Is ChatGPT GDPR Compliant? What Businesses Need to Do

Somewhere in nearly every small business, someone is using ChatGPT. Often on a personal account, often with client data, rarely with a policy. That’s not an accusation. Just how it goes: the tool arrives before anyone thinks about data protection.

The good news is that working with ChatGPT in a GDPR-friendly way isn’t a major project. It takes a handful of decisions you make once.

Is ChatGPT GDPR compliant?

The question is framed slightly wrong, which is why answers online contradict each other. No tool is compliant or non-compliant on its own. Under the GDPR, what counts is which data you put in, on which plan, and under which rules. The UK GDPR follows the same logic.

Three situations help sort it out:

Criterion	Personal account	Business plan with DPA
Text with no personal data
Pseudonymised content ([client], [company])	partly
Client names, emails, order details		partly
Health, HR or applicant data
Decisions about individuals

Our assessment based on the GDPR and German regulators' guidance. "Partly" means only where the purpose requires it and it's documented.

The top row is the everyday case. Sales blurbs, outlines, translations with no names in them: no data protection issue arises. Trouble starts when information about identifiable people enters the chat.

What regulators expect

German supervisory authorities set out their expectations in a joint guidance paper on AI and data protection in May 2024. It’s a useful reference beyond Germany because it’s concrete. Four points matter most for office use.

Employers should provide work accounts and devices, so staff don’t have to use personal accounts that build profiles about them. Settings should be chosen from the start so inputs aren’t used for training. Outputs must be checked for accuracy. And no decision with legal effect on a person may be left to the AI alone, which GDPR Article 22 generally prohibits. The paper’s example: an AI that screens job applications and sends interview invitations by itself.

Then come internal rules: what the tool may be used for, which data is off limits, who’s responsible.

The settings you should know

On personal plans, OpenAI may use your conversations to train its models unless you opt out. The toggle is called “Improve the model for everyone” and sits under Data Controls in settings. According to OpenAI’s help centre, it applies to new conversations on every device linked to the account.

Two details in the help pages often get missed. Giving a thumbs up or down on an answer can make the whole conversation eligible for training, even with the toggle off. And temporary chats aren’t used for training and, per OpenAI, are deleted after 30 days; save one, though, and it becomes a regular chat.

For ChatGPT Business, Enterprise and the API, OpenAI says it doesn’t use inputs or outputs for training by default. Just as important, only these plans come with a data processing agreement under GDPR Article 28.

Why the business plan makes the difference

When a service provider processes your clients’ data on your behalf, you need a processing agreement. It commits the vendor to use the data only on your instructions, protect it and delete it. Without one, sending client data to ChatGPT is hard to justify.

OpenAI lists ChatGPT Business at $20 per person per month billed annually, or $25 billed monthly (as of October 2026). For a small team that’s modest next to the cost of a data breach. Our ChatGPT vs Claude comparison covers how the two vendors’ plans differ. If you’re on Microsoft 365 already, look at Copilot as an alternative, since Copilot Chat often comes with enterprise data protection at no extra cost.

The paperwork that goes with it

The agreement alone isn’t the whole job. If you process client data through ChatGPT, it belongs in your record of processing activities under GDPR Article 30. Sounds bureaucratic. Really it’s five lines: purpose, data types, recipients, retention, safeguards.

Your privacy notice should list the vendor among recipients. And because OpenAI is a US company, the contract should state what transfers to the US rely on, such as the EU-US Data Privacy Framework or standard contractual clauses.

Full data protection impact assessments under Article 35 usually aren’t needed for occasionally drafting emails. It’s a different story if you systematically process lots of client data or the AI evaluates people.

What never goes into the chat

Whatever the plan, some data we’d keep out entirely:

• Health data such as sick notes, diagnoses, medical letters. GDPR Article 9 treats these as a special category.
• Job applications and personnel files.
• Login details, passwords, API keys.
• Client trade secrets shared with you in confidence.
• Anything covered by professional confidentiality, as with doctors, lawyers or accountants.

Regulated professions face rules beyond the GDPR. If you work in one, check with your professional body or data protection officer before using any AI tool with client material.

The Italian fine

Regulators are paying attention, as a case from Italy shows. The Italian authority, the Garante, fined OpenAI 15 million euros in December 2024. Its reasons: training on personal data without an adequate legal basis, breaches of transparency obligations, and a data breach from March 2023 that wasn’t reported. OpenAI called the decision disproportionate and challenged it in court. It won: according to Italian press reports, the civil court in Rome annulled the decision in March 2026. The reason was procedural. In the court’s view the lead authority under the GDPR one-stop-shop was Ireland’s regulator, where OpenAI has its EU establishment, not the Garante.

That doesn’t create a direct obligation for you as a user. It does show that the legal basis question isn’t theoretical, even though this particular fine didn’t stick.

A five-line team policy

An AI policy doesn’t need 20 pages. For a small team, one paragraph everyone knows will do:

1. For work we only use [tool] with a company account, never personal accounts.
2. We replace client names and contact details with placeholders
   when the task doesn't need them.
3. We never enter health, HR or applicant data.
4. Every output is checked before it leaves the company.
5. Questions go to [name].

Those five lines also cover part of the EU AI Act’s requirement to support AI literacy among staff. What that obligation actually asks for is explained in our piece on AI Act AI literacy.

One question before every prompt

Before you hit enter, ask: would I send this text by unencrypted email to a stranger’s company? If not, it doesn’t belong in a personal AI account. If yes, you’re usually fine.

That’s no substitute for proper advice, and in sensitive sectors a conversation with your data protection officer is worth having. For daily work, though, that one question catches most mistakes.

Companies that want AI built into their own processes, with access to client data in a controlled environment, need an architecture where security is designed in from day one. Studios like OnlineBuilders take on that kind of project.

Your next step takes two minutes: open ChatGPT’s settings, go to Data Controls and check whether “Improve the model for everyone” is switched off. Other risks and limits are covered under privacy and limits.

Questions people ask

Is switching off training in settings enough?

For personal use it's a sensible step. For client data in a business it isn't, because personal plans come without a data processing agreement. You need a Business or Enterprise plan for that.

Can I type client names into ChatGPT?

Not on a personal account. On a business plan with an agreement in place it's defensible if the task genuinely needs them. Often it doesn't: placeholders like [client] work for most jobs.

Has ChatGPT ever been fined over privacy?

Italy's data protection authority fined OpenAI 15 million euros in December 2024, citing among other things the lack of a legal basis for training. According to Italian press reports, the civil court in Rome set the decision aside in March 2026 because it found the authority lacked jurisdiction. The court did not rule on the underlying allegations.

Do I have to tell clients I use ChatGPT?

If you process their personal data with it, your privacy notice should list the vendor as a recipient. For drafting text that contains no personal data, there's nothing to disclose under the GDPR.