Indexed Search, KE-Search and comparable extensions have been providing reliable service for many years. But what visitors expect from a search today has changed fundamentally. Anyone who still merely matches keywords is sacrificing not only convenience, but also customer contacts.
When I analyse a website, sooner or later I take a look at the internal search. In most cases I find one of two tools there: Indexed Search, the standard solution from TYPO3, or KE-Search, the extension that has been popular for years. Both work reliably. Both do what they were originally designed to do. And both stem from a time when visitors still accepted having to figure out for themselves which terms they needed to search for in order for a website to deliver relevant results.
That time is over.
Why classic search reaches its limits
At their core, both of the tools mentioned are keyword searches. They break a website's content down into individual terms and match the visitor's input against this index. As long as the visitor uses exactly the word that also appears in the text, this works well. As soon as they phrase it differently, the process often breaks down.
An example: On the website of a tax firm there is an article on the “advance VAT return”. The visitor types: “When do I have to report sales tax to the tax office?” – and receives no result. From the search's point of view this is logical: the word “sales tax” does not appear in the text. From the visitor's point of view it is a dead end. They click back – if in doubt, to the competition.
On top of this comes a changed set of expectations. Anyone who uses a search today is accustomed to asking a complete question and receiving a direct answer in full sentences – including a source reference. Visitors now bring precisely this expectation to corporate websites as well. A plain list of hits, by contrast, feels like a step backwards.
What retrieval-based and context-based methods achieve
In recent years, two methods have become established that close precisely this gap: Retrieval-Augmented Generation and Cache-Augmented Generation – RAG and CAG for short.
In conversations I prefer to speak of “retrieval-based search” and “context-based search”, because the German terms better describe what actually happens.
The underlying idea is the same for both methods: a website's content is not only broken down into keywords, but captured semantically. When a visitor asks a question, the system does not search for identical words, but for passages within the data set that match in terms of content.
A language model formulates an answer from this in full sentences – supplemented by references to the relevant sources on the company's own website. As a result, the visitor to the tax firm receives an understandable answer about the advance VAT return along with a reference to the appropriate article. She stays on the website – and, if in doubt, submits an enquiry.
Three examples
A mid-sized mechanical engineering company had the problem that prospects were searching for use cases – not product names. The classic search delivered hardly any results. With a retrieval-based method, the website now answers questions such as: “Which machine is suitable for aluminium profiles?” – and leads prospects directly to the appropriate data sheet.
A law firm with an extensive article archive faced the question of whether its content was even still being used. With a context-based search, clients receive answers in full sentences – including a reference to the respective original article. Time on site and initial enquiries have risen significantly since its introduction.
A provider of software for the healthcare sector was struggling with a growing set of documentation that could hardly be searched meaningfully any longer via the classic search. Today, a context-based system answers both customer questions in the “Help and Instructions” area and internal enquiries from sales. The solution runs on its own infrastructure in Europe. Sensitive content never leaves the company at any point.
What it costs – and what it doesn't
The most common objection is: “That sounds interesting, but it's probably too expensive for us.”
This assumption often still stems from the early days of the large language models, when every API call to American providers incurred noticeable costs.
Today the situation is different. European models such as Mistral can be operated at comparatively low cost – in many cases even on your own infrastructure. The ongoing costs for a typical mid-sized company website often fall within an order of magnitude that is already amortised by a single additional customer enquiry per month.
In addition: the ongoing maintenance of classic search systems – synonyms, stop words, weightings and sorting rules – largely falls away. What remains is a one-off setup. And even this is usually more manageable than many assume.
My advice
Before you consider a new search, answer three questions:
- How often do your visitors use the search – and how often does it lead to no usable result?
- Which content on your website is particularly valuable, yet is hardly found today?
- Which enquiries would you receive more frequently if the search delivered genuine answers instead of mere lists of hits?
In most of the projects I support, the analysis of the search logs reveals a clear pattern: the classic search fails not just occasionally, but regularly. And it fails precisely where it becomes economically relevant – with prospects shortly before making an enquiry.
If you want to tackle the topic
I support small and mid-sized companies in modernising their TYPO3 websites – from analysing existing systems and developing concepts through to the technical implementation of retrieval-based and context-based search solutions based on European models.
For the switch from Indexed Search or KE-Search to a contemporary method, a field-proven solution is available that can be flexibly adapted to existing TYPO3 systems.
If you are considering how your website could deliver better answers to visitors – and thereby generate more enquiries – I would be glad to have a conversation.