Ask most business owners what they want from a website chatbot and the answer is "answer questions". That is a fair place to start, but it undersells the job. The assistants that make a real difference to a business do several quieter things as well, and they do them in every conversation.
Here is what we have learned from watching a very large number of real conversations, and what we think a good assistant should be doing all day.
1. It answers the question that was actually asked
This sounds obvious, yet it is where most chatbots fail. A visitor asks, "Can I return a sale item if it was a gift?" A weak assistant replies with your entire returns policy. A good one says, "Yes. Gifts bought in the sale can be returned within 30 days for store credit, as long as you have the gift receipt."
The difference is not the AI model. It is the content behind it. When your knowledge base has clear, specific answers, the assistant can give clear, specific replies. When it only has long, general pages, you get long, general answers.
2. It notices intent, not just words
"Do you work with companies in Germany?" is not really a geography question. It is someone checking whether they can become a customer. A good assistant answers the question and then moves the conversation forward: "Yes, we work with teams across Europe. Would it help if I showed you which plan fits a team your size?"
This is where lead scoring earns its keep. When the assistant recognises buying signals, such as questions about pricing, contracts, team sizes or start dates, it can mark the conversation as warm or hot, so your sales team knows where to look first on Monday morning.
3. It collects context without interrogating people
Nobody wants to fill in a form inside a chat window. But most visitors are happy to share their email address once they have been helped, especially if there is a good reason: "I can send you that price list. What is the best email to use?"
The rule of thumb we give customers is simple. Help first, then ask. Assistants that ask for an email before answering anything see far fewer conversations, and the leads they do capture are of lower quality.
4. It knows what it does not know
The most damaging thing an assistant can do is make something up. An invented discount, a delivery date that does not exist, a feature you do not have. Each one costs you trust, and sometimes money.
That is why we built strict mode. When it is on, the assistant only answers from your own content. If it cannot find the answer, it says so plainly and offers to bring in a person. In our experience visitors are surprisingly forgiving of "I am not sure, but I can get someone who knows". They are not forgiving of being misled.
5. It steps aside at the right moment
Some conversations should not be handled by software: a frustrated customer, a complicated quote, anything involving a complaint or a refund dispute. A good assistant recognises these and hands over gracefully, with the full conversation and a short summary, so the person who picks it up does not have to start from scratch.
We cover this in more detail in when to hand a chat to a human.
6. It teaches you about your customers
This is the benefit people rarely expect. Every conversation is a small piece of market research, in your customers' own words. After a few weeks you will know which questions come up most, which pages confuse people and which objections stop them from buying.
Imagine an online furniture shop that notices a large share of its chats are about delivery to upper floors, something its website never mentions. One added sentence on the product pages and most of those questions go away. Small fixes like this are hiding in almost every set of conversations.
What this means for you
If you are setting up an assistant, or reviewing one you already have, ask yourself these questions:
- Are the answers specific, or do they just repeat whole pages?
- Does the assistant move interested visitors towards a next step?
- Does it help before it asks for contact details?
- What happens when it does not know the answer?
- Can your team take over easily, with full context?
- Is anyone actually reading the conversations each week?
Get these right and the AI model behind it matters much less than you might think. If you would like a second opinion on your setup, talk to us. We are always happy to look.