Imagine a potential customer opens a beauty salon website and asks the website assistant:
“Do you have any appointments available for brow lamination on Friday afternoon?”
The assistant replies:
“After brow lamination, it’s best to avoid getting your brows wet.”
Erm… thanks. 😄
That might be useful information.
It just wasn’t the question.
And this is where the difference starts between an AI assistant that can generate answers and one that has actually been prepared for the job it is meant to do.
If we do not define what it is allowed to answer, what information it should use, and what it should do when it is unsure, it may simply try to say something anyway.
Sometimes the right response is not to guess.
The goal is not to have an answer for everything
It is tempting to think that the more customer questions a website assistant can answer, the better.
But in a real business, not every question is the same.
Some questions have a clear, approved answer.
Some need more information first.
And some are simply not the assistant’s job to answer.
A well-designed system needs to understand that difference.
Sometimes the best response is simply:
“I don’t have enough information to give you a reliable answer.”
That may sound less impressive than a long, confident reply.
But it is far more useful.
It does not need to know everything
A website assistant does not need to know everything about everything.
It needs to do its own job well.
That means deciding in advance what information it can use, which questions it can answer independently, when it should ask for more detail, and when it should not make assumptions.
It is very similar to training a new employee.
You would not tell a new colleague:
“If you don’t know, just make something up.”
You would show them what they can deal with themselves, where to find the right information, and when to ask for help.
An AI assistant needs the same kind of structure.
Sometimes it needs another question, not an answer
There are also times when the assistant does not know what to do simply because it has not been given enough information yet.
If someone says:
“I’d like to book an appointment.”
the best next step is not necessarily to start listing random times.
It may need to ask:
“Which service would you like to book?”
That sounds like a small difference, but it matters.
A good website assistant does not just produce answers.
It helps move the conversation forward.
And what if it genuinely does not know?
A customer may ask about a special request, an unusual service or a decision that is not covered by the assistant’s approved information.
In that case, there are several sensible next steps.
The assistant can say that it cannot give a reliable answer.
It can ask for clarification.
It can offer a contact option or a callback.
Or it can reach the point where a member of the team needs to take over.
The important thing is that this should not happen by accident.
It should be designed into the system just as carefully as the questions it is expected to answer.
Your website assistant is speaking on behalf of your business
This is one of the reasons guessing can become a real problem.
To the customer, the assistant is part of your business.
It appears on your website.
It gives information about your services.
It communicates in your company’s name.
So when it says something confidently, the customer may reasonably understand that as:
“This is what the business told me.”
That is why it is not enough to decide what the assistant should know.
We also need to decide what it should do when it does not know something.
How we think about this at NLS
When we design a website assistant, our aim is not to make it answer every imaginable question.
We first look at the role it is meant to have in the business.
What questions come up regularly?
What information can it provide with confidence?
Where does it need more detail?
In which situations would it be inappropriate for the assistant to make its own assumptions?
And what should happen when it reaches the edge of its role?
Because a well-designed website assistant does not need to know everything.
It does not need to answer everything.
And it certainly does not need to pretend that it knows the answer when it does not.
Sometimes the most helpful thing it can recognise is:
“This is as far as I can help. From here, someone else needs to take over.”
And that leads to the next important question:
When should an AI assistant hand a conversation over to a person?