Do you need a chatbot or a better page?

Look at the last fifty questions. Most of them have an answer that was never on the site.

Chatbots are usually bought to answer questions the site could have answered by displaying the information.

The last fifty enquiries grouped, with twenty questions covering most of the volume.
The last fifty enquiries grouped, with twenty questions covering most of the volume.

Do the count first

Take the last fifty enquiries, by whatever route they arrive, and group them by what was actually being asked.

The distribution is almost always the same: a small number of questions covering most of the volume, and a long tail of genuinely individual ones.

If twenty questions cover forty of the fifty, you do not have a conversation problem. You have twenty answers that are not on the site.

Pattern in the fifty What it points at
Same twenty repeat A page
Depends on their account A sign-in, or a bot with access
Genuinely different each time A person, or a bot
Asked because the page is unclear Rewrite the page
Asked to check it is still true Show a date on the page

Why a page beats a bot for the repeated twenty

It is findable. A page is indexed, so somebody searching finds the answer without arriving at your site at all. A bot answers only people who have already arrived and chosen to ask.

It is linkable. Your team can send the exact answer. Nobody can link to a conversation.

It is auditable. Everyone reads the same words and you can see what those words are. A bot's answers vary, and nobody reviews them once it is running.

It does not invent. A page says what you wrote. A bot occasionally says something plausible and wrong, in your voice, to a customer.

Then measure again

Publish the page, link it prominently, and count the next fifty enquiries.

If the volume drops substantially, the problem was information and it is solved. What is left is the genuinely varied remainder, and now you know how large it is and what it consists of.

That is the point at which building something makes sense, because you know what it has to do.

A questions page with dated answers, linked from the places people arrive.
A questions page with dated answers, linked from the places people arrive.

If you build one

Point it at the page.

A bot answering from a well-written source is substantially more accurate than one answering from a model's general knowledge about your business. The page has to exist and be correct either way, so it is not extra work; it is the same work in the right order.

And budget for maintenance. Answers go stale, the tone needs occasional supervision, and a confidently wrong answer is worse than no answer at all.

Closely related: Prompting or a no-code builder, and How to set up a client portal for the adjacent problem.

Ask the assistant for a page, not prose HTML Check what came back render it before trusting it publish Give it an address so others can read it
Markup from a chat, checked in a renderer, then given an address.

Put it at an address

Count the last fifty questions, write the repeated twenty as a findable page, measure what still arrives, build only for the varied remainder, and point anything you build at that page.

Questions people ask

When is a chatbot the right answer?

When questions are genuinely varied, depend on the person asking, or need to reach a live account. Then a search box will not do.

When is it the wrong answer?

When the same twenty questions repeat. Those belong on a page, answered directly, where they can be found and linked.

How do I decide?

Take the last fifty enquiries and count the distinct questions. If twenty cover most of the volume, write the page first.

What do bots cost beyond the subscription?

Maintenance. Answers go out of date, the tone needs supervision, and a wrong answer given confidently is worse than no answer.

What if we build one anyway?

Point it at a well-written page as its source. The page has to exist and be correct either way, so write it first.

Keep reading