AI Chatbot for Your Website: A UK Small Business Guide

Paper-cut illustration of a laptop showing a small shop website with a chat bubble holding a little robot, and a shop owner holding a sheet of paper with a tick
Paper-cut illustration of a laptop showing a small shop website with a chat bubble holding a little robot, and a shop owner holding a sheet of paper with a tick
Your chatbot speaks for your business, so check what it says. Illustrative artwork.

Short answer: A website chatbot can answer routine questions such as opening hours, delivery areas and how to book, at any time of day. Before you add one, decide what it may answer and what it must pass to a person, feed it only from pages you keep up to date, and tell visitors what happens to what they type. Treat every answer as your business speaking: in a 2024 Canadian tribunal case, Air Canada was held responsible for wrong information its website chatbot gave a customer. UK rules on personal data still apply, and the ICO's right-to-be-informed guidance means people should get privacy information when the chatbot collects their details. This is general information, not legal advice.

This is a documentation review of the ICO, ASA and tribunal pages listed under Sources, read on 18 November 2026. I haven't installed or tested a chatbot product for this page, and I don't quote prices because they change too often.

Chatbot, assistant or agent?

A website chatbot answers questions in a chat window. It becomes more like an AI agent once it can do things, such as take a booking, issue a refund or change an order. The more it can do, the more can go wrong. Our guide to the difference between an AI agent, a chatbot and an assistant explains the labels. If you want the phone answered rather than a chat window, see our AI receptionist guide.

What a chatbot is good for, and what to keep away from it

Good first jobs:

  • Opening hours, location, parking and delivery areas.
  • "How do I book?" and "How do I return something?", with a link to the right page.
  • Collecting a name and a question for a person to answer later.

Keep a person in charge of:

  • Prices, discounts and refunds, unless the bot only quotes a page you maintain.
  • Complaints, health, safety or anything legal.
  • Anything that commits the business to something, such as a delivery date or an exception to your policy.

Who is responsible when the chatbot gets it wrong?

Plan on the answer being "you". The clearest public example I've found is from Canada, not the UK, so treat it as a warning rather than UK law. In Moffatt v. Air Canada (14 February 2024), British Columbia's Civil Resolution Tribunal found the airline liable after its website chatbot gave a customer wrong information about bereavement fares. Air Canada argued, in effect, that the chatbot was "a separate legal entity" responsible for its own actions. The tribunal called this "a remarkable submission" and said: "It makes no difference whether the information comes from a static page or a chatbot."

Closer to home, the UK's four digital regulators wrote to ministers on 16 November 2026 naming "misleading recommendations, decisions or information" among the risks firms launching personal AI agents must address. That letter is aimed at agent providers, not at a café adding a chat window, but the direction is clear.

Practical steps that follow:

  1. Feed it from your own pages. Most chatbot tools let you point the bot at your website or a document. Keep those pages current, because the bot will repeat them.
  2. Give it a "don't know" answer. "I'm not sure, please email us at…" is better than a confident guess.
  3. Read a week of conversations after launch, and again after any change to prices or policies.
  4. Ask your supplier whether you can see and export transcripts, and how quickly you can switch the bot off. Our supplier due diligence checklist has the full list of questions.

UK GDPR: the checks before you switch it on

If the chatbot collects names, emails, order numbers or anything else that identifies a person, UK GDPR applies.

  • Tell people what happens to their data. The ICO's right-to-be-informed guidance says you must give people privacy information "at the time you collect their personal data from them", covering your purposes, how long you keep it and who it's shared with. A short line above the chat box with a link to your privacy notice is a common way to do it. The ICO notes this guidance is under review after the Data (Use and Access) Act.
  • Know where it's processed. Your chatbot provider, and the AI model behind it, will usually process what visitors type. Check the supplier's data processing terms and where data is stored; our UK data residency guide covers the big AI providers.
  • Decide how long you keep transcripts, and who in your business can read them.
  • Consider a DPIA if the bot will handle sensitive information or make decisions about people. Our AI policy and DPIA templates are a starting point, and our UK GDPR guide to AI agents explains what the ICO expects.

Should you say it's an AI?

Yes. I couldn't find a single UK rule that requires a chatbot label in every case ‡, but it costs nothing and avoids misleading anyone. The Advertising Standards Authority makes a related point about ads: there is "no blanket legal requirement in the UK to disclose the use of AI", but the usual rules against misleading people apply however content is made. If a visitor could reasonably think they're chatting with a person, tell them they aren't, and show them how to reach one.

Security: the bot reads what visitors type

Anyone can type anything into a public chat box, including instructions designed to trick the AI. Don't connect a public chatbot to tools that can pay, refund, delete or email customers without a person approving. The NCSC's advice on prompt injection is summarised in our AI agent security checklist.

Cookies: the chat widget may need consent

Added 18 November 2026. Many chat widgets are scripts from the chatbot company that store cookies or similar data in the visitor's browser ‡. The ICO's cookies guidance says you must get consent before storing a non-essential cookie, through a clear action such as ticking a box or clicking a link. The exemption covers only cookies that are strictly necessary to provide the service the visitor asked for. Cookies that are merely helpful, or that serve your own purposes, still need consent. So:

  • Ask your supplier which cookies or local storage the widget sets, what each is for, and whether any are third-party.
  • Add them to your cookie banner and policy, and don't load non-essential ones before the visitor agrees.
  • Consider loading the widget only when the visitor clicks a chat button. Whether that makes its cookies strictly necessary is a judgement for you and your adviser ‡.

The ICO says it is consulting on revised cookies guidance, so check its page before you rely on this.

A one-page plan

  1. Write the ten questions customers ask most, and the page that answers each.
  2. Choose a tool that can answer only from those pages and can say "I don't know".
  3. Add an "AI assistant" label, a link to a person, and a privacy line above the chat box.
  4. Check the supplier's data terms, transcript access and off switch.
  5. Read the first week of conversations, then monthly.

Frequently asked questions

Do I need to tell website visitors they're talking to an AI chatbot?

I couldn't find a single UK rule that requires a label in every case, but it's sensible to say so. UK advertising rules on misleading people apply however content is made, and if your chatbot collects personal data, the ICO's guidance says people must get privacy information when you collect it. This is general information, not legal advice.

Am I responsible if my website chatbot gives a customer wrong information?

Plan on it. In Moffatt v. Air Canada (2024), a Canadian tribunal held the airline liable for wrong information its website chatbot gave a customer, saying it made no difference whether the information came from a static page or a chatbot. That's a Canadian decision, not UK law, but it's a clear warning to check what your chatbot says.

Is a website chatbot covered by UK GDPR?

Yes, if it collects or processes information that identifies a person, such as a name, email address or order number. You need to tell people what happens to their data when you collect it, check your supplier's data processing terms, and decide how long you keep transcripts.

What should a small business chatbot not do?

Don't let it make commitments a person hasn't approved, such as refunds, discounts or delivery dates, and don't connect it to tools that can pay, delete or email customers without a person approving. Pass complaints, health, safety and legal questions to a person.

Sources

  • Civil Resolution Tribunal (British Columbia), Moffatt v. Air Canada, 2024 BCCRT 149, issued 14 February 2024, read 18 November 2026: https://decisions.civilresolutionbc.ca/crt/crtd/en/525448/1/document.do
  • ICO, "Right to be informed", opened 18 November 2026: https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/individual-rights/individual-rights/right-to-be-informed/
  • ICO, "Artificial intelligence" guidance hub, opened 18 November 2026: https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/artificial-intelligence/
  • ASA, "Disclosure of AI in Advertising: Striking the Balance Between Creativity and Responsibility", opened 18 November 2026: https://www.asa.org.uk/news/disclosure-of-ai-in-advertising-striking-the-balance-between-creativity-and-responsibility.html
  • DRCF, "Personal AI Agents" (16 November 2026 joint letter), opened 17 November 2026: https://www.drcf.org.uk/personal-ai-agents
  • NCSC blog, "Prompt injection is not SQL injection (it may be worse)", 8 December 2025, opened 18 November 2026: https://www.ncsc.gov.uk/blog-post/prompt-injection-is-not-sql-injection
  • ICO, "Cookies and similar technologies" (Guide to PECR), opened 18 November 2026: https://ico.org.uk/for-organisations/direct-marketing-and-privacy-and-electronic-communications/guide-to-pecr/cookies-and-similar-technologies/

‡ legend: ‡ = not stated on an official page I opened, or my own reading rather than the source's words. Check before relying on it.

Last checked: 18 November 2026 (Europe/London). Evidence: documentation review. No chatbot product was installed or tested. General information, not legal advice.

Scroll to Top