radius collapse, and the bubble close in.

Cycles

Exposure radius

1.00

Content diversity

100%

Diversity readout — collapsing100%

The recommender

Algorithm aggressiveness70%

How hard personalisation pulls the feed toward what you just engaged with. Higher = faster, tighter bubble.

Exploration noise5%

Share of the feed forced to be diverse — like YouTube's 5–10% "explore" injection. Below ~15%, bubbles still form.

How to use

Set algorithm aggressiveness and exploration noise, then hit Play. Each cycle the recommender surfaces the content nearest your last click, you engage with the most appealing item, and your exposure radius tightens. Watch the dashed bubble close and the diversity readout fall.

What's happening

Each cycle, the recommender ranks unseen content by distance to where you last engaged and serves the closest. Engaging with an item shifts your position toward it and shrinks the radius geometrically — the slice of opinion space you can ever reach narrows.

Exploration noise (the platform occasionally injecting something diverse) slows the collapse but doesn't stop it. Below roughly 15% noise, a bubble still forms.

Why platforms tune low noise

Diverse content lowers short-term engagement metrics. A platform optimising for watch time or scroll depth always converges toward a narrow feed. The bubble is a side effect of the clicked KPI — not a goal anyone set.

Bubble vs echo chamber

Nguyen (2020) draws a line worth keeping. A filter bubble is incidental: you simply never encounter the other views because the algorithm stopped surfacing them. An echo chamber is active: contrary voices are heard and then discredited. This sim models the bubble — exposure quietly narrowing, no one slamming a door.

Citations

Pariser, E. (2011). The Filter Bubble: What the Internet Is Hiding from You. Penguin Press. The original framing of personalisation-driven information narrowing.
Bakshy, E., Messing, S. & Adamic, L.A. (2015). "Exposure to ideologically diverse news and opinion on Facebook." Science 348(6239): 1130–1132.
Flaxman, S., Goel, S. & Rao, J.M. (2016). "Filter bubbles, echo chambers, and online news consumption." Public Opinion Quarterly 80(S1): 298–320.
Nguyen, C.T. (2020). "Echo chambers and epistemic bubbles." Episteme 17(2): 141–161. Distinguishes echo chambers (closed) from bubbles (incidental).

Built by Joshua Chua · Fable rebuild · vanilla JS + Canvas