The data behind a recommendation

When YouTube, Spotify or TikTok suggests something — where does the suggestion come from?

What the service knows about you:

The data	Example
What you have clicked on	which videos you opened
How long you stayed	the most important signal
What you skipped	says just as much
When and where	the time of day, your approximate location
Which device	a phone or a computer
What similar people liked	the great secret

The last row is the core. The service does not need to understand what a song is about. It only needs to know that people who liked A and B also tend to like C — and you liked A and B.

That is called collaborative filtering, and it is surprisingly powerful. It works without anybody ever having described the content.

What the service can guess wrongly about:

The situation	Why it goes wrong
You watched something once out of curiosity	the service thinks it is your interest
Somebody else borrowed your phone	their clicks become yours
You searched for something for a friend	the same thing
You have changed your mind	old habits linger for a long time
You are new	the service knows nothing — the «cold start problem»

The filter bubble is the larger question. The service shows more of what you already liked, because that is what keeps you there. Over time you see less and less of anything else.

That does not mean somebody has decided that you should see precisely that. It is a consequence of the system measuring what you stay with, and what you already like is what you stay with.

What you can do:

The measure	The effect
Look at your history	most services show it
Remove things you do not want to be influenced by	it actually works
Search actively instead of just scrolling	you steer instead of the feed
Use several sources	different bubbles

The honest picture: recommendations are often good. They help you find things you would otherwise have missed. The problem arises when they become the only way you find anything — and when what keeps you there is not what is good for you.

Task 1 — look at your own data.

Most services have a page where you can see what they have saved. Find it for a service you use:

• YouTube: the history and «my Google activity»
• Spotify: your listening history and the annual summary
• TikTok and Instagram: settings → activity

Look and answer:

• How far back does the history go?
• Is there anything there you did not know was being saved?
• Do you recognise yourself in what the service seems to believe about you?

Task 2 — test the system.

Deliberately watch three videos about something you never usually watch — cooking, fishing, old cars, anything at all.

Come back the next day. What is suggested now?

Task 3 — clear it and see what happens.

Remove the three videos from your history. Do the suggestions change back? How fast?

Task 4 — discuss.

• Is it good or bad that the service knows you so well?
• Who benefits from you staying longer?
• What would you like to be able to control that you cannot today?
• If you got to decide one rule for how recommendations may work — which one?

The last question is not hypothetical: EU rules now require large platforms to offer a feed that is not based on profiling. Have you tried it?