How to create an AI chatbot for your small business website

How to create an AI chatbot for your website: decide if you need one, build or buy, set it up to be honest with customers, and measure whether it works.

AI & Automation · 18 October 2026 · 7 min read

Your team answers the same questions every day: delivery times, prices, opening hours, how to return an order. A chatbot looks like the obvious fix, and plenty of tools promise to have one running by the afternoon. But knowing how to create an AI chatbot for your website is mostly about deciding what it should answer, what it should never touch, and how you will know it is working. The software is the smaller part of the job.

You would not be following the crowd. The government’s UK Business Data Survey 2026 found that only 6% of small businesses and 7% of medium businesses use AI for customer service chatbots. There is little established practice to copy, and a badly run bot will be noticed by customers.

Check that a chatbot is the right fix

Before you buy or build anything, look at what the government itself recommends. Its guidance on using chatbots and webchat tools tells organisations to do user research first and to consider whether improving existing website content or navigation would be more cost-effective than a chatbot.

For a small business, that research is simple. Pull the last three months of emails, call notes, live chat logs and anything customers typed into your site search. Tally the twenty questions that come up most. Then ask one question of each: is the answer already on the site, and can a customer find it in under a minute?

If the answer exists but is buried three clicks deep, fix the page. That costs an afternoon rather than a monthly subscription. A chatbot earns its place when customers phrase the same need in a dozen different ways, or when they do not know what to search for. Delivery queries, returns, eligibility checks and “which product suits me” questions are typical examples.

This list also deals with the problem that holds many firms back. Data from the British Chambers of Commerce, reported by TechRadar, shows over one-third of UK small and medium-sized enterprises now actively using AI, up from about 25% a year earlier. The barriers they cite are unclear use cases, cost and lack of skills. A written list of questions gives you a use case. It also lets you judge the cost against the staff hours it might save.

Build or buy when you create an AI chatbot for your website

The same government guidance points organisations to the Service Standard and the Technology Code of Practice to decide whether to build or buy. You do not need those documents, but the logic carries over. What problem are you solving, who will run it, what will it cost to keep going, and can you switch if it disappoints?

For most small firms with a narrow set of questions, buying a ready-made chatbot tool and feeding it your own content is quicker and cheaper than a custom build. You pay a monthly fee and accept less control. A custom build makes sense only if the bot must connect to your own systems, such as checking an order or booking an appointment, and you have a developer who will maintain it. Building also makes security your responsibility, and that is a real cost.

Whichever route you take, put these questions to any supplier before you sign:

• Can I see and edit exactly what the bot has been told, and limit it to my approved content?
• What does it do when it does not know the answer, and can it hand over to a person?
• Where is my customers’ data stored, and is it used to train anyone else’s system?
• Can I export the conversation logs, and what happens to them if I leave?
• What are the total costs: set-up, monthly fee, and any charge that rises with the number of conversations?

Once you have chosen, restrict the bot to content you have written and approved: your FAQs, delivery and returns policies, and pricing pages. Do not let it improvise on anything involving money or promises. Then test it before launch with fifty real questions from your list, including the awkward ones: an angry complaint, a request for a refund outside your policy, a question you do not cover. How it fails matters more than how it succeeds.

Be honest with customers and always offer a person

The government’s guidance on using AI in services sets a clear standard for its own chatbots. Users must be told plainly that they are talking to AI and not a human, told how AI may affect their data or outcomes, and given the option to contact a human. Service owners must understand the technology and its risks, involve cyber-security professionals, and keep monitoring the service.

That guidance is written for public services, but it is a sensible bar for any business. In practice it means four things:

• Label the bot. Make the first message say it is an AI assistant, and give it a name that does not pretend to be a member of staff.
• Explain the data. Add one line, with a link to your privacy notice, on what happens to what customers type in.
• Offer a way out. Put a visible “talk to a person” option on every screen, with your phone number or email and your hours. Pass the conversation to your team so the customer does not have to start again.
• Get the security checked. If you have no in-house expertise, ask your IT provider to review what the bot can access before launch. It should see your public content and nothing more.

The human route is also your safety net. A customer who is told “I can’t help with that” and then finds no way to reach anyone will not blame the software. They will blame you.

Measure it, fix it and be ready to switch it off

Even a well-resourced pilot does not satisfy everyone. GOV.UK’s own chatbot pilot, GOV.UK Chat, focused on business content. According to the AI Playbook for the UK Government, nearly 70% of users found the responses useful and about 65% were satisfied with their experience. A substantial minority therefore did not rate the answers as useful. Expect the same from yours at first, and plan to close that gap.

Track four things from the first day. Count how many conversations end with a hand-off to a person. Note which questions the bot could not answer. Ask customers for a quick thumbs up or down at the end. Log any complaint that mentions the bot. For the first month, someone should read a sample of transcripts every week. Each missed question is either a gap in your content, which you can fill, or a topic the bot should not handle, which you should route straight to a person.

The government’s guidance also expects service owners to keep iterating and to retire the AI if necessary. Take that seriously. Decide in advance what would make you switch the bot off: repeated wrong answers on price or policy, say, or a rise in complaints you cannot fix within a few days. Knowing that you can pull it quickly makes it much easier to launch.

Start narrow. Pick one topic, such as delivery and returns, launch the bot on that alone, and judge it after a month. If customers get answers faster and your inbox is quieter, widen it to the next topic on your list. If not, you will have spent weeks rather than a year finding out.

Common questions

How much does it cost to put an AI chatbot on my website?

Building a custom chatbot typically costs far more than buying a ready-made tool, because you pay for developers and ongoing maintenance as well as the software. With a bought tool, ask for the monthly fee, any set-up charge and whether pricing rises with the number of conversations. Get this in writing and compare the total with the staff time you expect to save.

Do I have to tell customers they are talking to an AI chatbot?

Yes, you should. Government guidance for public services says users must be clearly told they are talking to AI rather than a human, and the same is sensible practice for any business. Say so in the chatbot’s first message, explain briefly what happens to the information typed in, and offer an easy way to reach a person.

Can an AI chatbot replace my customer service staff?

No, and it is risky to try. A chatbot suits repetitive questions such as delivery, opening hours and returns policy. Complaints, unusual cases and upset customers need a person. In the GOV.UK Chat pilot about 65% of users were satisfied, so a meaningful share were not. Use a chatbot to free up your team’s time, not to replace them.

Sources

• Government Digital Service and Central Digital and Data Office: Using chatbots and webchat tools
• GOV.UK Service Manual: Using AI in services
• GOV.UK: UK Business Data Survey 2026
• TechRadar: AI confidence gap is stalling UK small businesses
• GOV.UK: AI Playbook for the UK Government