Practical

AI and Personal Data: How to Use These Tools with More Control

By Daniel Sardá · Published on

7 min read1,504 words

In this article · 7 sections

A practical guide to deciding what to share with an AI tool, minimizing personal data, and retaining control over how it is used.

Pasting an email into an AI assistant to summarize it may seem harmless. Yet that email may contain names, addresses, account details, work information, or a conversation the other person never expected to leave its original context. The important decision happens before you press “send.”

Using AI and protecting personal data are not incompatible goals. The problem arises when information is shared out of habit, without asking whether it is necessary, who may process it, or for how long. This guide offers a simple method for making that decision with more control. It is not legal advice: specific obligations vary by country, activity, and type of information.

Key idea: The useful question is not only “can AI do this task?” but “what is the minimum information it needs to complete it?”

First, identify the kind of information in front of you

Personal data is information relating to an identified or identifiable person. It is not limited to a name or an ID number. It can also include an email address, photograph, voice recording, location, employment history, or a combination of details that reveals whom the information concerns.

That definition is consistent, for example, with the European Union’s General Data Protection Regulation. It is a useful practical reference, but it does not make the European regime a universal rule. Each jurisdiction defines its own scope and obligations.

It helps to distinguish three groups:

The fact that data is public does not mean it is free of limits. A professional address visible online still has a context and may be subject to rules about purpose, reuse, or the person’s rights. Availability is not the same as unlimited permission.

A four-question filter before you share

Before pasting text, uploading an image, or attaching a document, pause for a few seconds and apply this filter:

  1. What task am I trying to solve? State a specific purpose: summarize a text, improve its style, or compare clauses. “See what AI says” does not justify exposing additional information.
  2. Which data is essential? Separate what the tool needs from what the file happens to contain.
  3. Could someone identify the person? Consider not only names, but roles, dates, locations, unusual events, and combinations of traits.
  4. Is there a lower-exposure alternative? You may be able to use an excerpt, a summary, fictional values, or an expressly authorized environment.

This process points to two useful principles: purpose limitation, knowing why information is used, and data minimization, limiting it to what is adequate and necessary. Both appear in the European GDPR and in Ibero-American guidance, such as this guide for public and private entities from Argentina’s Agency for Access to Public Information.

Minimizing data is not just deleting the name

Suppose you want to summarize an email from a client. Removing the signature helps, but the body may mention a small company, a unique role, a date, and a highly specific incident. Taken together, those elements may identify the person.

A more cautious version would replace names with roles—“client,” “supplier”—remove contact details and account numbers, generalize dates, and retain only the excerpt needed. If the goal is to improve the tone of a reply, describing the situation without pasting the original message may be enough.

Pseudonymization reduces the direct link by separating identifiers, but information remains personal data if a key or context can reconstruct the identity. Anonymization requires much more: identifying people or extracting their data must be very unlikely using reasonable means. The European Data Protection Board cautions that this assessment depends on the case; deleting a column of names alone does not provide that guarantee.

Caution: If the context still makes the person recognizable, replacing their name with “User A” does not automatically make the data anonymous.

The same logic applies to files. Before uploading a contract, check headers, signatures, attachments, comments, visible metadata, bank accounts, and third-party data. To compare a clause, isolate that clause and replace identifying values. If identity is essential to the task, do not improvise: use an approved environment or escalate the decision to the responsible person.

Reviewing the tool is also part of the decision

Reducing the content is only one layer. The other is understanding the service’s terms. Practices can vary by provider, product, plan, and settings, so do not assume that every tool trains on conversations—or that none does.

Before entering data that is more than trivial, check the current documentation for:

Spain’s Data Protection Agency advises users to limit the personal information they expose, review privacy notices, and not rely blindly on the accuracy of responses. A clear policy can help you decide, but it does not replace permission to share other people’s information.

Privacy and security are not the same thing. Security aims to reduce unauthorized access, loss, or disclosure. Privacy also asks whether the use is lawful, proportionate, and consistent with what the person could reasonably expect. A technically secure tool can still be used for an improper purpose.

When you act for an organization, the threshold is higher

Someone experimenting with fictional information faces a different situation from an employee uploading client, patient, or worker records. In an organization, it is not enough that a tool is convenient or that the user accepted its terms.

Before using real data, the team should know:

For example, preparing a job description does not require pasting real résumés: fictional profiles are often sufficient. If AI supports a decision about employment, health, credit, or rights, a responsible person must be able to examine, challenge, and correct the result. Human review improves control, but it does not undo a disclosure that has already occurred.

A clear allocation of responsibility protects individual autonomy and limits the informational power of companies and public authorities. Voluntarily accepting a service should not become an indefinite surrender of control over one’s own data or another person’s data. Anyone who decides to share information should be able to explain why and what safeguards were used; this approach is connected to individual rights.

Key idea: A provider’s terms govern a relationship with the service; they do not automatically authorize disclosure of clients’, colleagues’, or citizens’ data.

What to do if you already shared information you should not have

Do not try to hide the problem. Act quickly and retain the information needed to describe what happened without multiplying its exposure.

  1. Stop further uploads and note what information was shared, when, through which account, and in which tool.
  2. Review the available controls for deleting the conversation or file, without assuming that this removes every copy or record.
  3. Change credentials immediately if the content included passwords, keys, or tokens.
  4. Inform the internal privacy, security, or compliance lead if you were acting for an organization.
  5. Check the obligations that apply in your jurisdiction. Notification deadlines and procedures are not the same in every country.

The appropriate response depends on the sensitivity of the information, the number of people affected, and the possibility of misuse. An isolated name does not pose the same risk as a medical record or a database of credentials.

Checklist before you press “send”

Make one final review:

The goal is not to strip AI of its usefulness or avoid it out of fear. It is to preserve the capacity to choose. Sharing less, checking more, and retaining identifiable human responsibility makes it possible to use these tools without treating personal information as a resource available by default.

Do You Own Your Personal Data? Rights, Control, and OwnershipPersonal data is about us, but that does not necessarily make it property we can sell or reclaim like an object. The distinction matters for understanding our rights and the limits on companies and public authorities.AI and privacy: what data not to share and how to reduce riskA practical guide to classifying information, removing unnecessary data, and choosing the right channel before using an artificial intelligence tool.