A visitor opens the chat window on your site and types: “My son has a fever and a rash, can the doctor see him today? My number is…” In one message you now hold a child’s health details, a parent’s phone number and an implied location. That message has also travelled through at least one company you have never spoken to.
Chatbot data privacy is mostly about knowing where that message goes, who can read it, how long it stays, and what you have told the visitor about all of that. This post walks through those questions in plain terms. It is general information, not legal advice; if your business handles sensitive data, speak to a lawyer who knows the laws that apply to you.
Follow one message from start to finish
The clearest way to understand the risks is to trace a single chat message. For a typical hosted chatbot on a small business site, it goes roughly like this:
- The visitor’s browser. The chat widget is a script loaded from the chatbot vendor, not from your own server.
- The vendor’s servers. The message is sent to the vendor, stored in their database and shown in your inbox.
- An AI model provider. If the bot uses AI to reply, the message (and passages from your content) are usually sent to a model provider, which may be a different company from the chatbot vendor.
- Your connected tools. Leads may be pushed to your email, a Google Sheet, a CRM or WhatsApp.
- Your team’s devices. Staff read transcripts on laptops and phones, sometimes on personal ones.
Each stop is a place where the data is stored, copied or readable by someone. Draw this map for your own setup, with real company names. It does not take long, and it answers many of the questions below before you ask them.
Questions to ask a chatbot vendor
Most of these answers are in a vendor’s privacy policy, terms or data processing agreement. If you cannot find them, ask in writing.
Where is chat data stored?
Which country or region, and on whose infrastructure. Some businesses, or their clients, need data held in particular regions.
How long is it kept?
Look for a retention setting. Keeping every transcript forever is rarely necessary and increases what could be exposed in a breach.
Is chat content used to improve the vendor’s or model provider’s AI?
Policies differ between vendors and between plan levels, and they change. Find the current statement, and look for a setting to opt out if that matters to you.
Who are the sub-processors?
Many established vendors publish a list of the other companies that handle data on their behalf, including any AI model provider.
Can I export and delete data?
You need to be able to remove one person’s conversation on request, and to take your data with you if you leave.
Who at the vendor can see my chats?
Support staff access, logging, and how access is controlled.
What the visitor should be told
Being open with visitors is both fair and, under many privacy laws, expected. Practical steps:
- A short notice in the chat window before the first message, such as: “This chat is handled by an automated assistant. Please don’t share health, financial or ID details here. See our privacy policy.”
- Your privacy policy updated to mention the chat tool, what it collects, why, and the third parties involved.
- Clear purpose when asking for details. If the bot asks for a phone number, it should say why (“so the team can call you back about your enquiry”).
- Say it is a bot. Do not give an automated assistant a human name and photo that leaves people thinking they are talking to a person.
For Indian businesses, the Digital Personal Data Protection Act shapes how notice and consent should work. Our explainer on what the DPDP Act means for a business website covers the basics, and the post on consent on website forms applies directly to a chatbot that collects contact details. If you have visitors from the EU or UK, GDPR may also apply.
Collect less, on purpose
The simplest privacy protection is not collecting data in the first place. Look at your conversation flow and ask, for every question the bot asks: do we need this to help the person?
- A name and one contact method is usually enough for a callback.
- A full address is rarely needed at the enquiry stage.
- Date of birth, ID numbers, account numbers and health details should not be collected through a general website chat at all.
- If visitors volunteer sensitive information anyway, the bot can be instructed to reply without repeating it, and to suggest a safer channel.
The visitor-facing warning matters, but your own conversation design matters more. A bot that asks “Tell us your symptoms” is inviting sensitive data. When planning an ai chatbot for business use, trim the list of questions the bot asks before you polish anything else.
Chatbots and customer data inside your business
Chat data can leak just as easily on your side as on the vendor’s. Review how your team handles it:
- Separate logins. Each person who reads chats should have their own account, not a shared password. When someone leaves, remove their access the same day.
- Two-factor authentication on the chatbot account, the email it sends leads to, and any connected spreadsheet or CRM.
- Least access. Not everyone needs to read every transcript, and not everyone needs admin rights to change integrations.
- Exports. Downloading transcripts to a laptop “for reference” creates a copy nobody tracks. Avoid it, or delete exports when done.
- Personal phones. If staff use the vendor’s mobile app on personal phones, decide what happens when the phone is lost or the person leaves.
What you put into the bot matters too
Privacy runs in both directions. The bot answers from content you provide, and anything in that content might be shown to a visitor. Keep internal documents, customer lists and staff contact details out of the knowledge base. The guide on preparing website content for a chatbot covers what to include and what to leave out.
The widget itself: scripts and cookies
Even before anyone types, a chat widget is third-party code running on your pages. It may set cookies or use browser storage to remember returning visitors, and it sends basic technical information such as the page address and browser details to the vendor.
- Check the vendor’s documentation for which cookies the widget sets and what they are for.
- If your site uses a cookie banner, decide whether the chat widget should wait for consent or is treated as necessary for a service the visitor asked for. This depends on the laws that apply to you and how the tool behaves.
- Load the widget only from the vendor’s official plugin or snippet, and remove old chat scripts when you change tools, so data does not keep flowing to a service you no longer use.
Handling a request to see or delete chat data
Sooner or later someone will ask what you hold about them, or ask you to delete it. Prepare before it happens:
- Know how to search transcripts by email or phone number in your chat tool.
- Know every place that conversation was copied: email, sheet, CRM, WhatsApp.
- Delete or correct it in each place, or record why you must keep it.
- Reply to the person and note the date.
A separate guide on handling a customer’s data deletion request goes through this in more detail.
Sectors that need extra care
Chatbot data privacy gets harder when the subject itself is sensitive. Clinics, diagnostic centres, financial advisers, lawyers, schools and anyone dealing with children should be especially cautious. In these sectors it is often better to use the chatbot only for general information (timings, services, how to book) and route anything personal to a phone call or a secure channel designed for it. Sector rules may also apply on top of general privacy law.
A chatbot data privacy checklist
- Data-flow map drawn, with every company named.
- Vendor’s storage location, retention, AI training use and sub-processors checked.
- Retention set to the shortest period that works for you.
- Chat window notice in place, and the bot identifies itself as automated.
- Privacy policy updated to cover the chat tool.
- Bot collects only what is needed, and steers away from sensitive data.
- Individual logins with two-factor authentication.
- Knowledge base free of private documents.
- A tested process for finding and deleting one person’s data.
None of this needs expensive tools, but it does need someone to decide it deliberately. If you are setting up a bot and want a second pair of eyes on these choices, ask us to review your chatbot plan; setup work is quoted on request.
Next in this series: Live Chat Plugin for WordPress: Choose, Add and Fix
