The feed

Cover of The feed, Frontier Report No. 04

Frontier Report No. 04 · 16 pages · 9 sources

Do recommendation algorithms change what we like?

Partly. People come to like what they see often, and feeds decide what is seen often. In a Spotify experiment, personalised picks raised listening by 29 per cent but cut the variety each person heard by 12 per cent. The best studies find little short-run effect on political views.

What the report found

  1. Most choices now start with a recommendation. Netflix reported that its recommendations influenced about 80 per cent of the hours streamed on the service, with the rest coming from search.
  2. Taste was never formed alone. Older research shows that people come to like what they see often, and that showing people what others chose makes hits bigger and less predictable.
  3. Personal feeds narrow each person while widening the whole. In a Spotify experiment, personalised podcast picks raised listening by 29 per cent and cut the variety of each listener's choices by 12 per cent, while variety across all listeners rose.
  4. Feeds tuned to engagement amplify anger. On one large platform, the ranked feed showed users more angry and hostile political posts than a time-ordered feed, and users said they preferred the posts the ranking left out.
  5. Rabbit holes are contested. Switching people to a time-ordered feed for three months during an election cut their time on the apps but did not change their political views, and simulated users who followed only YouTube's recommendations saw less partisan content than real users chose.

What is inside

  1. What is new
  2. What older research tells us
  3. What early studies suggest
  4. What nobody knows yet
  5. Questions worth watching

Plus a diagram of the feedback loop, three older findings and what feeds change about them, a signals list and a graded list of claims.

Who it is for

For anyone who scrolls, streams or shops online, and anyone who builds, ranks or buys space in a feed.

How we work

Frontier Reports look at behaviour the textbooks have not yet caught up with. Every claim is graded Established, Emerging or Untested. Where researchers disagree, as they do on whether feeds pull people into rabbit holes, both sides are set out before we say where we stand.

Cite this report

The Why We Project (2026) The feed: How recommendation systems shape taste. Frontier Report No. 04. Dubai: Why We Tech Ltd. Available at: thewhyweproject.com