Start here — this part is the same everywhere

Platforms use different words, but enforcement follows a shared shape. Before anything platform-specific, four things decide how well the next week goes.

1. Read the notice for the category, not the tone

Every notice cites something — a policy, a section, a standard. That citation is the entire frame the reviewer is working in. A response that doesn't engage it has nothing to land on, no matter how sincerely it's written. Find the noun before you write a word.

And check whether you've actually been accused of anything. "Failure to complete identity verification" is not an accusation — it's an unfinished task, and the answer is verification, not argument. People routinely write impassioned defences of conduct in reply to a notice that was never about their conduct.

2. Find out whether there's anywhere to argue

There's a case from one of our own accounts in the Meta section below.

3. Never open a replacement account

It is the most natural reaction and the most expensive. Creating accounts to work around an enforcement action is treated as circumventing enforcement across every major platform, and it converts a temporary, reversible problem into a much more serious one. It can burn the new account and the appeal together.

4. Preserve the evidence now

Screenshot the notice itself, any "how we made this decision" disclosure, the dates, and the specific content referenced. Export whatever data you can while you still have access — some platforms reserve the right to delete account data on termination, and an appeal can take longer than the data survives.

Meta: one review that counts

Meta runs at least two very different processes under similar warning language, and the notice does not tell you which one you're in.

The 30-second check

Open the restriction notice and tap Request review — but don't complete anything yet. Just look at what appears.

• A box to type an explanation → a content decision. What you write is the whole game.
• A request for phone, email code, or ID → an identity check. There is no appeal to file here. Complete it promptly; for many people that's the entire fix, at no cost.
• Nothing appears → a third state. Check Account Status separately; the option sometimes surfaces later.

A case from one of our accounts

On 3 August 2026 an automated system put a thirty-day link-sharing restriction on one of our Instagram accounts. Nothing specific was named — no post, no link.

We opened "How we made this decision," which almost nobody taps. It read: "Our technology found your activity doesn't follow our Community Standards. As a result, our technology took action." Automated finding, automated enforcement, no person at either step.

Then we tapped Request review. There was no text box — just an email code and a phone number. We completed the verification. Same day:

We had a full rebuttal drafted. We never used it. There was nowhere to put it.

That's the whole reason to check first. In the other branch — where there is a box — what goes in it is the only thing that can change it.

That case was a feature restriction. A disabled or suspended account follows a different path, below.

If you do get a text box

Meta's Business Help Center states that appeal decisions are final. Functionally, the in-app review is your one structured chance to be heard — a rushed or vague appeal doesn't get a do-over. Meta's Help Center indicates you generally have 180 days to appeal a disabled account; confirm that against the live Help Centre, as these pages change.

The explanation field is where it's won or lost: what happened, why the flagged activity was legitimate, and why it doesn't violate the specific policy cited.

And there are usually two rounds, not one. Round one is the in-app Request Review, which queues to fast automated triage. Round two is the email reply — when Meta writes back and you respond, that response tends to reach a human with your full file in front of them. Write both as a single case. If the app text and the email say slightly different things, a reviewer reading both sees an inconsistency, and inconsistency is exactly what a misrepresentation flag is looking for.

Why ordinary businesses get caught

The most common scam on these platforms is the account-recovery scam — anonymous accounts promising to restore disabled profiles, taking payment, harvesting passwords. Meta polices it hard, and should. But a legitimate small business can share enough surface features with that archetype to trip the same classifier: a shield in the logo, words like "account" or "secure", a link in bio, a service to sell. The things that actually distinguish a real business — a registered company, published terms, never asking for passwords — are the things an automated system reads worst.

Real-estate agents, therapists, coaches and other licensed professionals get caught disproportionately, because credentials in a bio and high-volume local posting both resemble patterns the classifiers hunt. There's a fuller treatment of that in our guide for licensed professionals disabled for misrepresentation.

If the appeal is denied

• Oversight Board (worldwide). After exhausting Meta's appeals, a reference ID may appear in your Support Inbox allowing submission to the independent Board. It selects only some eligible cases.
• Appeals Centre Europe (EU only). Under the Digital Services Act, ACE is a certified out-of-court body covering Facebook, Instagram and Threads. No user fee at present; the decision must be within the last six months; ACE aims to decide within 90 days. Decisions are non-binding, but platforms must engage in good faith.

Amazon: routing decides the outcome

Amazon is not one-shot. That's the good news, and it's why panic-submitting is the costliest thing you can do — you have more than one move, so each one should count.

What a Plan of Action actually is

Not an apology. A structured document with a shape investigators read for: root cause, corrective action, preventive measures — mapped to the specific Code of Conduct or policy section cited against you. Generic POAs get auto-rejected, and each rejection makes the next one harder.

The mistake almost no free advice mentions: misrouting

Amazon enforcement isn't one funnel. Sending the right document into the wrong process wastes days and makes a straightforward case look evasive.

• ASIN-level (a single listing pulled) — often resolvable with a solid root-cause POA plus supporting evidence.
• Account-level deactivation (Section 3 / Code of Conduct) — heavier, frequently multiple rounds and a full business-documentation package.
• Intellectual-property complaints — often not a POA problem at all. The real fix may be a retraction from the rights owner or a counter-notice: different track, different timeline.
• Account-health metric breaches (order-defect, late-shipment, valid-tracking, cancellation rates) — these call for a performance-improvement plan tied to your actual order data.

Diagnosing which state you're in before you write is the highest-leverage move available to you.

Channels that stay open

A revised Plan of Action through Account Health; an Account Health Specialist call ("Call Me Now"), where a specialist can tell you how your POA is landing; Executive Seller Relations as a later escalation; and Brand Registry / rights-owner cooperation for IP takedowns, where a retraction can resolve things faster than any appeal.

Start gathering this now

Supplier invoices and authorization letters (critical for authenticity and IP cases); product photos, packaging, compliance certificates; screenshots of the listing and the exact notice; recent order and performance history if a metric is involved. A submission that shows the fix with documents beats one that claims it.

Timelines vary widely by category and completeness. Treat any number you see online — including ours — as a rough shape, not a promise.

App stores: you get essentially one good submission

The single most important distinction here is appeal versus fix-and-resubmit. They are different remedies for different states, and choosing wrong burns the shot.

Google Play

Know which enforcement state you're in — a rejected update, a removed app, and a terminated developer account are not the same problem and don't share a remedy. Two facts worth having: Google states appeals receive replies in only four languages (Chinese, English, Japanese, Korean), and abusive use of the appeal channel can cost you email-support eligibility.

A myth worth retiring: the widely repeated "1,000-character limit" on Google's appeal field appears on no official Google page we could find — it traces back to blogs and forum posts. Don't build a strategy on folklore.

A freshness trap: Google has reorganized and renamed policy sections. The old "Minimum Functionality" now sits as "Limited Functionality and Content" inside Functionality, Content, and User Experience, and several long-standing policy URLs redirect. Citing a stale section name tells the reviewer you're working from old blog posts rather than their current policy.

Apple

Apple's process is more conversational, which is an advantage if you use it in the right order.

Step one is not an appeal. The rejection in App Store Connect cites the specific guideline, and you can reply directly to the App Review team — with screenshots and supporting documents — from the App Review section of your app's page. The reply field allows 4,000 characters, and metadata-only rejections can be resubmitted once fixed. Many rejections resolve here; Apple states that on average over 40% of unresolved issues relate to Guideline 2.1 (App Completeness) — crashes, placeholder content, broken URLs, missing demo credentials.

Step two, if you genuinely disagree, is the App Review Board appeal — for when the app's concept or functionality was misunderstood. That's where the one-appeal rule bites, so it should carry your best guideline-specific argument, not your first draft. Apple also offers expedited review for extenuating circumstances and a bug-fix submission path.

What both platforms respond to

• Correct routing — appeal, resubmission, correspondence or escalation, matched to your actual state.
• A section-tied rebuttal against the current text of the policy cited, element by element — not a general plea about your livelihood.
• Evidence over assertion — a corrected build, screenshots, logs, licenses, and a clear account of what changed.
• Restraint — no repeat submissions of the same argument, no channel-flooding.

AI accounts: the leverage was decided on day one

If client delivery, support or code runs through an AI account, that account is infrastructure — and providers reserve broad discretion over it.

Each provider opens a different door

• Anthropic (Claude) — you appeal from inside the banned account. The Help Center directs you to log in with that account and complete the appeal form, and states you must be logged in to reach it. Separately, if an organization you belong to is on hold, the restricted screen carries a distinct "Request a review" button.
• OpenAI — use the link in the email notification, because it routes the request accurately. The general intake form is the fallback; if you got no email and can't log in, Help Center chat is the route.
• Google (Gemini Apps) — appeal via the link in the restriction notice. There's no standalone public form to find.

The pattern: the per-incident link in your notice carries routing information a generic form does not. Some providers publish no consumer appeal route at all — worth establishing early rather than losing a day looking.

Send what they asked for

OpenAI publishes the fields it wants: the User ID and Org ID, relevant context about your usage, the date range of any suspected compromise, card details if there were unauthorized charges, and steps already taken. Most appeals contain none of that — they contain a paragraph explaining the sender is a legitimate business. A reviewer can act on an Org ID; they can't act on that paragraph.

Export first, appeal second

Anthropic's Consumer Terms state that on termination it "may at our option delete any Materials or other data associated with your Account." Its Help Center is more encouraging in practice — banned Free, Pro and Max users can still log in to export data or delete the account, though what's accessible may be restricted depending on the violation. Read those together and the order is obvious. The appeal may take weeks; the data may not wait.

The part decided before anything happened

Anthropic's consumer terms are the outlier, in the customer's favour: Section 13 routes disputes to the state or federal courts in San Francisco under California law, with no arbitration clause and no class-action waiver. Its Commercial Terms — API and Console — are a different document and do arbitrate, with jury-trial and class waivers.

OpenAI's consumer terms take the more common path: binding individual arbitration with a class-action waiver, plus an opt-out window described in current summaries as 30 days from account creation or from an update to those terms. Confirm against the live terms for your own account — this is exactly the kind of clause that gets revised.

One more, before asking for a refund: Anthropic's terms state that if it terminates your access for a violation and you have a Subscription, you are not entitled to a refund.

More detail, by situation

One honest caveat

Reinstatement is the platform's decision in every case above — ours is never a promise and nobody else's should be either. What you control is whether your one submission is routed correctly, cites the current policy accurately, and proves its case with evidence. That is the part worth doing properly, and the part this guide can only take you so far into.