A German industry paper argues that knowledge graphs can make the information used by AI easier to trace. The wider issue is whether companies can see – and maintain – the knowledge behind their systems. 

A customer asks an AI assistant about a product and receives an answer based on information that is months out of date. Nothing has failed visibly. The problem sits behind the answer: an old document, a disconnected record, or data nobody is responsible for updating. 

A position paper published by Germany's KI Bundesverband in June 2026 makes the case for knowledge graphs as one response. Its authors argue that they can connect information held across business systems and make the sources behind an AI answer easier to identify. 

This is an industry position paper, not independent research. Its eight authors include representatives of AI and technology companies including TextVerstehen, niologic, deepset, ArtiQuare, SupraTix and ATVANTAGE. It also includes company examples from TextVerstehen and niologic. 

What a knowledge graph does 

A knowledge graph maps things – products, people, suppliers or events – and the relationships between them. Instead of recording a supplier as a single entry, it can link that supplier to a product, a site and a certification. 

The paper says this can create a shared layer across systems such as ERP, CRM and document repositories. Rather than working from several disconnected sources, an AI application can draw on information that has been linked and defined in one place. 

That can be useful when an answer needs to be checked. The paper argues that a graph can make it possible to trace an AI output to named sources, including information on when a source was current and which department was responsible for it. This could help a team identify whether an answer relied on information that was old, incomplete or no longer reliable. 

What it doesn’t solve 

A knowledge graph isn't the right fit for every organisation. The paper says it is most useful where information is closely connected, spread across several systems, changes frequently, or needs to support traceable and fact-based answers. Where existing systems already meet those needs and the data isn't especially interconnected, it says a relational or document database may be the better choice. 

Nor is it a system that can be built and left alone. The paper stresses that information must be updated, corrected and extended over time, with clear responsibility for data quality and currency. 

The bigger picture 

Knowledge graphs won't solve AI observability on their own. But they can make it easier to see where an answer came from and whether the information behind it is still current. 

For companies using AI with internal data, that can matter just as much as monitoring the model itself. 

For more information, read the KI Bundesverband's position paper on knowledge graphs.