Privacy and Consent in Website Chat: Cookies, Data and What to Tell Visitors
What a website chat stores, what notice to show, how it fits your cookie banner and what consent to get before moving the chat to text, email or WhatsApp.
In this article
Most teams add a chat widget in ten minutes and never open the privacy policy again. Then a visitor from Manchester asks "what do you do with what I type here?", the cookie-consent vendor flags an unclassified script, or a customer asks for their chat history to be deleted and nobody knows where it lives. Privacy and consent in website chat are not hard, but they are specific: a chat stores things a static page does not, and it is often the first step toward contacting someone by text, email or WhatsApp, which is where the real consent questions start.
This walkthrough covers what a website chat stores, where each item belongs in your policy, how the widget fits a cookie banner, what to say about AI and translation, and what permission you need before switching channels. It is general guidance for US and UK sites; laws differ by state and country, so confirm with your own counsel.
What a website chat stores (and why)
You cannot describe what you have not inventoried. The table is a starting point for the "data → purpose → basis" section of a policy; the basis column is illustrative, not advice.
| Data | Where it comes from | Purpose | Typical basis (confirm with counsel) |
|---|---|---|---|
| Visitor identifier in the browser | Set by the widget so a returning visitor resumes the chat | Continuity | Strictly necessary / functional |
| Page URL, referrer, UTM | Captured when the conversation starts | Context; attribution | Legitimate interest (UK) or disclosed use (US) |
| Messages and attachments | Typed or uploaded by the visitor | Answering; record of service | Contract / legitimate interest |
| Pre-chat fields (name, email, phone) | Entered by the visitor | Identification; follow-up | Consent or contract, depending on use |
| Logged-in user ID (via identify) | Your backend, signed with a secret | Linking chat to an account | Contract |
| Widget events (loaded, opened, started, sent) | Generated by the widget | Funnel; optionally GA4 | Analytics consent where required |
| Email transcript | Requested by the visitor at close | Visitor's own copy | Consent (they asked) |
State plainly that chat history is kept in your customer-service platform alongside other channels and may be read by the AI agent to answer. Both are normal; both should be disclosed.
Where the widget fits your cookie banner
The identifier a chat sets in the browser keeps the conversation alive between page loads. Whether that is "strictly necessary" or needs consent depends on your regulator's reading and on what else the widget does. Two workable patterns:
- Classify the widget as functional. The chat shows immediately; analytics events to GA4 fire only if analytics consent exists (Google's consent mode handles that side). Common on US sites and on UK sites that treat chat as a service the visitor deliberately uses.
- Load the widget after consent. Stricter: the loader runs only when the visitor accepts functional cookies. In Google Tag Manager that is a tag with a consent-based trigger.
The trade-off of the second pattern: visitors who decline never see the chat, including the ones with a billing question. Some sites add a static "Chat with us (enable functional cookies)" link as a fallback.
The Omnifox loader is a single line, so either pattern is straightforward:
<script src="https://omnifox.io/widget/loader.js" data-widget-key="YOUR_KEY" async></script>
If you install it through GTM, the Google Tag Manager guide covers consent-gated triggers without touching the theme.
Tell visitors about AI and translation
AI agent. If an AI answers first, say so in the welcome message or the widget name ("Ask Nova — AI assistant, a human can take over"). It sets expectations and, in some jurisdictions, is moving from courtesy to requirement.
Machine translation. When the agent writes in English and the visitor reads in Spanish, the text passed through a translation model. A short "translated automatically" note is enough; the website chat widget in Omnifox can show that notice when translation is on.
A short in-chat notice that covers both, suitable for the welcome message:
Hi! You're chatting with our AI assistant; a team member can step in anytime. Messages are stored so we can help you and may be translated automatically. Please don't share card numbers or health details here. Privacy policy: link.
Keep it to three lines; long legal text in a bubble is skipped.
Consent to continue by SMS, email or WhatsApp
The chat ends; the visitor left a phone number. Can you text them tomorrow? The rules are channel-specific and the widget cannot answer for you. In general terms:
- Email — in the US, a follow-up about the inquiry is usually fine under CAN-SPAM if it is honest and has an unsubscribe route; a newsletter is a separate decision. In the UK, service emails differ from marketing emails, which generally need opt-in.
- SMS — the US (TCPA and carrier rules) treats texting as sensitive: get clear, recorded consent before automated or marketing texts. The UK is similar under PECR.
- WhatsApp — the business can reply freely within 24 hours of the visitor's last message; after that, only approved templates, and the visitor should have agreed to be contacted there.
The practical fix is a consent question where the phone is collected, in the pre-chat form or in the agent's script:
Can we follow up by text or WhatsApp about this? Reply YES to agree; message rates may apply; reply STOP anytime.
Store the answer on the contact record so anyone in the shared inbox sees it before sending. The workflow for continuing the thread on another channel is in From Website Chat to WhatsApp or SMS; this article is about the permission.
Sensitive data: health, money and co-browsing
People paste things into chat they should not: a policy number, a diagnosis, card digits. Three rules that hold across industries:
- Ask for less. Never request card numbers, full account numbers or health details in chat. Instruct the AI explicitly; train the team the same way.
- Move regulated steps out of chat. Payment goes to checkout; medical intake goes to the patient portal. The chat links to them.
- Mask during co-browsing. If an agent needs to see the visitor's screen to help with a form, co-browsing should hide passwords and form fields by default and let you mask more. In Omnifox the visitor accepts or declines each session, the agent cannot type into masked fields, sessions are capped at 10 minutes and a log is kept.
Health and financial sites may have sector rules on top of general privacy law. Those are outside what a chat vendor can settle: confirm with your own counsel.
Deletion, transcripts and retention
Give visitors a way to get their data and a way to remove it:
- Email transcript. A "send me a copy" button at the end of the chat gives the visitor their own record. It can be turned off per channel.
- Deletion on request. Verify the requester, delete or anonymize the contact and its conversations, confirm in writing. Check for legal holds (tax, disputes) first.
- Retention period. State one; pick the number with counsel.
- Staff access. Say who can read chats (support and sales teams, plus the AI agent).
Frequently asked questions
Do I need cookie consent to show a chat widget?
It depends on how you classify the widget's browser identifier and where your visitors are. Many sites treat it as functional and load it immediately; stricter setups load it after consent. Either way, send analytics events only where analytics consent exists.
Do I have to tell visitors they are talking to an AI?
Tell them. It sets expectations, and disclosure is increasingly expected or required. A one-line note in the welcome message ("AI assistant; a human can take over") is enough for most sites.
Can I text or WhatsApp someone who gave their number in the chat?
Only with clear consent for that channel. Ask a plain yes/no question when you collect the number, record the answer on the contact, and honor opt-outs. Automated or marketing texts in the US and UK carry specific rules; confirm with your own counsel.
What happens to chat data if a visitor asks for deletion?
Verify the request, delete or anonymize the contact and its conversations, and confirm. Check for retention obligations first — a dispute or tax record may need to be kept even if the chat is removed.
Say what you store, ask before you switch channels
Privacy in website chat comes down to a short inventory, a three-line notice and one yes/no question before you move to text or WhatsApp. Build those in from day one and the rest is routine. Set up a widget with a pre-chat consent field and a visible AI notice on a 7-day free trial, or see how the website chat channel fits the shared inbox.
Ready to try it in your business?
Bring WhatsApp, Instagram, Messenger, email and calls into one inbox, and let AI handle the first reply. Set it up today.
7 days free. No credit card required.
Read more
-

AI for Automated Lead and Opportunity Follow-Up
-

Multi-Agent WhatsApp: Why the Phone App Falls Short
-

WhatsApp CRM in South Africa: 2026 Guide to Sell More
-

How to Generate Leads With SMS Campaigns: A Practical Guide
-
Appointment Booking and Reminders for Canadian Businesses
-

WhatsApp Calling in Malaysia: Providers and Availability
-

How to Close a Customer Service Conversation
-

WhatsApp Template Categories (Utility, Marketing, Auth): Which to Use and When
Comments (0)
No comments yet. Be the first to share your thoughts.
Leave a comment
Your email is never published. Comments are moderated before appearing.