--- title: "Why AI search needs more than just a chatbot | Search is the Interface" url: "https://www.dkd.de/en/blog/search-is-the-interface-why-ai-search-needs-more-than-just-a-chatbot/" description: AI-powered website search is changing the way users find information. Instead of traditional lists of results, they increasingly expect direct, relevant and clear answers. image: "https://www.dkd.de/_assets/processed/6GWy3UoD7125uTHRH8TtS_4IqLPbUtBGVVdvTf46fy8/q:85/w:1365/h:768/fn:Y3NtX2RrZC10eXBvMy1hZ2VudHVyLXBvZGNhc3Qta3VlbnN0bGljaGUtaW50ZWxsaWdlbnpfZWVmMTA3ODRiZQ:t/cb:e6c91e2513fb552b29ddb08fcbcf82084d081f16/bG9jYWw6L2ZpbGVhZG1pbi9CaWxkZXIvQmxvZy9Db250ZW50LzIwMjYvUG9kY2FzdF9BSV9TZWFyY2hfVFlQTzMvZGtkLXR5cG8zLWFnZW50dXItcG9kY2FzdC1rdWVuc3RsaWNoZS1pbnRlbGxpZ2Vuei5wbmc" date: 2026-10-30 modified: 2026-11-02 --- # Why AI search needs more than just a chatbot | Search is the Interface Olivier Dobberkau 29.10.2026 Search is the Interface: Why AI search needs more than just a chatbot ======================================================================= AI-powered website search is transforming the way users find information. Instead of traditional lists of results, they increasingly expect direct, relevant and transparent answers. Technologies such as Retrieval Augmented Generation (RAG), semantic search and Apache Solr are opening up new possibilities for making existing search functions more intelligent. But why isn’t a chatbot alone enough? In the DotFusion podcast, dkd founder Olivier Dobberkau explains the role played by high-quality content, TYPO3 and modern search technologies, and how companies can develop their website search to make it fit for the future. ![](https://www.dkd.de/_assets/processed/MWDuGXzqSwAnNlw3FayY6ByW7fByVTAviSrQD-Bw49w/q:85/w:592/h:333/fn:Y3NtX2RrZC10eXBvMy1hZ2VuY3ktcG9kY2FzdC1hcnRpZmljaWFsLWludGVsbGlnZW5jZV9mMjlkM2Q5YTg4:t/cb:42cf3068d28c186be6155d636b0441b3a81b9875/bG9jYWw6L2ZpbGVhZG1pbi9CaWxkZXIvQmxvZy9Db250ZW50LzIwMjYvUG9kY2FzdF9BSV9TZWFyY2hfVFlQTzMvZGtkLXR5cG8zLWFnZW5jeS1wb2RjYXN0LWFydGlmaWNpYWwtaW50ZWxsaWdlbmNlLnBuZw "Beyond the CMS") --- [Read the blog post](#c14884) [Straight to the podcast](https://www.youtube.com/watch?v=8UvvfA8p0EA&t=126s "Listen to the ‘Beyond the CMS’ podcast") ### In this blogpost - [From traditional website search to intelligent AI search](#c14890 "From traditional website search to intelligent AI search") - [Why good content is the foundation of any AI search](#c14891 "Von der klassischen Website-Suche zur intelligenten KI-Suche") - [RAG: How AI-powered website search works](#c14892 "RAG: How AI-powered website search works") - [And how can such mistakes be avoided?](#c14893 "And how can such mistakes be avoided?") - [Targeted responses instead of traditional chatbots](#c14894 "Targeted responses instead of traditional chatbots") - [The future of the website: answers rather than mere navigation](#c14895 "The future of the website: answers rather than mere navigation") - [TYPO3 and Apache Solr as the basis for intelligent search solutions](#c14896 "TYPO3 and Apache Solr as the basis for intelligent search solutions") - [How businesses can make their website search AI-ready](#c14897 "How businesses can make their website search AI-ready") - [The future of AI search starts with content](#c14898 "Die Zukunft der KI-Suche beginnt beim Content") How is artificial intelligence changing website search? What role do content quality, RAG and search technologies such as Apache Solr play? And why isn’t a chatbot automatically the best solution? [Olivier Dobberkau, founder of the digital agency dkd and President of the TYPO3 Association](https://www.linkedin.com/in/olivierdobberkau/ "Go to Olivier Dobberkau’s LinkedIn profile"), discusses these topics in the 50th episode of the DotFusion podcast, hosted by [Chris Bryce](https://www.linkedin.com/in/cbryce/ "Go to Chris Bryce’s LinkedIn profile"). The focus is on how traditional website search is evolving into an intelligent, AI-powered search. *Prefer to read rather than listen? No problem!* Of course, we realise that not everyone fancies listening to a podcast. Perhaps you’re in the office right now, have forgotten your headphones, or simply belong to the ‘I’d rather read – I’ll finish quicker’ camp. That’s why we’ve summarised the key takeaways from the conversation for you in this blog post. No play button required, but plenty of insights. From traditional website search to intelligent AI search -------------------------------------------------------- Search is one of the most important functions of a website. Users enter a term, press Enter and expect relevant results. So far, so familiar. However, various applications such as ChatGPT have changed our expectations of search. Instead of entering individual keywords, we are increasingly asking complete questions and expect answers that are as specific as possible. For example, someone who wants to know how to improve the [search function on a TYPO3 website](https://www.dkd.de/en/products/solr-enterprise-search/#c10355 "Enterprise Search: Apache Solr for TYPO3") doesn’t necessarily want to trawl through ten different subpages first; they want to find out as directly as possible what options are available. This is precisely where CEO Olivier Dobberkau sees a fundamental shift. Traditional website search is increasingly evolving into an intelligent interface between people and information. Whilst conventional search systems primarily return relevant documents and content, AI-powered search solutions can recognise the intent behind a query, link pieces of information together and formulate comprehensible answers based on this. The aim is not to completely replace existing search technologies, but to develop them further so that they can keep pace with users’ changing expectations. Added to this is another development that Olivier Dobberkau discusses in the podcast: in future, it will no longer be exclusively humans searching for information on websites. AI agents, too, will increasingly access digital content automatically. Websites must therefore present their information in such a way that it is understandable and accessible to both human visitors and machine systems. Why good content is the foundation of any AI search --------------------------------------------------- A key message from the discussion can be summed up with a well-known phrase: **‘Junk in, junk out.’** If you feed poor-quality content into an AI search engine, you shouldn’t be surprised to get poor results. Whilst this may sound obvious at first, it is often underestimated in practice. Many companies are already working intensively on new AI technologies, powerful language models and intelligent chatbots, whilst their existing website content often still has a whole host of other issues. Yet the quality of the available information is a key factor in determining how well an AI-powered search works. If texts are worded in an incomprehensible way, important information is missing or content is poorly structured, even a powerful language model cannot reliably compensate for these shortcomings. In the worst-case scenario, this results in answers that may sound very convincing but are actually incorrect. Olivier therefore recommends that companies first take a critical look at their existing content before making major investments in new search functions. This also involves examining the language used by their own target audiences. This is because companies often use internal technical jargon on their websites, whilst their visitors use completely different terms or phrase their questions differently. The wider this linguistic gap, the more difficult it becomes to find the right information. Good content should therefore not only be factually accurate and comprehensive, but above all should clearly explain what users actually want to know. Meaningful metadata and a clear content structure also help to make information more accessible to search engines and AI systems. RAG: How AI-powered website search works ---------------------------------------- Retrieval Augmented Generation, or RAG or short, plays a key role in the further development of website search. This term refers to an approach that combines traditional search technologies with the capabilities of large language models. In the podcast, Olivier explains how this can transform a conventional search query into a significantly more intelligent interaction with a website’s content. In a traditional search, documents and website content are first indexed, for example [using Apache Solr](https://www.typo3-solr.com/ "Alles Wissenswerte über Apache Solr für TYPO3 auf www.typo3-solr.com"). When a user enters a search term, the system searches the index and returns a list of relevant results. This principle has worked reliably for many years, but reaches its limits when people use phrasing that differs from that found in the existing texts. RAG extends this approach through, amongst other things, so-called [embeddings](https://www.dkd.de/en/blog/recap-seo-day-2025/#c10908 "Embeddings are replacing traditional keyword-based thinking"). This involves translating text content into numerical representations, so that search systems do not have to search exclusively for identical terms, but can also recognise contextual relationships. Here is a simple example: A website describes dogs and cats without explicitly using the term ‘pets’. A semantic search can nevertheless recognise that this content is relevant to the question of which animals are suitable as pets. If the information found is then fed into a language model, the model can formulate a comprehensible answer rather than simply producing a list of search results. However, RAG also presents challenges. Extensive documents are often divided into smaller sections of text, known as [‘chunks’](https://www.dkd.de/en/blog/recap-seo-day-2025/#c10902 "Content must be ‘chunkable’ and AI-readable"), so that they can be processed efficiently. However, this carries the risk that important contextual links may be lost. Furthermore, language models do not always interpret information correctly and may, in some cases, generate responses that are not at all supported by the available sources. In this context, Olivier refers to the well-known case of an Air Canada chatbot that provided incorrect information regarding company policies. And how can such mistakes be avoided? ------------------------------------- To avoid such errors as far as possible, a process known as ‘grounding’ plays a crucial role. This involves linking AI-generated responses to specific, reliable sources of information. For businesses, this means that an AI search should not simply produce texts that are as convincing as possible, but must provide verifiable answers based on actual existing content. Ideally, these answers are supplemented by source references and further links so that users can verify the information themselves. As soon as companies start thinking about [integrating artificial intelligence into their website](https://www.dkd.de/en/services/artificial-intelligence-ai/#c13164 "AI Transformation for Your Business"), one idea often springs to mind: We need a chatbot. Yet it is precisely this reflex that Olivier questions in the podcast. After all, just because a website has a chat window, that doesn’t necessarily mean users will be able to access the information they want any quicker. Many people have already had experiences with chatbots that pop up at the most inopportune moments, provide standardised responses or reach their limits when faced with slightly more complex questions. Consequently, there is a great deal of scepticism towards such applications. Olivier therefore argues that we should focus less on the user interface and more on the actual search task. Someone looking for a contact person needs different information to someone who wants to find a specific price, a technical data sheet or a service. An intelligent search system should be able to recognise these different intentions and respond to them appropriately. Targeted responses instead of traditional chatbots -------------------------------------------------- Technically, this can be achieved, for example, through specialised search functions that are linked via a so-called router. This recognises the intent behind a query and forwards it to the appropriate search component. Rather than overloading a single chatbot with as many tasks as possible, this enables companies to address different search requirements in a targeted manner and expand the system step by step. Another interesting alternative is AI-generated response texts, which are displayed directly within the [traditional website search function](https://www.dkd.de/en/products/solr-enterprise-search/#c10355 "Your Enterprise Search for the TYPO3 website"). Much like with Google, users are given a brief summary of the information they are looking for and can then delve deeper into the topic via the linked sources. The advantage of this approach is that it follows a familiar pattern of use. Visitors do not first have to get used to a completely new user interface, but can continue to use the search function in the way they are accustomed to. The future of the website: answers rather than mere navigation -------------------------------------------------------------- Olivier illustrates what such a development might look like in practice using a client project from Austria as an example. In that project, eight key subject areas were identified and around 350 typical questions were compiled that visitors might ask on the website. These included, for example, questions about official permits, areas of responsibility or application procedures. Rather than leaving users to their own devices with a long list of potentially relevant subpages, [an intelligent search function](https://www.dkd.de/en/blog/apache-solr-for-typo3-the-path-to-intelligent-search/#c11317 "Apache Solr for TYPO3 | The path to intelligent search") could, in future, directly explain which steps are required, where the relevant information can be found, or which organisation is responsible for a particular matter. This also changes the role of the website. It is no longer merely a collection of information that visitors have to navigate through themselves, but is increasingly evolving into an interactive point of contact. In the podcast, Olivier describes this concept as a sort of ‘mini-GPT’ specialising in the content and tasks of a particular organisation. Users ask their questions in natural language and receive answers that are as specific and clear as possible. This approach is particularly interesting for [large corporate websites](https://www.dkd.de/en/services/enterprise-websites/#c9132 "dkd develops and builds enterprise websites"), public institutions or organisations with complex information structures. In such cases, visitors often need to understand how a website is structured before they can find the content they are looking for. An intelligent search function could significantly simplify this intermediate step. However, this remains contingent on the required information being available and accessible in a reliable manner. TYPO3 and Apache Solr as the basis for intelligent search solutions ------------------------------------------------------------------- [At dkd, we have been working on professional search solutions](https://www.typo3-solr.com/ "Everything you need to know about Apache Solr for TYPO3 websites") and the integration of Apache Solr into TYPO3 for many years. Apache Solr is a powerful open-source search platform that is used in particular for large-scale websites and complex search requirements. In the podcast, Olivier discusses the long-standing development of the TYPO3-Solr integration, collaboration within the open-source community and the ongoing development of the technology. Particularly in the [context of AI](https://www.dkd.de/en/services/artificial-intelligence-ai/#c13164 "Your path to AI transformation with dkd"),it becomes clear just how important a reliable search infrastructure continues to be. Before a language model can formulate a good answer, the relevant information must first be found and made available. The quality of this so-called retrieval layer has a significant influence on whether an AI system can generate relevant and factually correct answers. Powerful search technologies, carefully curated content and a well-structured content architecture therefore remain a vital foundation for the further development of website search, even in the age of artificial intelligence. The integration of AI does not, therefore, necessarily mean that companies have to replace their existing search systems. It often makes more sense to expand existing technologies in a targeted manner and to deploy new functions where they offer genuine added value. This is precisely the step-by-step approach that Olivier follows in his recommendations for practical implementation. How businesses can make their website search AI-ready ----------------------------------------------------- According to Olivier, anyone wishing to improve their website’s search function should first find out what questions users are actually asking. To do this, it is worth taking a close look at \* existing search queries \* frequently searched terms \* searches with no results. Such analyses not only reveal technical weaknesses but also provide valuable insights into what information is missing or where the website’s language does not match visitors’ expectations. On this basis, companies should review their content and metadata and make targeted improvements. After all, an intelligent search function does not simply require as many documents as possible, but above all relevant, comprehensible and well-structured information. At the same time, Olivier recommends focusing initially on a reliable retrieval layer and only introducing additional AI functions afterwards. Rather than attempting to develop a comprehensive, end-to-end solution as quickly as possible, individual functions can be added, tested and optimised step by step. It is also important to continuously monitor the quality of search results. This involves systematically analysing search queries and gathering user feedback. Even simple rating options such as ‘thumbs up’ or ‘thumbs down’ can provide valuable insights into whether an answer was helpful or needs improving. A good website search is therefore not a one-off project, but a system that should evolve in line with the questions and needs of its users. The future of AI search starts with content ------------------------------------------- Artificial intelligence is transforming the way people search for information and interact with websites. Traditional search is increasingly evolving into an intelligent interface that not only finds relevant content but can also understand and answer more complex questions. Technologies such as RAG are opening up new possibilities for enhancing existing search systems and presenting information in a more user-friendly way. However promising this development may be, even the most intelligent AI search cannot function reliably without high-quality, comprehensible and well-structured content. Companies should therefore not start by asking themselves which chatbot they want to use on their website, but rather how they can better meet their users’ actual information needs. After all, what ultimately matters is not how much AI a website contains, but how well it helps users find the right answers. --- Whether it’s TYPO3, Apache Solr or AI-powered search functions: At dkd, we focus on how businesses can make their content more accessible and further develop digital search experiences. We help you optimise existing search solutions and make effective use of new possibilities. [Would you like to make your website search smarter?](https://www.dkd.de/en/contact/#c8746 "Would you like to make your website search smarter?") --- ### **Have a listen now!** Dot-Fusion Podcast, Episode 50 **Search is the Interface** – with **Olivier Dobberkau & Chris Bryce** Insights into the future of search, RAG, TYPO3 and open source [LISTEN TO THE PODCAST](https://www.youtube.com/watch?v=8UvvfA8p0EA&t=126s "Listen to the podcast") --- ### Sources & further links \[1\] [AI transformation? We are your AI agency](https://www.dkd.de/en/services/artificial-intelligence-ai/#c13164 "KI-Transformation? Wir sind Ihre KI-Agentur") \[2\] [The Path of dkd’s AI Transformation](https://www.dkd.de/en/blog/weg-der-dkd-ki-transformation/#c12851 "The Path of dkd’s AI Transformation") \[3\] Prompt, please! \[4\] [Starfruit AI](https://www.dkd.de/en/products/starfruit-ai/#c9213 "Starfruit AI | The secure AI interface for your business") \[5\] [Securing AI: The Self-Sovereign, Privacy-Compliant LiteLLM Platform](https://www.dkd.de/en/blog/securing-ai-the-self-sovereign-privacy-compliant-litellm-platform/#c13220 "Read the blog post: Securing AI: the secure, data-protection-compliant LiteLLM platform") \[6\] [Apache Solr for TYPO3 – The Path to Intelligent Search](https://www.dkd.de/en/blog/apache-solr-for-typo3-the-path-to-intelligent-search/#c11317 "Read the blog post: Apache Solr for TYPO3 – The path to intelligent search") ### Write comment ### Comments No Comments #### Tags - [\#apachesolr](https://www.dkd.de/de/blog/tag/apache-solr/) - [\#KI](https://www.dkd.de/de/blog/tag/ki/)