What an AI chatbot on your website can and can’t do, when an off-the-shelf tool is enough and when a custom build makes sense, and the guardrails, privacy points and checks to put in place before launch.
An AI chatbot for your website is worth it when visitors keep asking questions your own pages already answer, and when you can keep the bot to those answers. Done well, it answers common questions at any hour from your own content and passes everything else to a person. Done badly, it invents prices, makes promises you can’t keep and puts off the customers it was meant to help. Most of the difference comes from what you give it to work from and the limits you set, not from which tool you buy.
What an AI chatbot for your website can and can’t do
A modern AI chatbot reads a visitor’s question in their own words and replies in plain language, using the content you give it. Good uses:
- Answering repeated questions about your services, opening hours, the areas you cover and how to get started.
- Finding the right page on a large site, or the right section of a long policy.
- Gathering the details for an enquiry, such as what someone needs and the best way to reach them, and passing them to a person.
- Explaining a process in simpler words, such as what happens after someone books.
What it shouldn’t do:
- Quote prices, discounts or terms that aren’t in your approved content.
- Give advice that needs a qualified person, such as legal, medical, tax or financial advice.
- Make decisions about a customer, such as approving a refund or an application, without a person checking.
- Pretend to be human.
Off-the-shelf tool or custom build?
Off-the-shelf chatbot tools let you point the bot at your website, upload a few documents and add a chat widget to your pages with little technical work. That is the right choice when:
- Your questions are fairly standard and the answers are already on your website or in a few documents.
- You want to find out whether visitors use a chatbot at all before investing more.
- The tool lets you limit answers to your content, hand over to a person, and export or delete conversations.
A custom build makes sense when:
- The bot needs to work with your own systems, for example to check a booking, an order or a case.
- You handle sensitive information and need control over where conversations are stored and who can see them.
- The off-the-shelf tools can’t keep to your rules reliably, or can’t hand over the way your team works.
- The chatbot is central to how customers deal with you, not a nice extra.
Who shouldn’t buy either: if your website doesn’t yet answer the basic questions, a chatbot will only repeat the gaps, so fix the pages first. And if you get a handful of enquiries a week, a clear FAQ page and a quick reply from a person may serve you better.
What to train it on
Most chatbot tools don’t change the AI model itself. They search the content you provide for the passages that match a question and answer from those. So the real work is choosing and tidying that content:
- Your own website pages. Services, contact details and anything visitors often ask about.
- Your FAQs. Written as the questions customers actually ask, with short, definite answers.
- Your policies. Delivery, returns, cancellations, complaints and privacy, in their current versions.
- A list of things it must not answer. With the wording it should use instead, and who it hands over to.
Leave out drafts, old price lists, internal notes and anything you wouldn’t publish. If a document is out of date, the bot will repeat it with confidence. Someone should own the content and update it whenever prices, services or policies change.
Among the work I led as Head of AI & Automation at Alter Domus were knowledge agents that answered support questions from internal documentation. The lesson carries over to a website: the answers can only be as good, and as current, as the documents behind them.
Guardrails that keep it honest
- Answer only from your content. If the answer isn’t there, the bot should say so and offer a person, not guess.
- No made-up prices or promises. Prices, delivery dates and guarantees come only from approved content, or not at all.
- An easy way to reach a person. Visible at every point, with an honest message about when someone will reply.
- Say it’s an AI. Tell visitors at the start that they are talking to an AI assistant. It sets expectations, and if people in the EU use your site, the EU AI Act’s transparency rules may apply too: see what the EU AI Act means for UK businesses.
- Stay on topic. It should politely decline questions unrelated to your business, and resist attempts to make it say things it shouldn’t.
- Regulated firms. Agree with your compliance team what it may say about products, and keep it well away from anything that could read as advice.
- Keep logs, and read them. Review conversations regularly so wrong answers are caught early.
UK GDPR, cookies and consent
A chatbot collects personal data, often more than you expect, because visitors type names, email addresses and sometimes sensitive details into it. This is practical guidance, not legal advice, but these are the points to cover:
- Your privacy notice. Explain that the chatbot exists, what it collects, why, who provides it and how long conversations are kept. UK GDPR applies to what people type into it, and the ICO expects you to be clear when the chat is provided by a third party.
- A contract with the vendor. If a chatbot provider processes personal data for you, the ICO says there must be a written contract between you. Check theirs, and ask where conversations are stored and whether they are used to train anyone’s models.
- Cookies and similar technologies. The ICO’s guidance on storage and access technologies, finalised in April 2026, says that when a well-informed user chooses to use a chatbot, the storage it needs can count as strictly necessary. That covers only what the chat itself needs. If the widget also stores information for other purposes, such as usage monitoring or personalisation, you need consent, and nothing that needs consent should be set before the visitor gives it.
- Sensitive information. People will type health details, money worries or complaints into a chat box. Decide how the bot responds, who can read transcripts and how long you keep them.
How to measure whether it helps
Decide what success looks like before launch, and measure it with your own numbers:
- How many conversations end with a correct answer, checked by reading a sample every week.
- How many are handed to a person, and why. Repeated hand-offs on one topic usually mean a page needs improving.
- Enquiries or bookings that started in the chat.
- Fewer repeated emails and calls about the questions it covers.
- Wrong answers and complaints. One confident wrong answer about a price can cost more than the bot saves.
If it isn’t helping after a fair trial, switch it off. A clear FAQ page is a perfectly good outcome.
Pre-launch checklist
- Content chosen, current and owned by a named person.
- Answers limited to your content, with a fallback when it doesn’t know.
- No prices, promises or advice beyond approved wording.
- A visible route to a person, tested end to end.
- Visitors told at the start that it’s an AI assistant.
- Privacy notice updated and the vendor contract checked.
- Cookie behaviour tested before and after the chat is opened.
- Tested with real questions, including awkward, off-topic and hostile ones.
- Transcripts reviewed on a schedule, with a quick way to correct content.
- A date to decide whether it is earning its place.
I design and build business websites, and where it helps, a site can include an assistant that answers questions from your own content: see websites and digital products. A chatbot that needs to connect to your own systems is an agents and automation project, built on the same responsible AI principles for data and human review. If you are weighing up a chatbot, book a free 30-minute AI consultation and we can look at whether it would actually help your visitors.