The useful question
The question is not "should my business use AI". It is "which specific tasks in my week are repetitive, low-judgement, and currently eating hours". Those are the ones worth handing over. Everything else is a demo.
We build this into sites for a living, so treat what follows accordingly — but the list of things we recommend against is longer than the list we recommend, which should tell you something.
Where it genuinely earns its place
Answering the same six questions at 11pm. Every service business gets the same handful of questions: do you cover my area, what does it roughly cost, how soon can you come, do you do this specific thing. A well-configured assistant answering those out of hours — from your actual answers, not invented ones — keeps people engaged instead of moving to the next search result.
Turning a rambling enquiry into a structured one. Extracting the job type, urgency, address, and budget from a free-text message so the alert that reaches your phone is readable in three seconds.
First drafts of routine writing. Service descriptions, follow-up emails, review responses. A draft you edit is faster than a blank page. Publishing the draft unedited is where this goes wrong.
Triage and routing. Deciding whether something is an emergency, a quote request, or a supplier email, and sending it to the right place. Low-judgement classification is exactly what these systems are good at.
Summarising what happened. Turning a week of enquiries into "eleven enquiries, six for gutter work, four came in after 6pm". Useful, and nobody was ever going to compile it by hand.
Where it quietly costs you work
Quoting. An AI that estimates a price on your behalf will eventually give a number you cannot honour, and you will either eat the difference or start the relationship by revising the price upwards. Neither is recoverable. Give ranges and a mechanism, never a firm quote.
Anything where being wrong is expensive. Compliance questions, safety advice, whether a product is suitable for a specific situation. These systems produce confident wrong answers, and confident wrong answers about gas fittings are a different category of problem from a typo.
Pretending to be a person. A chatbot with a stock-photo face and a first name is a trust problem waiting to surface. Say it is automated. People are fine with automated; they are not fine with having been fooled.
Bulk content. Publishing hundreds of near-identical generated pages is the exact pattern search engines have built policies to catch, and it devalues the pages you did write properly. Volume without substance is not a strategy.
Handling complaints. Someone who is already annoyed and gets a bot has been told exactly what you think their problem is worth.
The guardrails to insist on
If you put an assistant on your site, these are not optional:
- It answers from your material only — your services, your areas, your policies — and says "I do not know, here is how to reach a person" rather than improvising.
- It is clearly labelled as automated in the first message.
- There is always a visible route to a human, on every screen, without asking for it twice.
- It never states a price it cannot stand behind.
- Conversations are logged and readable, so you find out what it has been telling people.
- You have read a week of transcripts before trusting it. This step is skipped constantly and it is where every unpleasant surprise lives.
What happens to the customer data
Anything typed into a chat window on your site is data you are now responsible for — names, phone numbers, addresses, sometimes details about someone's home or health.
Ask any vendor three questions and get the answers in writing: where is the data stored, is it used to train anyone's model, and how long is it kept. If the answers are vague, that is your answer. Your privacy policy needs to reflect whatever the honest answers turn out to be.
A sensible starting point
Pick the single most repetitive thing in your week and automate that. For most service businesses it is answering "do you cover my area and roughly what does it cost" for the four hundredth time.
Run it for a month. Read every transcript. Count how many conversations ended with a real enquiry and how many ended with someone giving up. If the numbers are good, add a second use. If they are not, turn it off — and notice that switching it off cost you nothing, which is exactly why starting small was the right call.
Questions people ask about this
Will an AI chatbot hurt my SEO?
The widget itself will not, provided it loads lazily and does not slow the page down. What does cause harm is generated content published at volume without editing — that is a separate decision, and a much riskier one.
Can it book appointments directly?
Yes, and this is one of the better uses, as long as it writes into your real calendar and respects your actual availability. Keep it to slot selection rather than negotiation about scope or price.
Is it worth it for a business with few enquiries a week?
Usually not for the chat side. At low volume the honest answer is that answering the phone quickly does more. The summarising and drafting uses can still save you time regardless of enquiry volume.
How do I stop it saying something wrong about my business?
Restrict it to your own documented answers, give it an explicit "I do not know" response, and read the transcripts. Restriction plus honest failure beats coverage plus confident invention, every time.
