The short answer: Comments are personal data, so moderating them is processing under GDPR. You need a lawful basis, usually legitimate interests. You are a joint controller with Meta for Page Insights, but a controller in your own right for what you do with comments. Any moderation tool you use is your processor and needs a contract.
This is written for page owners, not lawyers, and it is not legal advice. If your page carries real risk, as a public body, a publisher or a campaign does, get a qualified opinion.
Are comments really personal data?
Yes. A comment carries a display name, a profile link, often a photo, a timestamp and an opinion attached to an identifiable person. That is personal data under Article 4 of the GDPR, and reading, storing, scoring, hiding or replying to it all count as processing.
Some comments go further. People volunteer their politics, religion or health in comment threads every day, especially under news and political posts. Those are special categories under Article 9. Article 9(1) bans processing them unless a condition in Article 9(2) applies, and that condition is needed on top of your Article 6 basis. In practice page owners lean on Article 9(2)(e), data the person manifestly made public by posting it, and on Article 9(2)(f) where you keep something for a legal claim. Note the condition, do not assume it, and do not build anything that sorts commenters by those traits.
Are you a joint controller with Meta?
For page statistics, yes. In Wirtschaftsakademie Schleswig-Holstein (C-210/16, 5 June 2018) the Court of Justice held that a Facebook fan page administrator is jointly responsible with Facebook for visitor data, because setting the page's parameters helps determine why and how that data is processed. Fashion ID (C-40/17, 29 July 2019) confirmed the pattern for an embedded Like button and set the limit: you are a joint controller for the stages you actually influence, not for everything the platform does afterwards.
Meta puts this into a contract. Its Page Insights addendum names you and Meta Ireland as joint controllers under Article 26 for Insights events, says Meta takes primary responsibility for transparency, data subject rights and breach notification there, and leaves you three jobs: confirm your lawful basis, name yourself as a controller with contact details in your privacy information, and forward any Insights request or authority inquiry to Meta within seven calendar days.
The addendum covers Insights only. When you copy comment text into a moderation tool, a spreadsheet or a CRM, you are the controller for that processing and nobody shares the responsibility with you.
| Processing | Meta's role | Your role | Your tool's role |
|---|---|---|---|
| Page and post statistics (Insights) | Joint controller, primary responsibility | Joint controller: own lawful basis and privacy notice, forward requests within 7 days | None |
| Comments on the platform | Controller for its own platform processing | Controller for your moderation decisions | None |
| Comments pulled into a moderation tool | None | Controller: you set the rules and the purpose | Processor, acting on your instructions |
| A commenter asks what you hold | Answers for platform data | Answers for your tool and your records | Helps you export it |
| Hiding, replying, blocking | Executes the action | Decides and is accountable for it | Applies and logs it |
What is your lawful basis for moderating?
For most businesses, legitimate interests under Article 6(1)(f) fits. You have a real interest in keeping your comment section free of spam, scams, harassment and illegal content, and commenters expect a moderated brand page. Write the balancing test down rather than assuming it: the interest, why moderation is necessary for it, and why it does not override the commenter's rights. Proportionality helps: hiding is lighter than deleting, reversible, and leaves the evidence intact.
Two situations need more care. Public authorities cannot rely on legitimate interests for processing carried out in the performance of their tasks, which matters for ministries, councils and state agencies running pages; they need another basis, often a public task under Article 6(1)(e). Where a law requires you to remove content, a legal obligation basis may apply instead. Our guide to whether you are legally responsible for comments on your Page covers liability.
What changes when you use a moderation tool?
A tool that reads your comments and applies your rules acts on your instructions. That makes it a processor and you the controller, and Article 28 requires a written contract, usually a data processing agreement. Before you connect anything, check these points:
- A real DPA. It should bind the vendor to process only on your instructions, keep staff under confidentiality, help with data subject requests, and delete or return data when you leave.
- The sub-processor list. AI moderation involves at least one model provider. Ask who they are and whether you get notice before the list changes.
- Where the data sits. Hosting region, error monitoring region, and whether anything leaves the EU. If it does, ask which transfer mechanism covers it.
- Retention. How long comment text, author names and AI reasoning are kept, and whether deletion is actually deletion.
- Whether your comments train anyone's model. Ask directly and get it in writing.
- What the audit trail records. You need to show what was hidden, when, and on which rule.
Be skeptical of claims a vendor cannot back up: ISO 27001 or SOC 2 mean something only if they can hand you the certificate or report.
What about comments that expose other people?
Comment threads leak other people's data constantly. Someone posts a rival's home address, a screenshot of a private message, a photo of an employee, an order number, a phone number, or an accusation that names a specific person. Once that is on your page and you have seen it, doing nothing is itself a choice.
Handle it as its own rule category, not as general rudeness. Hide on sight anything with contact details, addresses, document numbers or screenshots of private conversations, and say so in your comment policy so the decision looks consistent rather than political. Hiding beats deleting here: the comment survives as evidence if the exposed person asks you for proof. For the differences between the actions, see hide vs delete vs restrict vs block.
How do you handle access and erasure requests?
Someone can ask what you hold about them, and ask you to erase it. For a page owner that splits three ways.
- Insights data. Forward it to Meta within seven calendar days, per the addendum. Do not answer it yourself.
- The comment on the platform. The commenter can usually delete their own comment. If they ask you to remove it, removing it is normally simplest.
- What your tool holds. This one is yours: be able to search by author, export what you have, and delete it.
Erasure is not absolute. If you keep a hidden comment as evidence of harassment or for a possible legal claim, Article 17(3)(e) gives you ground to stand on. Keep it narrow and time limited, and tell the person why. When you do delete, a minimal record of the decision, which rule applied and when, is usually defensible.
Do you have to keep records?
Probably, yes. Article 30(5) exempts organizations with fewer than 250 staff, but not where the processing is more than occasional, risks people's rights and freedoms, or involves special category data. A moderated comment section is continuous. So write a short record of processing for comments: what you collect, why, your lawful basis, who it is shared with, and how long you keep it. One page is usually enough, and it makes the privacy information easy to write: that you moderate comments, on what grounds, what happens to hidden ones, and a link to Meta's addendum.
Does the Digital Services Act apply to you?
Usually not in the way people fear. The DSA tiers its duties. A few basic ones, such as a point of contact and clear terms, apply to every intermediary service. Notice and action and statements of reasons apply to hosting providers. Internal complaint handling and the rest of the online platform duties apply to platforms, and Article 19 exempts providers that are micro or small enterprises from those. On Facebook, Instagram, TikTok and YouTube that provider is the platform, not you: you are a business user posting on it. Your own website is different. Storing readers' comments there can make you a hosting service provider in your own right, so notice and action and statements of reasons apply whatever your size, while the platform level duties usually fall away under that exemption. Our piece on how news publishers moderate Facebook comments covers the newsroom side of this.
A short compliance checklist
- Publish a comment policy, apply it the same way to supporters and critics, and write down your lawful basis and the balancing test behind it.
- Name yourself as a controller in your privacy information and link Meta's addendum.
- Sign a DPA with any moderation vendor and read the sub-processor list.
- Prefer hiding over deleting so decisions stay reversible and evidence survives.
- Keep an audit trail of what was actioned, when and under which rule, and know your route for requests: Meta for Insights, you for everything else.
Most of this is ordinary record keeping rather than legal engineering. If you want the record kept for you, Social Moderator AI is run by an EU company in Slovakia and acts as a processor for your comment and review content: rules are written in plain language, every AI action stores a written reason, and the audit log shows who or what did what and when. Analysis runs on Google Gemini, with Anthropic where configured; YouTube comment data is deleted or redacted after 30 days under YouTube's policy; error monitoring data stays in the EU. Facebook's keyword filters and Instagram's Hidden Words are free and keep the data inside Meta, which is a fair answer if your volume is low.
Frequently asked questions
Do I need consent from commenters to moderate their comments?
Normally no. Consent is a poor fit, since you cannot stop moderating if someone refuses. Most businesses rely on legitimate interests and document the balancing test.
Can I keep a hidden comment forever as evidence?
Keep it as long as you have a reason, not indefinitely by default. Set a retention period that matches the reason, for example the limitation period for a claim, and delete on schedule.
Is an AI moderation tool an automated decision under Article 22?
Article 22 concerns decisions with legal or similarly significant effects on a person. Hiding a comment is generally not in that bracket, but keep human review available, keep the reasoning visible, and let people appeal.
Does my tool need to be based in the EU?
No. Non-EU processors are allowed with a valid transfer mechanism and the right contract terms. An EU provider simply removes one question from the list.
Who answers if a commenter complains to a data protection authority?
Meta Platforms Ireland answers for the Insights processing it leads, under the Irish Data Protection Commission as its lead supervisory authority. That does not put you out of reach: as a joint controller with no establishment in Ireland, your own national authority can investigate your share, which is what happened in Wirtschaftsakademie. For your own moderation records and anything in your tool, you answer outright.