Platform ban case studies: what a resolution actually settles, and what it never does

The case studies circulating about account enforcement are almost all one of two things: a story from someone who got their account back, or a story from someone who did not. Neither is evidence, because you never see the accounts that did the same thing and got the opposite result.

The useful material is the shape of the process, the few matters a court has decided in public, and the write-up of your own incident. This page covers all three. It names a matter only where a court has published its reasoning, because an allegation attached to a name outlives every correction.

What to take away

Why the published ones do not transfer

Four defects, and most examples have all four.

Survivorship. The people who write these up are disproportionately the ones with a good ending. The base rate is invisible.

Missing variables. Outcomes are graded by account history, surface, reach, category and how the item was found. A write-up that omits those is describing a different account from yours. Which variables matter is set out in what moves the outcome.

Attribution. An account came back after somebody did five things. Which of the five mattered is unknowable, and the appeal may simply have reached a reviewer who read it properly.

Age. Rules, thresholds and routes change without announcement, so a write-up describes a system that may no longer exist.

None of that makes them worthless. It makes them useful for recognizing a shape and useless as a method.

How a matter proceeds

A decision is made, usually by an automated system, in bulk, with no human having looked at the item.

A notice is issued, or is not. Where one exists it typically names a policy area rather than a clause, and it may carry a reference number and a deadline.

An internal review runs, if you ask for one. This is not a hearing and not adversarial. It re-runs a classification against the rule as written, often quickly.

An outcome is returned, usually with no reasoning. A short turnaround does not tell you whether a person looked.

An external route may exist, where a service offers one for a market. It sits after the internal appeal rather than instead of it, and has its own eligibility rules published on its own site. Routes of that kind exist because of market regimes such as the European Union's Digital Services Act (DSA), which gives users the right to turn to out-of-court dispute settlement bodies if they think a platform's decision was wrong. Those bodies are certified by the Digital Services Coordinator of their country of establishment, cannot impose a binding settlement, and the regime applies within its own scope only.

The matter ends, which for the great majority means it stops rather than concludes. Nothing is published, nothing is explained, and there is no finding of any kind.

Knowing that shape stops you spending effort in the wrong place. A great deal of energy goes into arguing a case that no stage of this process is designed to hear.

What a resolution does and does not establish

Where a dispute produces a formal outcome, read what the document actually says.

An agreement to resolve establishes that a dispute existed and has ended, and a general explanation of settlement sets out how little that is in one legal system. A clause recording no admission means exactly what it says and supports nothing about the underlying conduct. A payment establishes that money moved on agreed terms, not that harm was measured.

A commitment to change something establishes that an obligation now exists, not that the earlier position was unlawful. Confidential terms mean you cannot know what was agreed, and that is not an inference in either direction.

Parties settle for cost, time, distraction and certainty, including matters they would probably win. Reading an outcome backwards into a verdict is the standard error. The document-by-document version of this, for disputes about the terms themselves, is in what the public record supports.

Two court matters and what their endings settled

Murthy v. Missouri (United States, 2024). State and individual plaintiffs sued federal officials and sought to stop them from inducing platforms to remove, suspend or otherwise act against speakers. The Supreme Court's opinion describes how platforms may suspend or ban users who frequently post content that violates platform policies, and one plaintiff's brother had his Twitter account suspended. The case was decided on June 26, 2024, and the Supreme Court held that neither plaintiff group had standing: neither the individual nor the state plaintiffs had established Article III standing to seek an injunction against any defendant. That ending settles who may sue and on what showing. It says nothing about whether any suspension was right, which is the point of reading the holding before the headline.

Facebook account suspensions (Germany, 2021). The users in this matter had their Facebook accounts partly suspended for a period and their comments deleted. The III. Civil Senate of Germany's Federal Court of Justice held that Facebook's terms of 19 April 2018 on removing posts and suspending accounts were invalid under § 307(1) sentence 1 of the German Civil Code (BGB), the standard-terms rule, at least because the provider did not commit to tell the user about a removal afterwards and about an intended suspension in advance, give the reason, and offer a chance to respond followed by a new decision. Where a post was deleted and the account partly suspended under those terms, the user could require the post to be restored. That ending settles a procedural duty under German law. It does not say the posts were acceptable, and it does not carry over to another service's terms or another country.

Both matters show the same thing as a private settlement: the outcome answers a narrower question than the one the parties started with.

The write-up worth having

Written the same week, while people still remember, and kept where the next person will find it.

Two things to keep out: blame, and a confident causal claim about what worked. Both make the document unusable by the people who need it most.

De-identifying it

If the write-up will circulate beyond the people involved, strip the account handles, the individual reviewer references, and the names of colleagues, and describe roles instead. Where a customer or a member of the public appears, remove anything identifying them. That includes the quoted text of their posts, which is identifying in practice even when the name is gone.

Keep the service unnamed too, unless the audience needs it. It rarely changes the lesson and it changes how the document can be shared.

What to do with it a year later

Reread it when the next incident happens and during your periodic checks. The value compounds.

Three write-ups show where the line sits for your account, unlike any published policy. They also show whether you made the changes you promised last time. What to change is usually already listed in the risks around an incident and in the tooling that genuinely helps.

Nothing here states what any service's process currently is, and none of it is legal advice. Routes, deadlines and eligibility change, so take those from the service's own current pages. Take anything involving money, employment or a legal order to someone qualified where you are. The outcomes themselves and their reversibility are set out in the eight outcomes.

Common questions

Can I use someone else's write-up to draft my appeal?

Use it for structure, not for content. Copying an argument that fitted a different account and a different policy area is how appeals get spent on the wrong point.

Should we publish ours?

Rarely worth it while the matter is live, and it commits you to a version of events before the facts are settled. Afterwards, a de-identified version can be genuinely useful to others.

How many of these do we need before there is a pattern?

More than you will have. Treat a small set as prompts for questions rather than as evidence, and pair it with the symptom-level view in fifteen situations.