Every chatbot claims lead generation. Most produce transcripts.
The difference is not the model or the interface. It is whether the thing was designed around a simple sequence: help first, qualify inside the help, ask once, record everything.
The design that kills conversion
The most common failure is asking for an email before answering anything. It appears in template after template because it maximises captured addresses in a demo.
In the real world it does the opposite. The visitor arrived with a question. Being asked to pay for the answer with their contact details, before knowing whether the answer is any good, reads as a toll booth. Most leave.
What works instead
Answer the question properly first. Establish that the thing is useful.
Qualify inside the answer, not as a form. "That depends on the property, how many bedrooms is it?" is a qualifying question that also genuinely helps. "Please complete the following four fields" is not.
Ask once, at the right moment. The natural moment is when the conversation has produced something they want: a quote, a callback, availability, or an answer you cannot give without more detail.
Take no for an answer. If they decline, keep helping. A visitor who was helped and not harassed comes back.
Record the whole thing. A lead with only a name and email is barely better than an anonymous form fill. A lead with the question, the qualifying answers and the context lets someone follow up warm.
How to judge one in ten minutes
Ignore feature lists. Four things decide whether it works for you:
- Where does it get its answers? It should answer from your own website and documents, and say so plainly when it has nothing. Anything that improvises about your prices will eventually invent one.
- Can it do anything, or only talk? Capturing a lead, opening a ticket and fetching a human are the difference between a chatbot and an employee.
- How does it price? Flat monthly is predictable. Per resolution or per conversation means your bill rises exactly when you have a good month.
- What happens at the limit? The honest failure — "I do not know, let me get someone" — is worth more than a confident wrong answer.
The metric to actually watch
Not conversations. Not captured emails. Qualified enquiries per 100 conversations, and then what share of those became customers.
That number tells you whether the thing is producing work or producing noise, and it is the only one worth reporting.
The only test that matters
Every product in this category demos well on its own website. Point one at your site and ask it the three questions your customers actually ask. You will know within a minute.
Build one from your URL free — no signup, and it deletes itself afterwards.
Related: what missed enquiries cost.