PrivaSearch

Rules

What is personal data? Examples, and what you may do with it in an AI chatbot

What personal data is, with examples from names to IP addresses, what falls outside the definition and what it means for a client file in an AI chatbot.

In brief

  • Personal data is any information about an identified or identifiable natural person: a living human, not a company.
  • It is not limited to names and addresses: a client number, number plate, voice recording or IP address can be personal data too.
  • Whether something is personal data depends on whether someone can reasonably be identified, including by combining data.
  • Pseudonymised data remains personal data for the organisation able to reveal the pseudonyms; only truly anonymous data falls outside the GDPR.
  • Putting a document into an AI chatbot processes all personal data it contains, including in subordinate clauses and attachments.

What is personal data?

Personal data is any information about a living human or traceable to that person. Names and addresses qualify, but so do a client number, photograph, IP address or note about someone's behaviour. It does not matter whether the information is public or work-related. What determines it is whether you can establish whom it concerns.

The definition is in Article 4(1) GDPR: “any information relating to an identified or identifiable natural person”. The legislation calls that person the data subject. On its page what is personal data, the Autoriteit Persoonsgegevens (AP), the Dutch data protection authority, summarises it this way: information either concerns someone directly or can be traced to them.

This article explains the concept. It is not legal advice. For a borderline case, you need a lawyer's or data protection officer's (DPO) judgement.

How do you read the definition of personal data?

In their 2007 opinion on the concept of personal data, European supervisory authorities divided the definition into four elements. The opinion concerns the directive preceding the GDPR, but the AP still references it.

Any information

The concept is deliberately broad. It includes facts, such as a date of birth, and opinions, such as an assessment in a performance review. The format does not matter: text, images, sound or a number in a spreadsheet.

Relating to a person

Information relates to someone if it says something about them, is used to assess or treat them, or may have consequences for them. The opinion gives the example of a garage's maintenance register. Its data concerns a car. Once the garage links it to the owner to send an invoice, it concerns the owner too.

Identified or identifiable

Someone is identified if you know who they are. Someone is identifiable if you can work it out. The legislation lists a name, identification number, location data, online identifier and factors specific to physical, physiological, genetic, mental, economic, cultural or social identity as means of identification.

Natural person

The GDPR protects people. A limited company, foundation or municipality is not a natural person. The legislation does not apply to deceased people either. Both are discussed below, because exceptions exist.

What is the difference between directly and indirectly identifying data?

The AP distinguishes direct and indirect personal data. Direct data concerns someone directly: a name or passport photograph. Indirect data does not directly concern someone, but combined with other data still says something about a person.

When is someone identifiable?

Recital 26 GDPR gives the test. Consider all means reasonably likely to be used by the organisation itself or anyone else to identify the person. Objective factors count: cost, time required, available technology and its development.

This means two things. A theoretical possibility nobody would reasonably use does not itself make information personal data. And what cannot be traced today may be traceable in a few years.

Combinations do the work

One item sometimes says little. Two or three together identify one person. The AP gives an example on its pseudonymisation page: a date of birth with the four digits of a Dutch postcode is often unique enough to find someone.

Client files work the same way. A fictional example: a report from bookkeeping firm De Meerkoet says “the owner of the bakery on the square” has payment arrears. Anyone knowing the village needs no name.

Which personal data examples occur in practice?

The AP gives obvious examples such as name, address, telephone number and passport photograph. It also mentions what someone does or buys online, whether they have allergies, client and personnel numbers, and camera footage or sound recordings in which someone can be recognised.

The table covers data organisations most often question. “Depends” means: consider who holds the information and what they can do with it.

Online data: IP addresses and cookies

Recital 30 GDPR mentions IP addresses, cookie identifiers and similar device identifiers. They leave traces that can be used, particularly with other information, to create profiles and identify people.

Images and sound

A photograph, meeting recording or security camera footage is personal data if someone is recognisable. A conversation transcript is too: it contains names, and what someone says concerns them. For what this means for automated reports, see AI meeting minutes and the GDPR.

What is not personal data?

The AP names two types not considered personal data under the GDPR: information about organisations and deceased people. Anonymous data also falls outside it. All three have a qualification.

Recital 14 says the GDPR does not concern legal persons' data, such as an undertaking's name, legal form and contact details. According to the AP, this changes if information also says something about a human. It gives three cases:

In those cases, it is personal data.

Information about deceased people

Recital 27 says the GDPR does not apply to deceased people's personal data and Member States may set their own rules. The AP points to the other side: information about a deceased person may say something about someone living, for example a hereditary condition. It then constitutes that living person's personal data.

Anonymous data

Data is anonymous if it does not relate to an identifiable person, or has been processed so nobody is identifiable any longer. Recital 26 says the GDPR does not apply to it. The threshold is high. Removing a name is insufficient if the rest of the document still points to someone.

Which personal data types does the law distinguish?

Not all personal data receives the same treatment. The more sensitive it is, the stricter the rules.

Sensitive is not a statutory category

The AP also uses sensitive personal data for data not given special status in the legislation but having greater impact: electronic communications data, location data, financial data and the BSN. No exhaustive list exists.

The eight special categories, exceptions and common mistakes are explained in special category data.

Is pseudonymised data still personal data?

Yes, for the organisation holding the key. Article 4(5) defines pseudonymisation as processing personal data so it cannot be attributed to a person without separately held additional information. Recital 26 adds that pseudonymised data attributable to someone using additional information counts as information about an identifiable person.

The AP calls pseudonymisation a security measure and explicitly says it is not anonymisation. It also warns about doing half the job: removing only directly identifying information such as names is not pseudonymisation if the remainder contains enough information to identify people.

Whether the same data is personal data for a recipient without the key is a question the Court of Justice addressed in 2025. What that judgement does and does not mean is explained in pseudonymising or anonymising. There is also a European proposal to amend the personal data definition. Until it becomes legislation, Article 4 applies as set out above. Check its status before relying on it.

What may you do with personal data in an AI chatbot?

A chatbot is no special case under the GDPR. The question is what you do with data, which is called processing.

Entering is processing

Under Article 4(2), processing is almost anything you can do with personal data, from collection and storage to consultation, disclosure and erasure. Pasting text into a chat window or uploading a file qualifies: you disclose data to the chatbot provider. In August 2024, the AP warned it had received several breach notifications because employees shared patients' or clients' personal data with an AI chatbot.

A document contains more personal data than you think

The name on the cover is not all that counts. The signature, client file number, Dutch number plate on a claim form, caseworker's name in a footer and comment about “the neighbour at number 12” are personal data too. Removing just the client's name therefore still means entering personal data.

What to arrange before personal data goes into a chatbot

  1. Establish the contents. Which people and data types, including subordinate clauses and attachments.
  2. Determine purpose and lawful basis. Does chatbot use fit why you received the data?
  3. Limit input. Article 5 requires no more data than necessary for the purpose. If the question can be asked without names, ask it without names.
  4. Arrange the provider. A business service with a data processing agreement and a transfer basis if data leaves the EU. See what is a data processing agreement.
  5. Assess risk. Sensitive data or large numbers may require a DPIA. See DPIA: meaning and implementation.
  6. Document it and tell employees. In your record of processing activities and AI arrangements.

What to arrange for each data type, from names to BSNs, is in client data in an AI chatbot: rules by data type. The main overview is on ChatGPT, Claude, Mistral Vibe and the GDPR.

What PrivaSearch does here

PrivaSearch sits between your documents and the AI model. Personal data such as names, addresses, BSNs, IBANs, Dutch number plates and dates of birth is detected inside PrivaSearch and replaced with pseudonyms. The steps are on how PrivaSearch works.

Automatic detection is never complete. A rare job title in a small village is not a name but still points to someone. A colleague therefore sees the original alongside the pseudonymised text, adds what the model missed and approves it. Only then is a document searchable. This review is mandatory.

A connected AI assistant receives only released, pseudonymised passages with source references, without original files, filenames or links. You can connect Claude, ChatGPT Business and Mistral Vibe (formerly Le Chat). What you type into the assistant goes directly to its provider: do not type names there.

For your organisation, pseudonymised passages remain personal data because you can reveal the pseudonyms. The lawful basis, record of processing activities and DPIA where needed therefore remain your responsibility. Your environment creates starter documents, not legal advice. More is on The GDPR and the EU AI Act. To see it with your type of client file, book a demo.

Frequently asked questions

What is personal data under the GDPR?

Personal data is any information relating to an identified or identifiable natural person. That is Article 4 GDPR's definition. It includes information directly concerning someone, such as a name, and information enabling identification, such as a client number or combined characteristics. Information about companies and truly anonymous data falls outside it.

Is an IP address personal data?

Often. The GDPR includes online identifiers in its definition, and its explanatory text names IP addresses as identifiers allowing people to be recognised, especially with other information. The Autoriteit Persoonsgegevens writes that an IP address can be personal data. Whether it is depends on who holds it and what they can do with it.

Is a business email address personal data?

An address in an employee's name is personal data even when used for work. A general organisational address concerns the organisation and in principle is not. According to the Autoriteit Persoonsgegevens, it becomes personal data if only one particular employee uses it. This often applies at sole traders or small firms.

Is information about a deceased person personal data?

No, the GDPR does not apply to information about deceased people. There is an important exception: if it also says something about someone living, it is that living person's personal data. The supervisory authority gives a hereditary condition as an example. Rules other than the GDPR may still apply; ask a lawyer if in doubt.

Is pseudonymised data personal data?

Yes, at least for the organisation able to reveal the pseudonyms. Pseudonymisation keeps the link to the person separately, but it still exists. The GDPR says data attributable to someone using additional information counts as information about an identifiable person. Only data processed so nobody remains identifiable is anonymous.

May you enter personal data in ChatGPT or another AI chatbot?

Only if arrangements are in place. Entering is processing: you disclose data to the provider. You need a lawful basis, a purpose compatible with why you received the data, a business service with a data processing agreement and a basis for transfers outside the EU. Stricter rules apply to special category data. A free or personal account does not provide those arrangements.

Read more

  1. The AI Act in autumn 2026: what applies now and what has been postponed?

    The AI Act timeline after the digital omnibus: what applies already, what has moved to 2027 and 2028, and what AI chatbot users need to arrange now.

  2. Deepfakes and the law: what the AI Act and GDPR do about them

    What a deepfake is, which rules apply now, which prohibition is coming and how organisations tackle fraud using a fabricated voice or video.

  3. The Algorithm Register and IAMA: what public bodies must record about their AI

    What the Dutch Algorithm Register and IAMA are, what is mandatory and what is not, and what a municipality arranges when staff use AI chatbots.

See in twenty minutes what happens to your documents.

We show it with fictitious client files from your sector: upload, detection, review, search and the connection with your AI assistant. Afterwards you'll know exactly what your team can use safely.

Book a demo →