Live chat is GDPR-compliant when you have a lawful basis for the chat data, tell visitors plainly what happens to it (including that they're talking to an AI, if they are), only store things in their browser without consent when it's strictly necessary, keep transcripts for a set period, honour access and deletion requests, and sign a data processing agreement with your chat vendor.
That's the short answer. The rest of this guide explains each part in plain English, then gives you two checklists: one for choosing a chat vendor, one for your own website.
It applies whether a person on your team or an AI chat widget types the reply. GDPR cares about the personal data in the conversation, not who answers.
If you want to know how Aiva Chat itself handles GDPR, our GDPR page covers it article by article. This guide is about what you, the website owner, need to get right.
This is general information, not legal advice. Rules vary between EU countries and the UK, and your situation may differ. For anything high-stakes, such as health data or large-scale processing, speak to a privacy lawyer or your data protection officer.
What personal data does a live chat collect?
More than most owners expect. A typical website chat handles:
- What visitors type, which can include names, phone numbers, email addresses, order details and, sometimes, things they shouldn't share at all.
- Pre-chat form answers, if you ask for details before the chat starts.
- Technical data, such as IP address, browser and the page the chat started on.
- Browser storage, such as a visitor ID that lets a conversation carry on across pages.
- Transcripts and summaries saved in your dashboard after the chat ends.
All of it is personal data if it can identify someone. So the full GDPR applies, from the principles in Article 5 onward.
What lawful basis do you need for live chat data?
Every use of personal data needs one of the six lawful bases in Article 6 GDPR. For live chat, three come up.
| Lawful basis | When it fits a chat | Watch out for |
|---|---|---|
| Contract, Art. 6(1)(b) | A visitor asks to book, order or get a quote, or an existing customer asks for help with their account | Only covers what's needed for that request |
| Legitimate interests, Art. 6(1)(f) | Answering general questions, keeping transcripts to follow up and improve answers | You must weigh your interest against the visitor's, and record that you did |
| Consent, Art. 6(1)(a) | Adding the visitor to a marketing list, or non-essential tracking | Must be freely given, specific and as easy to withdraw as to give |
Most businesses run chat on contract and legitimate interests. Consent is the basis people reach for by reflex, but it's often the wrong one for the chat itself.
Where you rely on legitimate interests, do a short written balancing test. The EDPB's guidelines on legitimate interest explain the three steps: a real interest, processing that's necessary for it, and a check that visitors' rights don't override it.
How do you tell visitors what happens to their chat data?
Article 13 says you must give people certain information at the time you collect their data. For chat, that means:
- Your privacy notice covers chat: what you collect, why, the lawful basis, how long you keep it, who processes it and whether it leaves the EU.
- A short line and link in or near the chat window, such as "We keep chats to answer you and improve our service. See our privacy notice."
- If you have a pre-chat form, the purpose of each field is clear.
Telling visitors they're talking to an AI
If an AI answers your chat, say so. Since 2 August 2026, Article 50 of the EU AI Act requires AI systems that interact directly with people to be designed so people are informed they're interacting with an AI, unless that's obvious from the context. It must be clear no later than the first interaction.
Strictly, that duty falls on the provider of the AI system. But GDPR's own fairness and transparency principles push you the same way. The simplest fix is a welcome message like "Hi, I'm the AI assistant for Studio Maya. A person can take over if you'd like." Don't give the AI a human name and photo and leave visitors guessing.
If someone opens a chat and asks a question, answering it doesn't need a tick box. Storing things in their browser might.
Does a chat widget need cookie consent?
This is where the ePrivacy rules come in, not GDPR. Article 5(3) of the ePrivacy Directive requires consent before you store information on, or read it from, a visitor's device. The only exceptions are storage that's purely for sending a communication, or storage that's strictly necessary for a service the visitor explicitly asked for.
Two points matter for chat:
- It isn't only cookies. The EDPB's guidelines on the technical scope of Article 5(3) confirm the rule covers similar technologies too. Local storage in the browser counts as storing information on a device.
- "Strictly necessary" is narrow. EU regulators' opinion on cookie consent exemptions treats "user-input" storage, such as keeping track of what someone entered during a session, as likely exempt. It also says that kind of storage should usually be short-lived, and that persistent storage is less likely to qualify. Where you're in doubt, it advises asking for consent.
So be balanced about it. Storage that keeps a chat running while a visitor moves between pages is a strong candidate for the exemption, because the visitor asked for the chat. Storage that follows a visitor for months, feeds analytics or advertising, or loads before they've touched the chat is a weaker one. Ask your vendor for the exact list of what its widget stores, why and for how long, then make the call with your adviser and list it in your cookie policy either way.
What should a pre-chat form ask for?
As little as possible. Data minimisation means collecting only what's adequate, relevant and limited to what you need.
- If the chat is for quick questions, you may not need a form at all.
- If you need to call or email someone back, ask for one contact method, not three.
- Make optional fields clearly optional.
- Keep any marketing consent as a separate, unticked box. Never bundle it with starting the chat.
How long can you keep chat transcripts?
GDPR doesn't give a number. The storage limitation principle says you keep personal data no longer than you need it for the purpose.
A sensible approach:
- Decide why you keep transcripts: follow-ups, complaints, training your team, improving answers.
- Pick a period that covers those reasons. For many small businesses that's months, not years.
- Write it in your privacy notice.
- Use a chat tool that deletes old transcripts automatically. A rule nobody enforces isn't a retention policy.
How do you handle access and deletion requests for chats?
Visitors have rights under Articles 15 to 22, including access to their data and erasure. Under Article 12, you have one month to respond, extendable by two further months for complex requests.
In practice you need to be able to:
- Find every chat linked to a person, usually by email or phone number.
- Export it in a readable format.
- Delete it, unless you have a legal reason to keep it.
If your chat vendor can't help you do this, you'll struggle to meet the deadline. If the AI makes decisions with legal or similarly significant effects on people, such as refusing a service, Article 22 adds further rights. Most booking and FAQ chats don't go that far, but keep a person in the loop for anything that does.
Controller, processor and the DPA
When you run chat on your website, you're normally the controller: you decide why the data is collected. Your chat vendor is the processor: it handles data on your behalf.
Article 28 requires a written contract between you, usually called a data processing agreement or DPA. It must cover, among other things, that the vendor only acts on your instructions, keeps data confidential, applies appropriate security, uses sub-processors only with your authorisation, helps you with data subject requests and breaches, and deletes or returns data at the end.
No DPA, no lawful processor relationship. If a chat vendor can't show you one, move on.
Sub-processors and international transfers
Most chat tools rely on other companies for hosting, email or AI. Those are sub-processors, and the vendor should publish a list and tell you before it changes.
If any of them stores or accesses data outside the EU or UK, that's an international transfer under Chapter V GDPR. The common routes are:
- The EU-US Data Privacy Framework, for US companies certified under it. The European Commission adopted it in July 2023. A legal challenge is currently on appeal at the EU Court of Justice, so it's worth knowing which route your vendor uses.
- Standard Contractual Clauses (SCCs), the Commission's model clauses, usually built into the DPA.
Remember remote access counts too. A support team in another country viewing your transcripts is a transfer.
Security for chat data
Article 32 asks for security appropriate to the risk. For a chat tool, the basics are encryption in transit, encrypted storage of credentials, separate sign-ins for each team member, a record of who did what, and limits on where the widget can load. If a breach does happen, you as controller generally have 72 hours to notify your supervisory authority, so your vendor must tell you fast.
Health and other sensitive data: a warning for clinics
Clinics, therapists, pharmacies and similar businesses need extra care. Health data is a special category under Article 9, and processing it needs both an Article 6 basis and one of the Article 9 conditions, such as explicit consent.
Visitors will type symptoms into a chat box whether you ask or not. So:
- Tell visitors not to share medical details in chat, and route clinical questions to a proper channel.
- Keep the chat's job to bookings and general questions.
- Shorten retention.
- Consider a data protection impact assessment before launch if health data is likely at any scale.
AI-specific questions: training and transcripts
AI chat adds two questions a human chat tool doesn't raise.
Is my visitors' data used to train AI models? Ask the vendor directly, and ask about the AI provider behind it. Training on chat data is a separate purpose that needs its own justification, and visitors won't expect it.
What does the AI keep? Transcripts and summaries are stored data like any other. Look for automatic removal of sensitive details such as card numbers, and check that summaries follow the same retention rules as transcripts.
Our article on data privacy and AI phone agents covers the same questions for calls.
Live chat vendor GDPR checklist
| Question to ask | What a good answer looks like |
|---|---|
| Do you offer a DPA under Article 28? | Yes, published, and it applies to every customer |
| Who are your sub-processors? | A public list with purpose and location, plus advance notice of changes |
| Where is chat data stored? | A named region, stated plainly |
| How are international transfers covered? | SCCs or the Data Privacy Framework, named in the DPA |
| Exactly what does the widget store in the browser? | A full list with purpose and lifetime |
| Can I set retention and have old chats deleted automatically? | Yes, in settings |
| Can I export and delete one person's chats? | Yes, or the vendor does it on request within a set time |
| Do you train AI models on our chats? | A clear no, in writing |
| How fast will you tell us about a breach? | A fixed time, such as within 72 hours |
| Which security claims are independently audited? | An honest answer, including "none yet" |
Website owner checklist
- Lawful basis recorded for chat (and a balancing test if you use legitimate interests)
- Privacy notice updated to cover chat, the vendor and any transfers
- Notice link visible in or near the chat window
- Welcome message says the visitor is talking to an AI, if they are
- Widget storage reviewed and listed in your cookie policy
- Pre-chat form trimmed to what you need; marketing consent separate and unticked
- Retention period set and automatic deletion switched on
- Process for access and deletion requests within one month
- DPA in place with your chat vendor
- Team members on their own sign-ins
- Clinics: visitors told not to share medical details; DPIA considered
Where Aiva Chat stands
Here's what our own published documents say, so you can check them against the lists above.
- DPA: Our Data Processing Addendum applies automatically to every customer, with a countersigned copy on request. You're the controller and Aiva Chat is the processor.
- Sub-processors: Aiva Chat itself uses three: Amazon Web Services, Stripe (for our own subscription billing) and Firecrawl (for public web pages you import into knowledge). The sub-processor list also names the services that run on your own accounts, and we give 30 days' notice of changes.
- Where data lives: On AWS in the United States (US East, N. Virginia). We don't offer an EU region today. Our team in Rajkot, India may access data to provide support. Transfers of EU personal data rely on the Standard Contractual Clauses in our DPA, and UK customers can request the UK International Data Transfer Addendum.
- AI: We don't train AI models on your customer conversations. AI processing runs on your own OpenAI account, under its data terms.
- Retention and scrubbing: You set how long recordings and transcripts are kept, and a daily job deletes older ones. Card numbers, SSNs, dates of birth and terms you add are removed from transcripts and summaries.
- Rights and exit: Privacy requests are handled within 30 days. Call logs, appointments, contacts and form submissions can be downloaded as CSV, and data is deleted within 90 days of termination.
- Breaches: We notify you within 72 hours of discovery.
- The widget: It sets no cookies. It uses local storage for a visitor ID, a phone number a visitor enters (expires after 30 days) and whether they closed the greeting bubble. Our cookie policy lists each one; it's yours to declare in your own cookie policy.
- Security: Team roles with their own sign-ins, an activity log, encrypted stored keys and an allowed-websites list for the widget. See our security page.
What we don't have yet: an EU hosting region and any third-party certification such as SOC 2 or ISO 27001. If either is a hard requirement for you, we'd rather you knew now. Our privacy policy and GDPR page have the full detail, and we provide a DPIA support package on request.
The website chat itself lets you add a pre-chat form with your own consent text, set the welcome message, and have your team take over any chat.
Ready to set up a chat you can stand behind?
Start with the owner checklist above, then put the vendor questions to every tool on your shortlist, including ours.
Start free and try the website chat on your own site, book a demo and we'll walk through the privacy settings with you, or see pricing.
Written by
Jagdish Parmar
AI Security Team Lead
Jagdish leads AI security at Woyce AI, the team that builds Aiva Chat. Jagdish secures and runs AI systems in production: agent permissions, prompt-injection defences, AWS infrastructure, deployment pipelines and monitoring.
More posts

