Website search that answers questions – explained simply
A search returns links. An answer box also adds a short sentence that answers the question straight away – with a source. What lies behind it, without any jargon.
The box answers the question in a few sentences and names the pages the answer comes from.
The usual results remain visible below.
If there is no suitable source, the box says so openly.
The problem: searching is often tedious
Visitors looking for something on a website usually type a few words into the search field. The result is a list of links. Then the real work begins: opening a page, reading, going back, trying the next page. Often the information sought is buried somewhere in the middle of a long text.
Many visitors give up quickly. Some then send an email to the support team, others leave the website altogether. Both are a shame, because the answer was there all along – it just could not be found quickly enough.
The idea: the answer first, then the links
An answer box reverses the order. First comes a short text that answers the question directly. Below it are small cards showing the sources. A click on a card leads straight to the section containing the information. The normal search results stay below, as usual.
What makes a good answer box
- Short: a few sentences instead of a long text.
- Backed up: every answer names the pages where the information can be found.
- Honest: without a suitable source there is no guessed answer, just a clear notice.
- Labelled: a visible notice shows that the text comes from an AI.
- Unobtrusive: the existing search stays in place and keeps working.
Who benefits?
An answer box is particularly helpful on websites with a lot of information: company websites with many products, online manuals, knowledge portals or shops that get many questions before a purchase. The more text a website has, the greater the gain when the right passage is found immediately.
An answer box makes less sense for confidential content. The answers are public. So only texts that anyone is allowed to read on the website anyway belong in it.
And the technology?
In the background, a program reads the website’s public pages and notes what each section is about. When a question comes in, it picks out the relevant sections. A language model – an AI program for text – then turns them into a short answer. The last step is the important one: the answer may only be based on these sections. There is more on this in the post about RAG on this blog.