Skip to content
PROOF48
All answers
ai

What your privacy policy has to say once an AI assistant reads client messages

Once an AI assistant reads client messages, your privacy notice needs a few new lines: what the chat collects, which company processes the text, how long conversations are kept, whether staff read them and who decides exceptions. GDPR Article 13 sets the list, and the chat window needs a short line saying it is an AI. A lawyer checks the final wording.

Founder, PROOF48Updated 2026-11-197 min read

01What an AI assistant adds to your privacy notice

Your privacy page probably describes a contact form: a name, an email, a message. A chat assistant changes three things. First, the input is free text, so people can type anything, including things you never asked for. Second, that text goes to a company that runs the chat tool or the AI model behind it. Article 4(9) of the GDPR defines a recipient as any body to which personal data are disclosed, whether a third party or not, and Article 13(1)(e) asks you to name the recipients or their categories. Third, conversations pile up, and Article 13(2)(a) asks for the storage period or the criteria used to set it. Our post on whether a small business website needs a privacy policy covers the whole Article 13 list. This one covers only what an assistant adds.

  • Data: what people type in the chat, plus any contact detail they leave
  • Recipients: the chat tool and the AI model provider, named or described by category (Article 13(1)(e))
  • Transfers: say so if a provider sends data outside the EU (Article 13(1)(f))
  • Retention: how long conversations are kept, or the rule that decides it (Article 13(2)(a))

02Say why the assistant reads the message, and on what basis

Article 13(1)(c) asks for the purposes of the processing and its legal basis. Write one purpose for each use, in plain words: answering a question, booking a time, passing the request to a person and, if you do it, reviewing conversations to correct the assistant. The legal basis is a choice you make with your lawyer. Article 6(1) lists consent in point (a), a contract or steps taken at the person's request before one in point (b), and legitimate interests in point (f). If you rely on legitimate interests, Article 13(1)(d) asks you to state them. Skip the catch-all line 'to improve our services'. Whether the vendor trains its models on your chats is settled first in the vendor contract, and the notice then describes whatever the contract says. If you want to see what an assistant would collect on your site, tell PROOF48 how clients contact you today and it shows what the AI handles and what it hands off.

  • One purpose per use: answering, booking, handing off to a person, quality review
  • A legal basis named for each purpose (Article 6(1) lists the options)
  • Legitimate interests spelled out if you rely on them (Article 13(1)(d))
  • No catch-all line such as 'to improve our services'

03Say who reads the chat and who decides

Two statements need to be true before you print them. The first is who reads the conversations. If someone on your team reviews them, the notice should say that staff can read them. PROOF48's AI assistant page describes a weekly review in which you see every conversation the AI handled, sorted into answered, escalated or unanswered. The second is who decides. Article 13(2)(f) asks you to disclose automated decision-making of the kind described in Article 22(1) and (4), with meaningful information about the logic involved. Article 22(1) covers a decision based solely on automated processing that produces legal effects or similarly significantly affects the person. An assistant that gives opening hours and offers a free slot is not obviously that. One that refuses a booking, declines a refund or screens out an applicant might be, and that question goes to a lawyer before launch. The cleaner design is that complaints, refunds and exceptions go to a person, and the notice says so.

  • Say staff can read conversations, if they can
  • Say a person handles complaints, refunds and exceptions, only if your setup does that
  • If the assistant could refuse, approve or rank people, ask a lawyer about Article 22 before launch
  • Never write that no automated processing happens unless that is true

04Put a line where the conversation starts

Article 13 information is due at the time the data is obtained, so a link in the footer arrives late. The AI Act adds a second duty that sits beside the privacy law, not inside it. Article 50(1) says AI systems intended to interact directly with people must be designed so that those people are informed they are dealing with an AI, unless that is obvious to a reasonably well-informed person. Article 50(5) says the information comes in a clear and distinguishable manner at the latest at the first interaction. The European Commission's timeline page says the Article 50 transparency rules apply from 2 August 2026. Article 50(1) is addressed to providers, so which duties fall on you as the business using the tool is a question for your lawyer. Two short lines above the chat box can say it is an AI and point to the full notice. The same lines belong wherever else the assistant answers, such as a DM or an email auto-reply. If you are weighing an assistant for your website, DMs or email, the AI assistant page says what it handles and what it hands off.

  • Above the chat box: it is an AI assistant, conversations are kept, staff can read them, link to the full notice
  • The AI Act line and the privacy line come from different laws and can sit together
  • Repeat the lines on every channel where the assistant answers
  • A starting line for your lawyer to edit: 'You are chatting with an AI assistant. We keep this conversation for [period], our team can read it, and [company] processes it for us. Details are in our privacy notice.'

05Where PROOF48 is the wrong choice

PROOF48 builds AI assistants. It is not a law firm and it does not write your legal text. AUTOMATE, at 2,997 EUR a month, includes an AI website chatbot that qualifies leads and books, AI responders for DMs and email, booking automation and a CRM. It does not cover WhatsApp or phone calls. The AI assistant page says the build starts with a week of mapping every message type clients send, and in our view that map is the best raw material for the notice, because it shows what people actually type. This site publishes no data processing agreement template, no sub-processor list and no security certification, and our privacy page covers proof48.com, not a client's chatbot. If you need any of those today, or the chat would give medical, legal or financial advice, PROOF48 is the wrong choice. A lawyer checks the final text either way.

  • AUTOMATE, 2,997 EUR a month: website chatbot, DM and email auto-responders, booking automation, CRM
  • WhatsApp and phone calls are not covered by AUTOMATE
  • No DPA template, sub-processor list or security certification is published on this site
  • Wrong choice if you need those today, or if the chat would give medical, legal or financial advice
From a build we shipped

Not an assistant: a hotel site that publishes the route to a person

None of our eight written-up builds is an AI assistant, so there is no assistant case to quote. The nearest is Santorini Hotel, a seafront hotel in Eforie Nord. Its site ships a booking enquiry flow and publishes the reception, restaurant and complaints phone lines. For a notice that promises a person handles complaints, that is the useful part: the human route has to exist and be easy to find. The case study records no chatbot, no privacy wording and no booking figures, so we claim none.

See the build: Santorini Hotel

Questions, answered

Does an AI chat need its own privacy policy?

It needs the Article 13 information, not necessarily a second document. A new section in the notice you already have, plus a short line at the chat window, is one way to give it. Your lawyer decides the form.

Do I have to tell people they are talking to an AI?

Article 50(1) of the AI Act says systems meant to interact directly with people must be designed so people are informed, unless it is obvious. It is addressed to providers, so ask your vendor and your lawyer what falls on you. In our view one line at the chat window is cheap and clear.

Do I have to disclose automated decision-making?

Article 13(2)(f) asks for it where the processing is the Article 22(1) kind: a decision based solely on automated processing with legal or similarly significant effects. A bot that answers questions and offers slots is not obviously that. A bot that refuses refunds or screens applicants might be, so ask a lawyer.

Can my notice say we do not store chat conversations?

Only if that is true. If anyone reviews conversations, or the chat tool keeps a log, they are stored. Article 13(2)(a) asks for the storage period or the criteria used to set it.

Where PROOF48 fits

Once an AI assistant reads client messages, add to the privacy notice what it collects and why, which company processes the text, how long conversations are kept, whether staff read them and who decides exceptions, and put two short lines at the chat window. Article 13 sets the list, Article 50(1) of the AI Act covers telling people it is an AI, and a lawyer checks the final text. PROOF48's AUTOMATE plan, 2,997 EUR a month, includes the website chatbot and the DM and email responders, not WhatsApp or phone.

· Founder, PROOF48

Cosmin has worked in online marketing and software development since 2018, and has built with AI every day since January 2023. He still writes code from scratch daily, in Python, React, Rust and C#. PROOF48 ships business websites in 48 hours and invoices afterwards, so every opinion on this blog was paid for in real builds, not in theory.

Related articles