Rule-based or AI: one of them can invent a price
This comparison usually gets written as old technology versus new technology. That framing is wrong. The real difference is what happens when a customer asks something nobody planned for — and which answer your business is legally on the hook for.
Same question, two chatbots
Ask something a plumbing company's website would get asked. Watch what each one does with it — especially the ones where the honest answer is "I don't know."
Moffatt v. Air Canada
A passenger asked Air Canada's website chatbot about bereavement fares after his grandmother died. The chatbot told him he could apply for the discounted rate retroactively, within ninety days of booking. He screenshotted it, bought full-fare tickets, and applied.
Air Canada refused. Its actual policy did not allow retroactive bereavement claims — the chatbot had made that up.
Air Canada's defense was that the chatbot was effectively a separate entity responsible for its own statements. The tribunal member called that submission remarkable, and found that a chatbot is simply part of the company's website — it makes no difference whether information comes from a static page or a chat window. The airline was found to have failed to take reasonable care to ensure its chatbot was accurate.
The practical lesson for a small business: anything your chatbot says is a representation your business is making. Eight hundred dollars is nothing to an airline. A fabricated quote, warranty term, or code-compliance claim on a contractor's site is not nothing to you. This was a Canadian tribunal rather than a court and rules vary by jurisdiction — but the reasoning is not exotic, and it is the risk model worth planning around.
Which one do you actually need?
Six questions. No email, no results-gate, no sales call. It will tell you to use the free tool if that is the honest answer.
What each one is genuinely good at
RB Rule-based
- Cannot state anything a human did not approve
- No per-conversation cost, at any volume
- Runs in the visitor's browser — no server to go down
- Live in about two minutes
- Predictable: the same question gets the same answer forever
- Easy to audit — you can read every possible reply
- Weak at anything phrased unusually or outside the script
- Practical ceiling of about six visible options
AI AI-powered
- Handles questions nobody thought to anticipate
- Understands messy phrasing, typos and long rambles
- Can be trained on your documents, pricing and policies
- Genuinely conversational over multiple turns
- Costs real money per conversation or per resolution
- Can state something false with total confidence
- Needs guardrails, testing, and someone to maintain them
- Days to set up, not minutes
The hybrid is usually the right answer
This gets framed as a choice, but the way a well-built custom bot actually works is both at once. High-risk questions — pricing, warranty, code compliance, insurance, anything with legal weight — get hard-scripted answers that the model is not permitted to improvise on. Everything else goes to the AI layer, which can handle the open-ended half of the conversation.
That structure gives you real conversation without betting your business on a language model's mood. It is also why "just use AI" and "AI is too risky" are both lazy answers.
How to choose in one question
Ask yourself: what are the twenty questions we actually get? If you can write them down and the answers are stable, rule-based covers it and you should not be paying monthly. If you genuinely cannot predict what people will ask — deep product catalogs, technical support, complex eligibility rules — you need the AI layer, and you need it governed.
Most small businesses who try the exercise find they can list all twenty in about ten minutes. That is the answer.
Eye To Ad Media
A chatbot only pays off if people are landing on your site in the first place. Eye To Ad has been doing SEO, AIO and GEO for small businesses out of Denver since 2012 — the search side of the same operation that built this tool.
Common questions
What's the actual difference?
A rule-based bot matches what a visitor says against answers a human wrote in advance, and admits it when there is no match. An AI bot generates a fresh reply each time using a language model — so it can answer things nobody anticipated, and it can also state things that are not true.
Can an AI chatbot really make things up about my business?
Yes. In Moffatt v. Air Canada, a British Columbia tribunal found the airline liable for negligent misrepresentation after its chatbot gave a customer incorrect bereavement fare information, awarding $812.02. The tribunal rejected the argument that the chatbot was a separate legal entity.
Am I liable for what my chatbot says?
That decision treated the chatbot as part of the company's website, drawing no distinction between a static page and a chat window. It was a Canadian tribunal rather than a court and rules vary by jurisdiction, so treat anything your bot says as a representation your business is making, and get real legal advice if the stakes are high.
Which is better for a small business?
Usually rule-based. The questions filling your phone line are repetitive with fixed answers, and a scripted bot delivers them at no monthly cost without improvising a price or a warranty. AI wins when customers ask open-ended things you cannot all anticipate.
Can you combine both?
Yes, and it is often correct. Scripted replies handle the high-risk questions — pricing, warranty, policy — while a guardrailed AI layer handles open-ended conversation. That is how we structure custom builds.
Are rule-based chatbots outdated?
No. Being older technology is not a drawback when the job is delivering a fixed answer accurately. No per-conversation cost, no server, and no way to say something a human did not approve.
How do I add guardrails to an AI bot?
Hard-script the answers to anything with legal or financial weight, restrict the model to your own approved content, test it adversarially before launch, and log conversations so you find problems before a customer does. That is work, which is why it is a custom build and not a free tool.
Start with the one that can't lie to your customers
If the free rule-based bot covers your twenty questions, you are done and it cost you nothing. If it does not, you will know exactly what the AI layer needs to handle — and we build those, with the guardrails on.