AI integration for your website.
First the benefit, then the technology.
We integrate AI-supported search and digital assistants into your website and automate recurring tasks. What matters is where these functions help in everyday work and which content, data and interfaces are needed for them.
Together we check what fits your existing systems and processes. AI makes sense for some tasks; for others, automation according to fixed rules is sufficient.
The knowledge exists. Access is often the problem.
A project rarely begins with a concrete AI solution. Usually, very practical challenges are on the table first:
- The website search delivers numerous hits, but no helpful answer.
- The service team answers the same questions again and again, even though the information is already available online.
- Dates or other data are transferred manually because the systems involved do not communicate with each other.
At the same time, modern language models arouse interest in using these possibilities for one's own website or internal processes. For us, however, the decisive question is not first where AI can be used, but where time is unnecessarily lost today.
That is why we first look at the processes of your clients and employees. Only when the concrete hurdles and their causes are established do we talk about the right technical solution – with artificial intelligence or without.
AI is not the same as automation.
The two terms are often used together, but describe different approaches. Anyone who separates them cleanly can choose the right tool for each task.
Automation for fixed processes
If a process follows clear rules, it can be automated in the classic way: data is imported, checked, filed and then a notification is sent. Such processes are traceable, easy to test and deliver reproducible results under the same conditions.
A language model usually brings no advantage here. It would rather make the process more laborious, more expensive and less predictable.
AI for linguistic and content-related contexts
Artificial intelligence can make sense when language and context play an important role – for example with freely worded questions, extensive documents or content that describes the same topic with different terms. In such cases, fixed search terms and rules are often not enough.
We look at both approaches and check which of them suits the respective task.
Two foundations are decisive
- AI does not replace reliable content. If sources are incomplete, outdated or contradictory, the risk of unreliable answers also increases. A convincing wording is not yet proof of correct content.
- AI needs clear guard rails. Which sources may a digital assistant use? Which content is excluded? And how should it react if there is no reliable answer? These rules are set by people – not by the system.
How both approaches work together
An interface takes over event dates from your booking system and makes them available in TYPO3. This eliminates duplicate maintenance: classic automation.
On this structured data basis, an AI-supported search can then help to match even differently worded questions about the events with suitable information. Which content is approved for this and how uncertain results are dealt with remains under the control of your editorial team.
Automation or AI – what suits the task?
The overview offers initial orientation. If the assignment is not clear, we look at the concrete process together with you.
| Feature | Automation with fixed rules | AI and language models |
|---|---|---|
| Fits for | Always the same process | Language, context, interpretation |
| Inputs | Structured data, fixed formats | Free texts, documents, questions |
| Result | Reproducible with the same inputs and conditions | Can vary and be wrong |
| Testability | Rules and results can be tested specifically | Automated tests and professional assessment |
| Running costs | Dependent on scope, infrastructure and maintenance | Additionally dependent on model and usage |
| Typical example | Taking over event data from a third-party system | Answer to a freely asked question from approved pages |
| Poorly suited for | Tasks whose rules cannot be defined sufficiently | As the sole solution for calculations with a clear result |
What we implement for you.
Start small. Decide on a sound basis.
We start with a clearly defined use case. Using a first working version, we check together what benefit it brings and where its limits lie. On this basis, you decide whether and how the project is continued.
Your contacts accompany you through all phases and openly address it when effort and expected benefit are no longer in a sensible proportion.
Select a use case
Together we identify where time is lost today or access to information is made difficult. To start, we choose a concrete use case with recognisable benefit and manageable risk.
Check content and data
We examine whether the required sources are up to date, unambiguous and approved for the intended use. Often opportunities to improve content and processes already become apparent in this phase – even before new technology is used.
Implement on a small scale
We develop a first working version that you can test with real questions and processes. This creates a tangible result early on, on which benefits, limits and necessary improvements can be assessed.
Evaluate and decide
Even before the test, we jointly define how we recognise a good solution – for example by the quality of the answers, the time saved or the number of correctly processed cases.
We then discuss what has proven itself and which adjustments are still necessary. Together we decide whether the solution is developed further, revised first or not pursued further.
What improves in everyday work.
For your visitors
Users receive suitable information on their questions with traceable source references. As a result, they have to work through long lists of hits less often. If a personal conversation is necessary, it can already begin with a more concrete question.
For your organisation
Dates, product data or member information are imported from the respective leading systems and made available where they are needed. This reduces multiple maintenance, saves time and reduces the risk of transfer errors. How we implement such connections is described on the page Interfaces and extensions.
For your editorial team
Your responsible staff define which content is used as the basis. Corrections are made directly at the original source and are available to the assistant after the next update.
This keeps the responsibility for content with the people who know your topics. The AI becomes a tool within clear processes – not an additional system that stands independently alongside.
Our AI search in use
Ask our website a concrete question. The search uses our approved website content and an editorially maintained knowledge base as its basis and shows the sources used.
- How does a TYPO3 upgrade work?
- What does the BFSG mean for our website?
- When is an AI search worthwhile?
Nothing is formulated in the process: a language model only selects which of our text blocks match the question, and the answer consists literally of these texts. Below each answer it says whether it was formulated by us or selected by AI. If suitable information is missing or a question lies outside our topics, the search says so and refers to personal contact. A selection can also be off the mark; we point this out.
For recurring questions, for example about prices, dates or availability, fixed answers that we have written ourselves apply. They take precedence over the selection. Changes to website content are automatically transferred to the search index when saved.
Limits on costs and the number of requests help to keep operation under control. Questions without suitable hits are collected for editorial evaluation. This shows us which information is still missing on the website.
According to our configuration, the selection is carried out by a provider within the EU with a data processing agreement, and only if you have consented to the AI function in the cookie settings; without consent, the search finds pages via keywords. Details can be found in our privacy policy. What matters in implementation and how we measured the quality is described in the article AI search and chat on our website.
Interfaces and automation in use.
The following projects show how we make data from existing systems usable on websites and reduce duplicate maintenance. The focus is on interfaces and rule-based processes – not necessarily AI.
Frequently asked questions.
The running costs depend, among other things, on the model used, the scope of the content, the number and length of requests and the chosen infrastructure.
That is why we start with a clearly defined use case. The usage values determined in pilot operation then allow a more reliable forecast of the running costs. We define the effort for concept, development and integration transparently with you in advance.
We clarify and document this before implementation. This includes the model used and the providers involved, the place of processing and storage, the data transmitted and its intended retention period.
If there are special data protection or security requirements, we examine processing within Europe or on infrastructure controlled by you. Which solution is suitable depends on the use case and the technical framework.
Yes. Even an assistant limited to approved sources can assign content incorrectly or summarise it inaccurately.
To reduce this risk, we limit the knowledge base, show the sources used and define how the system should react to missing or uncertain information. In addition, we test typical questions and regularly check the results in ongoing operation. Incorrect answers cannot be completely ruled out.
In many cases, the new function can be integrated into an existing TYPO3 platform. The prerequisite, however, is that the required content is technically accessible and available in sufficient quality.
If information is only available in documents that are difficult to evaluate, in several places or in contradictory versions, structuring the content comes first. Classic search and editorial work also benefit from this.
Before the project, we jointly define which change is to be achieved and how we measure it. Possible criteria are the quality of the information found, the proportion of successfully answered questions, fewer recurring service requests or a shorter processing time.
We compare the results of the pilot operation with the previous situation. This creates a clear basis for decisions on further development.
Yes. You can try out our search and the chat via the magnifying glass in the header. The search uses our website and an editorially maintained knowledge base. An AI selects suitable texts but does not formulate new answers. Website sources used and further links are marked accordingly. You can find further insights in the article AI search and chat on our website.
Yes. This is about the findability of your website in external search and answer systems, not about an assistant on your own website. We check technical accessibility, content and structure and observe, using selected questions, whether and how your website appears as a source. We cannot guarantee a specific mention or recommendation. You can find out more under AI visibility and GEO.
Describe the concrete process to us.
One example is enough: a question that regularly lands with your team, or a list that is transferred manually every week.
Together we check what the cause is and which approach suits it – clearer structures, rule-based automation or the targeted use of artificial intelligence. If another path makes more sense, we tell you that just as openly.