Who decides what comes down, and what you are owed when it does.
This is the first piece in our Who Decides? series.
It does not always disappear. Sometimes a post is simply there and yet nobody finds it. The numbers drop, the replies stop. And you are left with a question you cannot answer: did I do something wrong, or am I imagining it?
That uncertainty is built into the process. Decisions get made about you, and nobody tells you they were made.
What follows is what actually happens when content comes down in Europe: who takes the decision, and what the law says you are entitled to know.
The claim
Platforms give two very different reasons for removing something, and they are easy to confuse on purpose.
The first is: this is illegal. Somewhere a law was broken, and the platform is acting on it.
The second is: this breaks our rules. No law is involved. A company wrote a policy, and your post fell outside it.
They arrive in the same grey notification box, but they are not the same thing. One is enforcement of public law. The other is enforcement of a private rulebook that the company wrote and can rewrite tomorrow.
The European Commission runs a public database of these decisions, and the pattern in it is worth sitting with. The ground platforms state most often is their own rulebook. The label the database uses is “Other violation of provider’s terms and conditions.” Most of what comes down, comes down under private rules.
Who decides?
Almost half the time, nobody does. Not in the sense of a person reading your post and forming a view.
In the six months to 8 August 2026, 363 active platforms filed more than 3.47 billion statements of reasons to the EU’s public DSA Transparency Database, and 42 percent of the decisions they reported were fully automated. Those are not running totals since the law took effect. The database keeps a rolling six-month window and updates daily.
Behind any single removal there may be a classifier that flagged it, a subcontracted moderator with seconds per case, a “trusted flagger” that reported it, a state authority that ordered it, or a policy team that wrote the rule years ago and never met your situation. In most cases there is no name attached to the decision, because there was no decision-maker in the way we normally mean the word.
There is a reasonable defence for this. At the volume platforms handle, humans cannot read everything. But speed is not the same as being right. An automated system can be fast, consistent and confidently wrong about a piece of documentation, a joke, a quotation, or a photograph of something that needed to be recorded.
By what rule?
Under the Digital Services Act, a platform that restricts your content owes you a statement of reasons, and Article 17(3) is unusually specific about what it must contain:
- what kind of restriction it is, including its territorial scope and duration;
- the facts and circumstances the decision relied on;
- whether automated means were used;
- the legal ground, if the claim is that the content is illegal, or the contractual ground, if the claim is that it broke the terms;
- information about your redress options.
Read that list again, then compare it to the last notification you received. “Your content violated our Community Standards” satisfies almost none of it.
One detail matters more than the rest, and most English-language coverage skips it: Article 17(1)(a) explicitly names demoting content as a restriction that triggers this duty. You do not have to be deleted to be owed an explanation. If your reach was cut, that is a decision, and it is a decision you are entitled to have explained.
Who pays the price?
Three groups, unevenly.
People judged by machines. Nearly half of these decisions arrive without human review. The cost of an automated error is not borne by the system that made it.
People outside Europe. This is the part that has to be said plainly, or this becomes an advertisement for Europe. Article 2(1) states that the DSA applies to recipients of the service located in the Union. The same classifier, trained on the same data, moderates a user in Lagos, Manila or São Paulo, and they are owed none of this. The right to be told why you were silenced should not depend on where you happen to live. At the moment, it does.
People documenting harm. Footage of violence is, to an automated system, violent content. Evidence and atrocity look identical to a classifier. Archives of abuses have been erased by systems working exactly as designed.
What does it mean for you?
First, work out what actually happened. Removal, demotion and a technical fault look the same from the inside, and the remedies differ. If your reach fell without notice, that is not automatically a moderation decision, but it may be one, and you are entitled to ask.
Then, if you are in the EU, you have two routes.
Internal appeal: Article 20. Free, electronic, and open for at least six months from the day you were notified. Platforms must handle complaints “in a timely, non-discriminatory, diligent and non-arbitrary manner.”
Out-of-court dispute settlement: Article 21. A certified body reviews the case. It must decide within 90 days, or up to 180 for complex disputes. If you win, the platform pays the fees. If you lose, you pay nothing, unless you acted in manifest bad faith.
One exception worth knowing: Article 19 exempts micro and small enterprises from the section containing the internal appeal duty. They still owe you a statement of reasons under Article 17. That obligation sits elsewhere in the regulation.
And a warning about the transparency database: under Article 24(5) platforms have to make sure the statements they submit contain no personal data, so it never goes in. You will not find your own case there. It is built for studying patterns. You cannot look yourself up in it.
Is there another way?
The European framework is a real improvement on nothing, and it is worth saying so. But two gaps in it are worth naming, because both were choices.
Article 20 sets no hard deadline. “Timely” is not a number of days. A complaint can be handled slowly and still comply.
The appeal body cannot bind anyone. Certified out-of-court dispute settlement bodies have no power to impose a settlement on the parties. So there is a review process, and it has no teeth. You can win and still lose.
Less invasive alternatives exist and are not exotic: temporary demotion instead of deletion, where the harm is disputed; preservation of removed material as evidence rather than erasure; mandatory human review above a defined threshold of impact; and appeal decisions that actually bind the platform that lost.
None of this is an argument that nothing should ever come down. Some content is genuinely illegal, and some rules are reasonable. It is an argument about who gets to decide, and whether they have to explain themselves. We have grown used to a handful of companies deciding what we see and what we may say. Most of us only start asking how the system works at the moment it grinds us.
So it is worth asking now, before it does: who decides what you are allowed to see and say online, and who checks them?
Written by Pavel Kadlec. Sources: Regulation (EU) 2022/2065 (Digital Services Act), full text at EUR-Lex, CELEX 32022R2065; DSA Transparency Database. Database figures retrieved 8 August 2026.
