You watch one political video on Instagram or TikTok. Perhaps it is about immigration, a war, an election, climate change, or a political scandal. You pause for a few seconds. Maybe you watch it twice. You do not follow the creator or search for the subject again.

And yet, over the next few days, something changes.

Another video appears. Then another. Soon, your feed seems full of the same issue. Different speakers, different headlines, different clips—but the same anxiety, controversy or political question keeps returning.

After a while, it becomes easy to think: Everyone must be talking about this.

But are they?

Or has an algorithm simply learned that you are likely to keep watching?

This seemingly ordinary experience contains one of the most important political questions of the AI age.

We often worry about algorithms because they might give us false information. But an algorithm does not need to lie to influence us. Sometimes, it only needs to decide what we see repeatedly—and what quietly disappears from view.

That is what makes recommendation systems so politically powerful.

When recommendation becomes agenda-setting

Platforms such as YouTube, TikTok, Instagram and X contain more information than anyone could possibly consume. Something therefore has to sort it.

Recommendation systems perform that task.

They decide which video appears next, which post rises to the top, which controversy receives another few seconds of our attention and which story remains buried beneath thousands of others.

This resembles what communication scholars have long called agenda-setting. The traditional media did not necessarily tell people what to think, but newspapers and broadcasters possessed considerable influence over what people were encouraged to think about.

Digital platforms have taken this much further.

The newspaper editor once selected a front page for millions of readers. The algorithm can now create millions of different front pages—one for each of us.

And this is where recommendation becomes political power.

Consider what happened around Romania's 2024 presidential election.

After the unexpected first-round success of candidate Călin Georgescu, concerns emerged about coordinated online activity and the amplification of political content on TikTok. Romania's Constitutional Court later annulled the election amid allegations of electoral manipulation and foreign interference. In December 2024, the European Commission opened formal proceedings against TikTok under the Digital Services Act, specifically examining, among other issues, risks connected to its recommender systems and the possibility of their coordinated manipulation. TikTok said it had worked to protect election integrity and cooperated with the investigation.

The point is not that an algorithm can simply “choose” a president.

The more unsettling point is that recommendation systems have become important enough to democratic life that regulators must now ask whether the architecture of a social-media feed can affect the conditions under which political opinions are formed.

Foucault would recognise this kind of power

Michel Foucault gives us a particularly useful way of understanding what is happening.

For Foucault, power is not limited to governments issuing orders or police enforcing laws. Modern power can be much quieter. It operates through institutions, classifications, norms and systems of knowledge that shape how people understand reality.

An algorithm rarely says:

Believe this.

Instead, it says:

Here is something you might like.

Then it says it again.

And again.

This is power without the appearance of command.

Nobody forces us to watch the next video. Yet the informational environment around us has already been organised. Some voices become more visible. Certain issues appear urgent. Some interpretations become familiar simply because we encounter them repeatedly.

The politics lies in the ordering of attention.

And attention is scarce.

What receives our attention has a better chance of entering our conversations, shaping our fears and eventually influencing what we demand from political institutions.

Habermas and the disappearing common world

Jürgen Habermas offers another warning.

Democracy, in his conception of the public sphere, depends upon citizens being able to encounter arguments, discuss common concerns and participate in forming public opinion.

That never meant everyone had to agree.

Democracy needs disagreement.

But meaningful disagreement requires something else first: enough of a shared world to disagree about.

Algorithmic personalisation complicates that.

Imagine two neighbours opening their phones over breakfast; one encounters videos about unemployment, housing prices and public healthcare, and the other sees immigration, crime and national security. Neither feed necessarily contains false information.

Yet after months of this, the two individuals may develop entirely different answers to the question: What is going wrong with our country?

This is why the political challenge of recommendation systems goes beyond “fake news.”

Even truthful information can distort our understanding of reality if certain truths are endlessly amplified while others remain practically invisible.

AI governance must govern attention, too

Governments are beginning to recognise the problem.

In October 2024, the European Commission asked YouTube, Snapchat and TikTok for detailed information about how their recommender systems operate and how they may amplify risks involving elections and civic discourse.

The scrutiny has continued. In January 2026, the Commission expanded its investigation into X's compliance with obligations relating to recommender-system risks. And in February 2026, it preliminarily found TikTok's addictive design—including its highly personalised recommendation system, infinite scrolling and autoplay—in breach of aspects of the Digital Services Act.

These developments point towards a larger question for AI governance.

We already ask whether AI systems are accurate, biased, transparent or safe.

We should also ask:

Who gets to organise our attention?

Who decides what “relevant” means? Why does one political story travel further than another? What incentives shape the systems deciding what billions of people encounter every morning? And should citizens have greater control over the algorithms constructing their informational worlds?

Because political power does not always arrive wearing the uniform of the state.

Sometimes it arrives as a recommendation.

You might also like this.

Watch next.

Suggested for you.

And perhaps that is precisely why it deserves our attention.

The most powerful algorithm may not be the one that tells us what to believe.

It may be the one that quietly decides what we will have the opportunity to think about at all.