Can I Put Customer Data Into ChatGPT? The GDPR Answer

If your staff are using a convenient online tool to rewrite emails, summarise notes or prepare client documents, uncertainty about privacy is completely normal. The important next step is not to ban every use or assume every use is safe. It is to understand when customer information becomes personal data, what checks the GDPR framework points to, and how to create a rule your team can actually follow.

Start with the real business situation

Consider an operations manager at a 15-person professional services firm. Before any policy exists, staff copy full client emails and case notes into ChatGPT to make them shorter, clearer or easier to summarise. Names, contact details, commercial disputes and confidential instructions travel with the text because removing them feels like extra work.

After a basic control is introduced, staff first remove names, account numbers, addresses and other details that are not needed. Low-risk drafting can use fictional or generalised facts, while work involving identifiable records goes through an approved business workspace and a documented review. The useful change is not simply buying a different subscription. It is deciding which information is necessary for the task and controlling the route it takes.

What counts as customer personal data?

Under the GDPR, personal data is information relating to an identified or identifiable person. That includes obvious details such as a name, email address, telephone number or customer number, but it can also include a combination of facts that makes someone identifiable. The official definition appears in Article 4 of the GDPR.

A client company name is not always personal data by itself, but the name of its director, a sole trader's account, an employee complaint or a contact person's purchasing history may be. Pseudonymising a record by replacing a name with a code can reduce risk, although it remains personal data if your business can reconnect the code to the individual.

The GDPR question is about processing, not just storage

Pasting customer information into ChatGPT is a form of processing and disclosure. Your business is usually deciding why the information is being used and what result is wanted, which commonly places it in the controller role. The AI provider may process submitted business data under the service contract, but the exact allocation of roles depends on the product and the relevant terms.

This is why a simple statement such as “the data is encrypted” does not settle the GDPR question. Encryption is an important security control, but the GDPR also addresses lawfulness, fairness, transparency, purpose limitation, data minimisation, accuracy, retention, security and accountability. Those principles are set out in Article 5.

You need a lawful basis, but consent is not the automatic answer

The GDPR lists several possible lawful bases in Article 6. Depending on the circumstances, a business might examine consent, performance of a contract, compliance with a legal obligation, protection of vital interests, a public task or legitimate interests. Choosing one is not a box-ticking exercise. The purpose, necessity and effect on the individual matter.

Consent can be difficult in an ongoing customer relationship because it should be specific, informed, freely given and withdrawable. Legitimate interests also requires more than saying the use is convenient. The EDPB explains that an organisation considering legitimate interests should identify the interest, test whether the processing is necessary and balance that interest against the person's rights and reasonable expectations. Its AI model opinion also stresses that the assessment is case specific. See the EDPB Opinion 28/2024.

For a small business, the useful question is: could the task be completed without identifiable data, or with less data? If yes, sending the full record is difficult to justify as necessary. Removing unnecessary details is usually a better first move than searching for a broader legal argument.

Sensitive and confidential information needs a higher barrier

Some information receives additional protection under Article 9 of the GDPR. This includes data revealing racial or ethnic origin, political opinions, religious beliefs or trade union membership, as well as genetic data, biometric data used for identification, health data and information about a person's sex life or sexual orientation. Criminal conviction and offence data is separately addressed by Article 10.

Health notes, employee grievances, identity documents, financial hardship details and legal case material should not be treated as ordinary drafting content. Even where a lawful route may exist, professional secrecy, sector rules, contractual confidentiality and the possible harm from disclosure can make the risk much higher. A business should route these uses to its data protection lead, legal adviser or relevant professional body rather than letting an individual employee decide in the moment.

The processor contract matters

Where a supplier processes personal data on your behalf, Article 28 of the GDPR describes the contractual safeguards expected between controller and processor. Businesses commonly look for a data processing agreement, instructions for processing, confidentiality commitments, security measures, rules for subprocessors, assistance with rights requests, deletion or return provisions, and audit information.

OpenAI states that it can enter into a Data Processing Addendum for ChatGPT Business, ChatGPT Enterprise and the API. Its current Data Processing Addendum sets out processing terms, subprocessors and transfer mechanisms. That is materially different from allowing employees to use unmanaged personal accounts under consumer terms.

A DPA supports a business review, but it does not certify that every proposed use is lawful. Your organisation still needs to decide what data staff may submit, whether the use is necessary, how access is managed, what retention setting applies, and how the activity is explained in your privacy information where relevant.

Does ChatGPT train on customer data?

The answer depends on the service and settings. OpenAI states that business data submitted through ChatGPT Business, ChatGPT Enterprise and the API is not used to train its models by default. Its enterprise privacy page also describes business security and privacy controls.

For individual ChatGPT services, OpenAI provides data controls that allow users to choose whether content may be used to improve models. Its EU services privacy policy explains those controls and other processing. A consumer opt-out is useful, but it is not a substitute for organisational approval, a suitable contract or an assessment of the data being disclosed.

What about transfers outside the EEA?

GDPR Chapter V applies when personal data is transferred to a country outside the European Economic Area. Article 44 and the following provisions describe routes such as an adequacy decision or appropriate safeguards. Standard Contractual Clauses are one widely used safeguard, but organisations may also need to consider the circumstances of the transfer and any supplementary measures.

OpenAI's DPA incorporates transfer provisions, including the European Commission's Standard Contractual Clauses where they apply. OpenAI also offers European data residency options for eligible business customers and services. Its announcement states that eligible API customers can choose European processing for supported endpoints, while eligible ChatGPT business offerings can support European storage options. Review the exact product scope because “European data residency” does not automatically mean that every service function, support activity or subprocessor remains exclusively within the EEA. See OpenAI's European data residency information.

A practical yes, no and review framework

This table is a planning aid, not a compliance determination. Context can move an example from one column to another. A seemingly harmless paragraph may still identify a person when combined with details already known to the recipient or contained elsewhere in the prompt.

What your business can do now

1. Find out what staff are already doing. Ask which AI tools they use, which tasks they perform and what information they paste or upload. A short, non-punitive survey usually produces better information than announcing a ban first.

2. Create a prohibited-data list. Name the categories that should not enter an unapproved AI tool, such as passwords, payment data, identity documents, special-category data, legal advice, confidential investigations and full customer records. Give staff examples from their actual work.

3. Prefer de-identification and synthetic examples. Remove names, addresses, reference numbers and unnecessary background. Replace real facts with placeholders where the task is only about style, structure or wording.

4. Select an approved business service. Review the contract, DPA, subprocessor list, retention controls, access management, logging, security information and transfer mechanism. Do not assume a paid plan is automatically suitable merely because it is labelled for business.

5. Record the purpose and legal reasoning. Document what the tool is used for, what data categories are involved, which lawful basis is being considered, why the processing is necessary and what safeguards are applied. Keep the record proportionate to the risk.

6. Consider whether a data protection impact assessment is appropriate. Article 35 describes a DPIA for processing likely to result in high risk. Large-scale sensitive data, systematic evaluation, monitoring or novel uses with significant effects deserve particular attention.

7. Train staff with a decision rule. A useful rule is: if the person is identifiable, the information is confidential, or disclosure would surprise the customer, stop and use the approved escalation route. Reinforce the rule with examples rather than relying on a long policy nobody remembers.

Industry-specific implications

Legal and professional services: Client confidentiality and professional secrecy may create concerns beyond GDPR. Drafting with fictional facts is very different from uploading case files, witness statements or privileged communications.

Accounting and financial services: Customer records can contain identity, income, tax and banking information. Summaries should be created from minimised inputs or within a specifically approved environment, with contractual and sector obligations reviewed.

Healthcare and social care: Health information is special-category data. Casual use of a general chatbot for clinical notes, patient messages or care records should be treated as high risk and escalated for specialist review.

Education: Student information may involve children and vulnerable individuals. Schools and training providers should consider age, transparency, parental expectations, access controls and the purpose of the processing before deploying AI workflows.

The decision is not simply free versus paid

A managed business plan can provide better contractual and administrative foundations than separate consumer accounts, but price alone does not decide suitability. For a 15-person firm, the commercial question should be framed at team level: compare the total monthly cost with the staff time saved, then weigh that benefit against implementation, training and privacy review costs.

Pricing and plan features can change, and OpenAI does not publish a single universal EUR price for every business arrangement. As of July 2026, confirm the available plan, billing currency, data residency options and contract terms directly with the provider before building a customer-data workflow around it.

Frequently asked questions

Related reading: our AI governance by region.

Try our free AI Privacy Risk Scorer to score your current AI tool setup against data-privacy best practice.

Is it illegal to put any customer data into ChatGPT under GDPR?

No blanket rule makes every submission illegal. The assessment depends on factors including purpose, lawful basis, necessity, transparency, the type of data, the service contract, security and international transfers. A business should avoid treating this as general permission to paste full customer records.

Does customer consent make it safe?

Consent is one possible lawful basis, but it must meet GDPR conditions and may not be the most appropriate basis for an ordinary service relationship. Consent also does not remove other issues such as data minimisation, processor terms, security, retention or overseas transfers.

Can staff use ChatGPT if they remove the customer's name?

Removing a name helps, but the remaining details may still identify the person. Staff should remove or generalise all information that is unnecessary for the task and consider whether the person could be identified from the combination of facts.

Is ChatGPT Business automatically GDPR compliant?

No service can make a customer's use automatically compliant. OpenAI offers business privacy controls and a DPA that can support an organisation's assessment, but the business still decides the purpose, data submitted, lawful basis, access, retention and internal controls.

Do we need a DPIA before using ChatGPT?

Not every use necessarily triggers a DPIA. Article 35 links a DPIA to processing likely to result in high risk, particularly when new technology, sensitive data, scale, monitoring or significant effects are involved. Check guidance from your national supervisory authority and obtain specialist advice for higher-risk deployments.

What is the safest rule for a small team?

Use approved business accounts, prohibit sensitive and highly confidential data, minimise or fictionalise customer details, and require escalation whenever a person remains identifiable. This creates a safer operational default while more complex uses are assessed properly.