Do You Need an AI Chatbot for Your Small Business Website?
11 min read
An AI chatbot for a small business website is worth adding only after two things are true: visitors ask enough questions to justify it, and you've written the answers down. This lesson gives you the test, the free options, and four rules to follow if you go ahead.
Chatbot vendors quote impressive numbers about how many conversations their bots resolve. Those figures come from companies with thousands of chats a month and a documentation team. Your situation is different, and the honest answer for many small businesses is "not yet".
The two-question test
Do you get at least 20 website questions a week? Count contact form entries, "I was on your site and couldn't find" emails, and calls that start with "I saw online that". Below that volume a chatbot has almost nothing to do, and a clear FAQ plus a contact form serves people better.
Is your FAQ written? A modern AI chatbot answers from the content you feed it. No content, no answers, or worse, guessed ones. If you haven't done the last lesson, do that first. It's the prerequisite, and it helps even if you never add a bot.
Two yeses: read on. One or none: skip to "Start with live chat, no bot".
What changed since the old chatbots
The chat widgets you hated five years ago ran on decision trees. Someone had to script every branch, and anything off-script got "Sorry, I didn't understand that."
The current kind reads your FAQ, policies and product pages, then answers in natural language from that material. Setup is mostly uploading documents. That makes it practical for a small team, and it also raises the risk, because a bot that can phrase anything can phrase something wrong. You saw in Chapter 1 who carries the responsibility when it does.
The free options, honestly
Checked in September 2026:
| Tool | What's free | The catch |
|---|---|---|
| tawk.to | Live chat, ticketing, knowledge base, no seat limit. AI Assist has a free tier for sites with low chat volume | Its badge shows on the widget. AI add-on starts around $29 a month once you outgrow the free tier |
| Tidio | Live chat for up to 50 conversations a month, plus 50 conversations with its Lyro AI agent | The 50 AI conversations are a one-time trial, not a monthly allowance |
| Crisp | Chat widget and shared chat inbox for 2 seats | No AI and no knowledge base on the free plan |
| HubSpot free tools | Live chat with simple scripted bots, tied to your contact records | Scripted, not generative. HubSpot branding. Two users |
If your customers message you on WhatsApp more than they visit your site, the next lesson matters more than this one.
No free plan gives you an unlimited AI agent. Answering with AI costs the vendor money on every message, so the AI part is where free tiers stop. Budget roughly $25 to $50 a month if you decide a website bot earns its place. Set that against what it replaces: if it answers 30 routine questions a week at two minutes each, that's an hour of your time.
Start with live chat, no bot
Here's the option most lists skip. Install free live chat with no AI at all. tawk.to and Crisp both have phone apps, so a visitor's question reaches you like a text message. When you're busy or closed, the widget turns into a "leave a message" form.
Run it for a month. You'll learn what visitors ask, in their words, which is the raw material a bot would need anyway. And many owners find that ten chats a week, answered by a human in two minutes from a phone, convert better than any bot would. Your assistant still helps: paste the question and your notes, get a tidy reply, send.
Four rules if you add a bot
1. Known answers only. Feed it your FAQ, hours, service area, pricing pages and policies. Tell it, in its instructions, to hand off anything about refunds, complaints, quotes, medical or legal questions, and anything it can't find in its sources. Most tools have a setting for "only answer from my content". Turn it on.
2. It says what it is. First message: "I'm an automated assistant for [BUSINESS]. I can answer common questions, and I'll connect you with a person for anything else." Customers forgive a bot that's upfront. They don't forgive one that pretended.
3. A person is one tap away. A visible "talk to a human" option during hours, and after hours a promise with a time in it: "Leave your name and number and Sam will call you before 10 a.m. tomorrow." A bot with no exit is a trap, and trapped customers write reviews.
4. Read the transcripts every week. Fifteen minutes, every Friday. Look for wrong answers, questions it couldn't handle, and people who gave up. Each one is either a fix to your FAQ or a new handoff rule.
Test it before customers do
Before you switch it on, try to break it. Ask your assistant for the test set:
I'm about to launch a website chatbot for my [TYPE OF BUSINESS].
It should answer only from our FAQ and policies, and hand off
everything else to a person.
Write 25 test questions a real customer might type, including:
- 10 routine questions
- 5 worded badly or with typos
- 5 that try to get a promise we don't make (a refund outside
policy, a discount, a guaranteed date)
- 5 that must go to a human (a complaint, an injury, a legal
threat, a request for advice we're not qualified to give)
For each, say what a safe response looks like.Type all 25 into the bot. If it invents a policy or hands out a discount even once, tighten its instructions and run the set again. Launch when it passes.
Where this breaks
Bots fail quietly. Nobody emails to say "your chatbot gave me a wrong answer, so I went elsewhere". If you won't commit to the Friday transcript check, don't run one. A contact form that says "we reply within one business day", and then does, beats an unattended bot every time.
Try it now
Answer the two test questions with real numbers from last week. If you passed both, install one free tool from the table in live-chat-only mode and run it for two weeks before deciding on AI. If you didn't, put the hour into your FAQ instead.