A chatbot is software that holds a text conversation with people, usually in a chat window on a website or in a messaging app such as WhatsApp or Messenger. It answers questions, collects details and passes conversations to a person when needed, without anyone having to be at a keyboard.
How a chatbot works
There are two broad types, and many products mix them:
- Rule-based chatbots follow a decision tree. The visitor picks from buttons or types something the bot recognises, and it moves to the next scripted step. They are predictable and cannot say anything you did not write, but they cope badly with questions outside the script.
- AI chatbots use a large language model to understand free text and write replies. Good ones are grounded in your own content, such as your FAQs, service pages and policies, through a method called retrieval-augmented generation, so they answer from your information rather than from general knowledge.
Behind either type sit the connections that make it useful: a handover to live chat with a person, a link to a booking system, and a route for passing collected details into your CRM or inbox.
Why it matters
A chatbot can answer routine questions instantly at any hour, catch enquiries that would otherwise be lost in the evening and at weekends, and qualify leads before a person picks them up. In messaging apps, chat flows are also how click-to-message ads turn a tap into a conversation.
The risk is what it says. Under UK consumer protection law, which since April 2025 sits in the DMCC Act 2024, your business is responsible for what your chatbot tells customers, just as it would be for a member of staff. If it invents a refund policy, quotes a price you do not offer or promises a delivery date you cannot meet, that is your business misleading a consumer. A Canadian tribunal ruled against an airline in 2024 after its chatbot gave a passenger incorrect refund information, rejecting the argument that the bot was responsible for its own words. The principle carries over.
AI chatbots can also hallucinate, producing confident, plausible answers that are simply wrong. And chat transcripts often contain names, phone numbers and sometimes health or financial details, so your privacy notice must explain what is collected through chat, why, who the provider is and how long transcripts are kept.
Common mistakes
- Letting an AI chatbot answer from general knowledge instead of limiting it to your own approved content.
- No easy way to reach a person, which turns a quick question into a complaint.
- A pop-up that opens over the page on every visit, blocking content on mobile.
- Giving the bot a human name and photo so visitors think they are talking to staff.
- Forgetting to update the bot when prices, opening hours or policies change.
- Leaving chat data out of the privacy notice.
How to act on it
Start by listing the questions your inbox and phone get most often. If most are simple and predictable, a rule-based bot or a well-organised FAQ may be enough. If questions vary widely, an AI chatbot grounded in your own content makes sense, with instructions not to answer outside it and to hand over to a person when unsure.
Test it hard before launch: ask about refunds, prices, edge cases and things you do not offer. Review transcripts weekly for the first month. If you plan to start chats from ads, the flow design matters as much as the ad itself, and I build that into click-to-WhatsApp and Messenger ads.