For most small local businesses, an AI chatbot is worth it only if it answers a short list of real questions and still makes it easy to phone or send a form. It is not worth it if it sits in front of a customer who was ready to call. A form, a tappable number and clear copy on the page still win more work than a clever widget. Treat a bot as an extra, not as the way in.
This is about the box on your website, not about using ChatGPT in the office. The wider habit of drafting and summarising is in AI for small business.
What can an AI chatbot actually do on a small site?
A modern bot can read a plain question and reply from the pages and notes you give it. That is a real step up from the old scripted widgets that only understood five buttons. In practice, a well-fed bot can:
- Answer hours, areas covered, lead times and what is included.
- Point to the right page, the phone number or the form when the person is ready.
- Take a name, a number and a short description of the job after hours.
- Repeat the same three answers all week without getting bored.
It cannot visit a roof, price a rewire from a photo, or know you are booked until Thursday unless you told it. It will still sound sure when it is guessing. If the site already answers those questions in ordinary English, a lot of the “need” for a bot disappears.
When is a chatbot genuinely useful?
It earns its keep in a few narrow cases.
You get the same simple questions, often. A garden designer who is asked “do you do small courtyards” twenty times a month can put that answer in the bot and on the page. A one-off odd question is not a reason to buy software.
People write at 9pm. If your inbox already fills overnight, a bot that captures a name and a job description can stop that person bouncing to whoever answers first in the morning. Capture is useful. A long chat that never leaves a number is not.
You have enough traffic to justify the upkeep. A bot that nobody opens is a monthly fee and a page that looks busy. If you get a handful of visits a week, spend the time on the phone number and the first two paragraphs instead.
Someone owns the answers. The bot is only as good as the last time you updated it. Prices change. Areas change. You stop doing a service. If nobody will edit the source notes, skip the widget.
Useful also means measurable. If you cannot say, after a month, how many chats became a call or a form, you do not know if it paid. Count those, not “engagement”.
When is a form and a phone number better?
Most local trades and many small shops should start here and stay here.
People hiring for a leak, a consumer unit or a quote often want a human on the first pass. A chat that asks three qualifying questions before it shows the number feels like a call centre. They leave.
A form says: tell us the job, we will come back. A phone number says: we answer. Both beat a bot that half-answers and then asks for an email you will not check until Monday.
Use a form and a phone when:
- The job needs a site visit before any price is real.
- Customers already call, and you want more of that, not less.
- You cannot staff the chat inbox the same day.
- The answers depend on what is behind the plaster, not on a FAQ.
You can still add a bot later. You cannot un-lose the people who closed the tab because they could not find the number behind a popup.
What makes a chatbot annoy customers?
The failures are boring and common.
It hides the phone number. Any design that treats chat as the “modern” way and the call as a last resort will cost you work. Keep the number in the header, on mobile, and in the bot’s first reply.
It cannot answer, and says so badly. “I’m not sure, please email us” after four messages is slower than a form. If you do not have a reliable answer, do not put the question in the bot’s brain. Send the person to the form on the first miss, not the fifth.
It invents. AI will fill gaps. A bot that quotes a price you do not offer, or claims you cover a town you do not, becomes your problem when the customer repeats it. Short, approved answers only. If a fact is not on the site, the bot should not invent it.
It feels like a wall. Long qualification trees and a refusal to show a human option read as cost-cutting.
It nags. A slide-in on the second scroll, on a phone, while someone is reading your prices, is not helpful. Delay it. Let people dismiss it.
The test: does this make contacting you easier than last month? If not, turn it off.
How should you set one up if you still want one?
Do the unglamorous work first.
Write a one-page list of questions you already answer by phone. Hours, area, typical lead time, what you do not do, how to book, how you price (visit first, or a published band). Put those answers on the site in plain English. Then, and only then, feed the same text to the bot. Do not give it your entire blog and hope.
Keep the first bot message short. Offer the phone number and the form in that first message. Allow a human handoff that creates an email or a CRM note you will see.
Review the chat log once a week for a month. Every wrong answer is a page you should write, or a line you should remove from the bot. Every chat that ends with “just call me” is a hint that the widget is in the way.
Do not paste customer addresses, photos of interiors, or account details into a public bot tool if you would not put them on a postcard. The same caution as any public AI tool applies. The office-side version of that habit is in the AI post linked above.
If you care about assistants recommending you, that is a different job. Clear pages and consistent details help ChatGPT and Google describe you. A chat widget does not. That work is AI optimisation and GEO, not a popup.
What should you do instead of buying a bot first?
Put the number where a thumb can hit it. Answer the five questions people actually ask, on the page they land on. Keep the form short. Reply the same day, even with “we will look at this tomorrow”. A slow page with a chatbot is still a slow page.
A bot can wait until those are true and repeat questions still arrive after hours. Then you are buying a capture tool, not a personality for the brand.
Page Forge will give you a straight view of that order as advisory work, quoted per project. We do not run the bot afterwards, and we are not a helpdesk. If the honest answer is “do not install one”, that is the answer.
Frequently asked questions
Do I need an AI chatbot on my small business website?
Usually not as a first job. A fast site, a visible phone number and a short form matter more for most local firms. Add a chatbot only when you have a short list of stable answers and someone who will keep them true.
Will a chatbot lose me phone calls?
It can, if it hides the number or forces a chat before a call. For many local jobs, people want to speak to you. If you use a bot, put the number in the header and in the first bot reply so the widget adds a route rather than removing one.
Are AI chatbots better than the old scripted ones?
Yes, they handle plain questions more naturally. That does not change the test. A fluent wrong answer is worse than a stiff correct one. You still have to feed it facts and give people a human option.
Can a chatbot book jobs in my diary?
Not in a way I would trust unsupervised. Diaries, prices and access details need a person. A bot can collect a preferred window and a description. You confirm. If a vendor promises “fully automated booking” for site work, treat that as a sales line.
How do I know if mine is working?
Count chats that left a real contact and became a job or a booked call. Ignore time-on-widget. If a month of chats produced nothing you would have missed from the form, switch it off and keep the form.
Should I put a chatbot on the site to look modern?
No. People hiring a trade or a local shop do not award points for a widget. They award points for a clear offer and a way to reach you. Looking modern is a poor reason to sit between them and the number.
If you want a plain view of whether a chatbot would help, or whether the time is better spent on the page and the form, that is a short advisory piece of work. See IT and AI consulting, or get in touch. Bring the questions customers already ask. That list decides it.