Skip to content
AI-SNS.IO — CHAT SUPPORTAI-SNS.IO — CHAT SUPPORT
Security & Compliance

GDPR and AI Chat: What to Know Before You Deploy

AI SNS CHAT EditorialPublished 7 min read

A practical guide to GDPR for AI chat on your website, WhatsApp and Messenger: lawful basis, data minimisation, retention, vendors and human oversight.

If you run an AI chat agent on your website, WhatsApp or Messenger, you are processing personal data from the first "hello". Names, email addresses, order numbers, phone numbers and free-text messages all count. This article walks through the GDPR questions that matter most for AI chat, what to decide before launch, and how to set up your tooling so the answers are easy to defend. It is general guidance for planning, not legal advice.

Why AI chat raises GDPR questions

A chat conversation looks informal, but under the GDPR it is a record of personal data. Customers also tend to over-share: they paste addresses, health details or account numbers into a chat box without being asked.

AI adds three things on top of a normal live-chat tool:

  • Free text goes to a model provider. The message is processed by a language model, often run by a third party.
  • Answers are generated, not looked up. You need to be able to explain what the bot is allowed to say and where its knowledge comes from.
  • Scale. An agent can talk to thousands of people, so a small configuration mistake is repeated thousands of times.

The GDPR applies if you offer goods or services to people in the EU or UK, or monitor their behaviour, wherever your company is based. That is why businesses in the US and Asia-Pacific often need to care about it too.

Know your roles: controller, processor and sub-processors

Most of the legal weight depends on who plays which role.

RoleWho it usually isWhat it means
ControllerYour businessDecides why and how chat data is used
ProcessorYour AI chat vendorProcesses data on your instructions
Sub-processorModel providers, hosting, messaging platformsEngaged by the processor to do part of the work

In practice, this means you should:

  1. Sign a data processing agreement (DPA) with your chat vendor.
  2. Ask for the list of sub-processors, including which AI model providers may see your customers' messages.
  3. Check where data is stored and what transfer mechanism applies if it leaves the EU or UK.
  4. Remember that WhatsApp and Messenger are run by Meta, which has its own terms and its own role for data on those platforms.

Choose a lawful basis and be transparent

Every use of personal data needs a lawful basis. For support chat, the common ones are:

  • Contract: answering an order question from a customer is part of delivering the service.
  • Legitimate interests: offering fast support on your website, balanced against the person's rights.
  • Consent: needed for things like marketing follow-ups, or where local rules on cookies and tracking require it.

Whatever you choose, tell people what is happening. A good chat disclosure is short and visible:

  • Say that they are talking to an AI agent, not a person.
  • Link to your privacy notice.
  • Explain that they can ask for a human.
  • Warn them not to share sensitive details they do not need to share.

Data minimisation: collect less, keep less

The GDPR expects you to collect only what you need and keep it no longer than necessary. For chat, that translates into specific decisions.

What the agent asks for

Only ask for fields that serve a stated purpose. If the bot can answer a shipping question from an order number, do not also ask for a date of birth. If you capture leads, each field should have a reason.

What the agent knows

Your knowledge base can leak personal data. Internal documents, old tickets and exported spreadsheets may contain customer details that a stranger could coax out with the right question. Review sources before you connect them, and prefer curated answers over raw dumps.

How long you keep it

Decide a retention period for conversations and stick to it. "Forever" is rarely defensible for support logs. Pick a window tied to a real need, such as handling complaints or warranty claims, and delete after that.

  1. 01Customer sends message
  2. 02Agent replies from approved sources
  3. 03Human reviews if flagged
  4. 04Conversation deleted after retention window
Example data lifecycle for a support chatIllustrative example

Rights requests, automated decisions and human oversight

Chat logs are personal data, so people can ask for access, correction or deletion. Plan ahead so that you can find a person's conversations by their contact details and delete or export them within the one-month response window the GDPR sets.

Article 22 limits decisions that are made solely by automated means and have legal or similarly significant effects. A support bot that explains opening hours is nowhere near that line. A bot that approves or denies refunds, credit or contracts with no human involvement may be. Keep high-impact decisions with people.

Good practice for oversight:

  • Offer an obvious route to a human at any point.
  • Review a sample of conversations regularly, using ratings and flags.
  • Correct wrong answers at the source so they do not repeat.
  • Keep a record of who changed the agent's configuration and when.

Security, transfers and outbound messaging

Security measures

Article 32 asks for "appropriate" technical and organisational measures. For an AI chat deployment, ask about encryption of stored secrets, two-factor sign-in for staff, lockout after failed logins, role-based access, protection against prompt injection and abuse, and log redaction. Ask the vendor to be specific, and be wary of vague badges. Certifications you do not see claimed are worth asking about directly.

International transfers

If your chat vendor or its model providers process data outside the EU or UK, you need a valid transfer mechanism, such as an adequacy decision or standard contractual clauses. Ask which providers handle which traffic, and whether failover can move a conversation to a different vendor.

Proactive messaging

Chat is not an excuse to skip consent rules. Unsolicited marketing over WhatsApp, SMS or email is regulated by the GDPR together with ePrivacy rules. Safer patterns include messaging only people who contacted you first, including an opt-out in every message, and respecting platform windows such as WhatsApp's 24-hour customer service window.

A pre-launch GDPR checklist for AI chat

Risky setup
Defensible setup
No notice that it is an AI
AI disclosure in the welcome message
Raw internal exports as knowledge
Curated, reviewed sources
Conversations kept forever
Defined retention window
Bot handles refunds alone
Human approves significant decisions
Anyone on the team can edit
Roles and an activity log
Risky setup versus defensible setupIllustrative example

Before you go live, confirm:

  1. You have a DPA and a sub-processor list from your vendor.
  2. Your privacy notice mentions AI chat and names the purposes.
  3. The welcome message discloses the AI and links to the notice.
  4. Knowledge sources are free of data you would not show a stranger.
  5. A retention period is set, and someone owns deletion requests.
  6. A human handoff exists and works outside office hours.
  7. Staff accounts use two-factor authentication and sensible roles.
  8. You have run a data protection impact assessment if your use is large-scale or involves sensitive data.

Doing this in AI SNS CHAT

AI SNS CHAT is built so that several of the checklist items are settings rather than projects. Its practices are designed with reference to GDPR and APPI. The site states clearly that it is not SOC 2, ISO 27001 or HIPAA certified, and it holds PrivacyMark No. 20002676. Details are on the security page.

Disclosure and handoff. Write your AI notice into the welcome message, which is auto-translated into every language the widget speaks. Set shifts under Helpdesk → Scheduling (Standard plan or higher) so handoffs go to whoever is on duty. Handoff alerts can reach Slack, Teams, LINE WORKS, WhatsApp, email or a webhook. The customer message about what happens next is built from the rota and never invents a wait time. In Activity → Chat logs, or the email Inbox, staff can take over a chat and the bot pauses.

Minimising what the agent knows. Under Data sources, you control the files, snippets, websites and Q&A the agent learns from. Passages shown to strangers are scrubbed of personal data, and secrets are redacted from logs. For lead capture, the "Collect leads" action and Custom attributes let you describe each field so the agent asks only when appropriate.

Retention and rights. A conversation retention window is available as an opt-in limit; the default is to keep conversations, so set it deliberately. In Chat logs you can filter, export, and bulk delete conversations, which helps with access and erasure requests. Contacts also support CSV export.

Access control and audit. Roles include Owner, Co-owner, Member, Support and Viewer, with per-member overrides. Workspace settings → Activity records every create, edit, delete and upload with name, email, feature and time. Users can enable TOTP two-factor authentication, accounts lock for 30 minutes after 10 failed logins by default, and each agent has prompt-injection and abuse filters.

Model choice. Text replies can use Anthropic, OpenAI, Google or OpenRouter, with failover in that order after the agent's own provider. If you want every reply to stay with one vendor, that is an admin setting, so raise it with your platform admin. Review which providers you list as sub-processors accordingly.

Outbound messaging. Outbound campaigns (Standard or higher) reach only people who messaged first, one to one, with a STOP opt-out in every message. There are no SMS or email marketing campaigns by design.

Not found in the product: SSO/SAML. Plan your staff access around email, Google or phone SMS sign-in with 2FA. Compare plans on the pricing page, or review features before you sign up.

Frequently asked questions

Do I need consent to run an AI chat on my website?

Not always. Answering support questions can often rest on contract or legitimate interests. Consent becomes important for marketing follow-ups and for any cookies or tracking that local rules cover. Decide your basis per purpose, and record it.

Must I tell visitors they are talking to an AI?

Transparency is a core GDPR principle, and other rules, such as the EU AI Act, also point toward disclosure. The simplest approach is a clear line in the welcome message with a link to your privacy notice and an easy way to reach a person.

Is a vendor's GDPR claim enough?

No. Ask for a DPA, a sub-processor list, details on transfers, and a description of security measures. Be careful with wording: practices "designed with reference to" a law are not the same as a certification, and a good vendor will say which is which.

How long should I keep chat logs?

Only as long as you have a real need, such as handling complaints or warranty claims. Pick a specific period, document the reason, and delete after it. Keeping everything indefinitely is hard to justify.

Can the bot handle deletion requests?

The bot can point people to your process, but a person should verify identity and carry out the request. Make sure your chat tool lets you find, export and delete a person's conversations.

#GDPR#AI chat#privacy#compliance#data protection

Put this into practice with AI SNS CHAT

Build an AI agent trained on your own content and connect it to your website, LINE, WhatsApp and more. Free to start, no credit card required.