AI Chatbot for Your Website: What It Costs, What It Can’t Do, and When It’s Actually Worth It

Most US small businesses can run a capable AI website chatbot for $30 to $150 per month on an off-the-shelf platform, plus a one-time setup cost of roughly $500 to $5,000 if someone else builds the content and integrations for you. A genuinely custom chatbot — one wired into your booking system, inventory, or CRM with logic no platform offers — starts in the five figures and is rarely justified below a few hundred conversations a month. And a chatbot is worth it only when three things are true at once: you get enough website traffic that inquiries are being missed, you have a repeatable set of questions with real answers, and someone is available to pick up the conversations the bot can’t finish.

That last condition is where most small-business chatbots fail. Nobody sells you that part.

What a website chatbot actually costs in 2026

Pricing has split into two models, and knowing which one you’re buying matters more than the headline number.

Subscription pricing gives you a fixed monthly fee and a message or conversation allowance. Predictable, easy to budget, and the right fit for most small businesses.

Outcome pricing charges per resolved conversation. Intercom, for example, publishes its Fin AI agent at $0.99 per outcome on top of seat pricing that starts at $29 per seat per month and runs to $132 per seat on its top tier. That model is honest — you pay when the bot works — but your bill moves with your traffic, and a seasonal spike moves it fast.

Here are current list prices from three platforms a small business would realistically shortlist. These are the vendors’ own published figures as of this writing; check them before you sign, because this category re-prices often.

Two patterns repeat across the whole market. First, the AI is almost always a separate line item from the chat widget. The $25 plan you saw advertised is usually live chat with a human — you. Second, removing vendor branding is a paid upgrade on most low tiers, and it can cost more than the plan itself.

Why raw AI costs are not the driver

Owners often assume the language model is the expensive part. It isn’t, and hasn’t been for a while. OpenAI currently lists its small models at fractions of a cent per thousand tokens — gpt-5-nano at $0.05 per million input tokens and $0.40 per million output tokens. A typical support answer is a few hundred tokens. You could run thousands of conversations a month and the model bill would land in single-digit dollars.

What you’re actually paying for is the wrapper: the widget, the retrieval system that keeps the bot anchored to your content, conversation logs, handoff to a human, CRM sync, and someone maintaining it. That’s the real cost, and it’s the part cheap tools skimp on.

The costs nobody quotes you

Budget for these or the project stalls at month two.

• Writing the source content — 5 to 15 hours. A chatbot is only as good as what you feed it. If your pricing, service area, turnaround times, and policies aren’t written down anywhere, you’re writing them now. This is the single biggest hidden cost and it lands on the owner.
• Escalation coverage. Every bot generates conversations it can’t finish. If those sit unanswered for two days, the chatbot has made your service worse than having no chat at all.
• Monthly review — 1 to 2 hours. Read the transcripts. Find the questions it fumbled. Update the source content. Skip this and quality decays quietly.
• Integration work. Booking a job, checking an order, or pushing a lead into your CRM is real development, not a toggle. This is where off-the-shelf ends and a build begins.

What an AI chatbot genuinely cannot do

Demos are staged. Here is what breaks in production.

It cannot invent facts you never gave it — but it will try. Ask about a policy that isn’t in the source material and a poorly configured bot will produce a plausible answer instead of saying “I don’t know.” This is not hypothetical liability. In Moffatt v. Air Canada (2024), the British Columbia Civil Resolution Tribunal found the airline liable for negligent misrepresentation after its website chatbot gave a customer incorrect information about bereavement fares, and rejected the argument that the chatbot was somehow a separate entity. Damages were small — around $812 — but the principle is the point: what your bot says, your business said.

It cannot close a complicated sale. A bot is good at qualifying and scheduling. It is not good at a $40,000 conversation with a nervous buyer.

It cannot resolve everything. A Gartner survey published in August 2024 found only 14% of customer service issues are fully resolved in self-service. Gartner separately predicts agentic AI will autonomously resolve 80% of common customer service issues by 2029 — but note the word common, and note that 2029 is not today.

It cannot fix a bad website. If people can’t find your prices or your phone number, a chat bubble is a bandage. Fix the site structure and page content first.

When a chatbot is not worth it

We turn this work down regularly. Skip the chatbot if any of these describe you.

Your traffic is under roughly 1,000 visitors a month

At that volume you might see a handful of chats a week. You can answer those yourself, faster and better, from your phone. Put the money into getting found in search instead — a chatbot amplifies traffic you already have, it doesn’t create it.

You can’t list 15 questions with settled answers

Try it right now on paper. If your answer to most questions is “it depends on the job,” a bot will either give a wrong number or deflect every time. Both are worse than a contact form.

Nobody is watching the inbox

A chatbot creates a channel customers expect to be live. If escalated chats sit for 48 hours, you have built a machine for disappointing people. This is the disqualifier we see most often in one- and two-person businesses.

Your inquiries are emotional, regulated, or high-stakes

Medical, legal, financial, funeral, emergency services. Route to a human immediately and use the bot for hours, directions, and intake only.

When it does pay back — run the math yourself

There are two credible returns. Don’t count both for the same conversation.

Return one: deflected questions. Count how many repetitive questions you answer per week — hours, pricing, service area, order status. Multiply by your hourly value. Twenty questions a week at five minutes each is about seven hours a month. Gartner’s benchmarks put the median cost per contact at $1.84 for self-service versus $13.50 for an assisted channel, which gives you a sense of the gap even if your own numbers differ.

Return two: captured leads, mainly through speed. This is the bigger one for most service businesses. The classic Harvard Business Review study “The Short Life of Online Sales Leads” (Oldroyd, McElheran and Elkington, March 2011) found firms contacting a prospect within an hour were nearly seven times as likely to qualify that lead as those that waited even an hour longer. A chatbot’s real advantage is that it responds in seconds at 9pm on a Sunday.

So the arithmetic is simple. Take your average job value, multiply by your close rate on inbound inquiries, and ask how many extra captured leads per month it takes to cover $75. For a service business with an $800 average job and a 20% close rate, one additional captured lead per quarter covers the subscription. For a business with a $60 average order, the answer is very different — and probably no.

Off-the-shelf versus custom

The honest rule: start off-the-shelf. Go custom only when you can point to a specific transaction the platform genuinely cannot perform — checking live inventory, quoting from your own pricing engine, writing into a system with no public integration. “It’ll feel more like us” is not that reason. Branding removal is a $99 add-on, not a $30,000 project.

A 30-day test before you commit

1. Week 0. Log every inbound question for one week. If you can’t fill a page, stop here — you’ve saved yourself the money.
2. Write the answers. One page. Plain language, real numbers, and an explicit list of what the bot must refuse to answer.
3. Build on a monthly plan. No annual contract in month one. Configure it to say “let me get a human” rather than guess.
4. Set the handoff. Escalations go to a phone someone actually checks, with a stated response window.
5. Read every transcript for 30 days. Count three things: conversations handled end to end, contact details captured, and answers that were wrong. That third number decides everything.

If after 30 days it captured leads you would have lost and got nothing badly wrong, keep it and go annual. If it mostly deflected people who then called you anyway, cancel it. That’s a real outcome, not a failure.

Where to go from here

We build chatbots for US small and mid-sized businesses, and we start every conversation with the qualifying questions above rather than a demo — because a bot deployed on thin traffic with no escalation plan reflects badly on both of us. If you want to think it through with someone who’ll tell you when the answer is no, our AI solutions and automation practice is the place to start. You can see how we approach conversational AI on the Genie’s AI page, and our pricing page shows what engagements typically look like before you get on a call. Common questions are answered on our FAQs, and if you’d rather just talk it through, get in touch.

Platform prices cited are the vendors’ published list prices at the time of writing and change frequently. Verify current pricing directly before purchasing.

Frequently asked questions

How much does an AI chatbot cost for a small business per month?

Most small businesses land between $30 and $150 per month on an off-the-shelf platform. Chatbase’s paid tiers start at $40/month for 700 message credits and $150 for 4,000. Tidio’s Starter is $24.17/month but meters its Lyro AI agent separately from around $32.50. Expect setup work on top, whether you do it or pay someone.

Is an AI chatbot worth it if I only get a few hundred website visitors a month?

Usually not. Below roughly 1,000 visitors a month you’ll see only a handful of chats a week, and you can answer those yourself faster and better. A chatbot amplifies traffic you already have; it doesn’t generate it. Spend that budget on search visibility or your website’s clarity first, then revisit.

What’s the difference between a custom chatbot and an off-the-shelf one?

Off-the-shelf platforms go live in days for a monthly fee and handle questions, qualifying and booking well. A custom build starts in the five figures and is only justified when the bot must act inside your own systems — checking live inventory, quoting from your pricing engine, writing to software with no integration. Wanting it to look more branded isn’t a reason to build custom.

Can my business be held responsible if the chatbot gives a customer wrong information?

Yes. In Moffatt v. Air Canada (2024), the British Columbia Civil Resolution Tribunal held the airline liable for negligent misrepresentation after its website chatbot gave incorrect fare policy information, rejecting the argument that the chatbot was a separate entity. Treat bot output like any other statement on your website: configure it to say it doesn’t know rather than guess.

How do I know whether the chatbot is actually working?

Track three numbers for 30 days: conversations it handled end to end, contact details it captured that you would otherwise have lost, and answers it got wrong. The third number matters most. If it deflected people who then called you anyway and made mistakes, cancel it — that’s a legitimate result, not a failure.