AI receptionists and website chatbots: what a small business can trust them with

The phone rings while you are up a ladder, with a client or closed for the night. By the time you ring back, the caller has tried someone else. A website enquiry that arrives at ten at night waits until morning, and by then some of those people have booked elsewhere.

An AI receptionist or a chatbot on your website promises to answer every one of them, at any hour. Part of that promise holds up. The real question is not whether the technology works. It is which conversations you can safely hand over, and which still need you.

What an AI receptionist actually does

There are two common kinds, and they work the same way underneath.

• An AI receptionist, sometimes sold as an AI phone answering service, picks up the calls you cannot take, talks to the caller in ordinary speech, and sends you a summary or a transcript afterwards. Some can book straight into your diary.
• A website chatbot does the same job in writing, in a chat box on your website.

Both answer from information you give them: your services, the area you cover, your hours, your prices and your policies. Underneath sits a general AI model that is very good at sounding natural. That is its strength and its risk: it answers smoothly whether or not it knows the answer.

What they do well

• Answering the questions you hear every day. Opening hours, the areas you cover, whether you do a particular job, how to book. If the answer is already on your website and rarely changes, an AI assistant can give it at two in the morning as easily as at two in the afternoon.
• Taking details properly. Name, number, what they need and how urgent it is, captured the same way every time, instead of a rushed voicemail with half a phone number in it.
• Booking, where it is connected. If it can see your real diary, it can offer real slots. If it cannot, it should take the request and leave the confirming to you.
• Covering the hours you cannot. Evenings, weekends and the middle of a job. A caller who gets an answer and a clear next step has a reason to wait for you. A voicemail beep gives them none.
• Sorting. Flagging the urgent call, so it is the first thing you see in the morning.

What these have in common: none of them commits you to anything a person has not already decided.

Where they go wrong

• Wrong answers, given confidently. Asked something it does not know, an AI assistant may fill the gap with something plausible: a price you do not charge, a slot you do not have, a refund policy you never wrote. The caller has no way to tell.
• Callers who need a person. Someone upset, someone confused, someone with an unusual job, or someone who simply wants to speak to a human. Making them argue with a machine first spends goodwill you worked hard for.
• Complaints. A complaint handled well can keep a customer. One answered by something that does not understand it rarely does.
• Anything sensitive. Health details at a clinic, a customer in financial difficulty, a legal matter, a bereavement. These need judgement and discretion, and a careful decision about where the information ends up.
• Mishearing. A name, a postcode or a phone number misheard on a bad line is a lost enquiry that looks like a captured one.

Your business is responsible for what it says

On 14 February 2024, the Civil Resolution Tribunal in British Columbia, Canada, decided a small claim against Air Canada. A customer had asked the airline's website chatbot about bereavement fares, and it told them they could claim the reduced fare after buying their tickets. The airline did not allow that. Air Canada argued it could not be held liable for what its chatbot said.

The tribunal rejected that. The chatbot was part of the airline's website, and the airline was "responsible for all the information on its website", whether it came from an ordinary page or a chatbot. The airline had not explained why a customer should have to check one part of its website against another, and it was ordered to pay the customer the difference in fare.

It is a Canadian small claims decision, not UK law. But it is how your customers will see it too: if your chatbot or AI receptionist said it, your business said it.

How to decide what to hand over

Sort the conversations your business has into three groups.

1. Let it answer. Facts that are already on your website and rarely change: hours, areas covered, services, how to book, where to park.
2. Let it collect, and a person decides. Anything that commits you to a price, a date or a promise: quotes, availability it cannot see, exceptions, refunds. It takes the details; you confirm.
3. Send straight to a person. Complaints, emergencies, anything sensitive, anyone upset, and anyone who asks for a person.

Then set a few house rules before it goes live:

• It says it is automated, at the start, and tells people how to reach a person.
• It answers only from your information. If the answer is not there, it takes details instead of guessing.
• It never gives a price you have not given it.
• You test it first. Ask it the questions you hear most often, then the awkward ones, and read exactly what it says.
• Someone reads the conversations. Every week at first, so a wrong answer is found by you rather than by a customer.
• Changes happen in one place. When a price or a policy changes, the assistant's information changes the same day as your website.

The data protection questions to ask first

An AI receptionist has to process what a caller says in order to answer at all, and many keep a recording or a transcript. A chatbot keeps what people type. That is personal data, and the Information Commissioner's Office (ICO) sets out what UK law expects.

• Tell callers when a call is recorded. The ICO's guidance on monitoring calls says you must tell the people on a recorded call that it is being recorded and why, and calls a recorded message good practice. The rest of your privacy information can follow, for example as a link to it on your website.
• Tell people what happens to their details. Under the right to be informed, people must be told, when you collect their details, why you are using them, how long you keep them and who they are shared with. Check that your privacy notice covers the AI tool.
• You stay responsible for the supplier. A supplier that handles callers' details on your behalf is what the law calls a processor. The ICO says you must only use a processor that can give sufficient guarantees, put a contract in place that meets the law's requirements, and keep checking it.
• Take extra care with health details. Information about a person's health is special category data, which UK law protects more strictly. If callers might describe their health, decide before go-live whether the assistant should take those details at all.

Before you sign up, ask the supplier where calls and chats are processed and stored, how long recordings and transcripts are kept, whether your customers' conversations are used to train its AI, and how you delete them. When we checked on 6 November 2026, the ICO said much of this guidance was under review following the Data (Use and Access) Act, so look for any update.

Often the first fix is the website

Many missed calls come from people who wanted something simple: to check a price, ask whether you cover their area, or book. If your website answers those questions plainly and lets people book or enquire at any hour, a good share of those calls never need making. An AI assistant then has less to do, and less to get wrong.

At The Anchor, our own venue, booking used to mean a phone number during service. The fix was not something to answer the phone. It was a booking route on the website that turned interest into a confirmed table, with confirmations and reminders after it. Table bookings grew by 403%. The Anchor is a pub, but every business that loses work to an unanswered phone has the same problem. The full case study explains the changes.

For the wider picture of where AI pays in a small business and where it is a distraction, read AI for small businesses.

Where we fit

We do not name or rank AI receptionist products: there are many, they change quickly and we have not tested them all. Judge any of them against the three groups and house rules above.

What we do is add AI to a client's website, or the systems behind it, where it takes real work off the owner: sorting enquiries, drafting replies for a person to check, organising customer details. Our AI integration work starts by agreeing the task, the information it can read and which answers a person checks, before anything is built. If a clearer page or a booking form would do the job, we will say so.

Building an AI feature is a project, scoped and quoted on its own. Once it is live, Care with AI is £100 a month plus VAT, which replaces standard Care and includes fair-use AI running costs.

If your business is in Staines or Spelthorne

Orange Jelly is based in Stanwell Moor, Staines, and businesses nearby are the first we look to work with. If yours is in Staines or Spelthorne, here is how we work with local businesses.

The short version

• An AI receptionist or website chatbot is good at the routine: common questions, taking details, booking into a connected diary and out-of-hours cover.
• It goes wrong when it guesses, and when the caller needs a person: complaints, anything sensitive, anyone upset.
• Let it answer facts you publish, let it collect details for anything that commits you, and send the rest to a person.
• Whatever it says, your business said. Test it first and keep reading its conversations.
• Tell callers when calls are recorded, check how your supplier handles their details, and take extra care with health information.
• Often the first fix is a website that answers the common questions and takes bookings at any hour.