What is AI Mode? Meaning, how it works and its impact on search
A knowledge graph represents knowledge as a network of entities and their relationships. Rather than storing information in isolation, it reveals how people, places, companies, products or topics are connected.
In this way, individual facts are combined to form an interconnected knowledge model capable of representing relationships and meanings. Knowledge graphs are used by search engines, AI systems and businesses, amongst others, to better organise information, answer complex questions and make knowledge usable in a context-sensitive manner.
Key facts about the Knowledge Graph at a glance
- A knowledge graph links entities, properties and relationships.
- It is not the same as a graph database or the Google Knowledge Panel.
- Google uses its own Knowledge Graph to categorise real-world entities and their connections.
- Unique entities and semantic relationships are relevant for SEO and GEO.
- Knowledge graphs can also support AI, RAG and GraphRAG applications by providing structured context.
What is a knowledge graph?
A knowledge graph is a model for the structured representation of knowledge. Individual objects are recorded as entities or nodes and linked to one another via defined relationships.
A simple example is the entity ‘Berlin’. It can be linked to other entities and pieces of information:
Berlin → is the capital of → Germany
Germany → is situated in → Europe
Berlin → has a landmark → the Brandenburg Gate
Such statements are often represented as triples: they consist of an entity, a relationship and another entity or piece of information. Together, many of these links form a knowledge network. The key added value therefore lies not in collecting as much data as possible, but in explicitly mapping connections and meaning. This enables systems to recognise how different pieces of information relate to one another.
How does a knowledge graph work?
Put simply, a knowledge graph works as follows:
- Capturing entities: People, places, companies, products or topics are created as independent objects.
- Add properties: Entities are assigned characteristics such as location, sector or population.
- Establishing relationships: Links show how entities are connected, for example:
Berlin → is the capital of → Germany → is located in → Europe. - Merging data: Different sources and names can be associated with the same entity.
- Querying relationships: More complex questions can be answered on the basis of these links.
The key advantage: A knowledge graph maps not only individual facts, but also their meaning and relationships. This enables search engines, AI systems or business applications to utilise knowledge in a context-aware manner.
What role do ontologies play?
In more complex knowledge graphs, an ontology describes which types of entities and relationships are permitted. For example, it specifies that a city is situated in a country or that a company offers certain services. This creates a structured set of rules for the semantic linking of knowledge.
What technologies are used for knowledge graphs?
Various data models, standards and query languages are used for the technical implementation of knowledge graphs. The technology used depends, amongst other things, on how the information is to be structured, stored and subsequently queried.
The most important building blocks include:
- RDF (Resource Description Framework): A widely used standard for the structured description of information and its relationships.
- Triples: RDF often represents statements as a triplet structure consisting of a subject, a predicate and an object – for example: ‘Berlin – is the capital of – Germany’.
- SPARQL: A query language that allows RDF-based knowledge graphs to be searched in a targeted manner and relationships between entities to be queried.
- Property Graphs: An alternative graph model in which both nodes and relationships can possess additional properties.
- Graph databases: Specialised databases that can efficiently store highly interconnected information and quickly traverse relationships between entities.
It is important to draw a distinction: a knowledge graph is not automatically a graph database. The graph database primarily provides the technical infrastructure for storage and processing. A knowledge graph supplements this structure with the semantic meaning of entities, relationships and rules.
Knowledge graph, graph database and vector database – what is the difference?
Term | Main Task | Typical strength |
|---|---|---|
Knowledge Graph | Semantic Modeling of Knowledge and Relationships | Context and Meaning |
Graph Database | Storing and Retrieving Highly Interconnected Data | Efficient Traversal of Relationships |
Relational Database | Managing Structured Data in Tables | clear, stable data schemas |
Vector Database | Finding Similarities Between Vectors | Semantic Similarity and Retrieval |
A graph database can form the technical basis of a knowledge graph. However, it is not the same as the knowledge graph itself.
Vector databases, too, serve a different purpose: they identify semantically similar information. A knowledge graph, by contrast, explicitly maps how entities are connected to one another. In modern AI systems, both approaches can be combined.
What a Knowledge Graph is not
A knowledge graph is often confused with related terms. The key differences are as follows:
- Knowledge graph ≠ graph database: A graph database is a technical database for storing and querying interconnected data. A knowledge graph, on the other hand, describes a knowledge model consisting of entities, properties and semantic relationships.
- Knowledge graph ≠ vector database: Vector databases store mathematical representations of content and enable similarity searches. Knowledge graphs, on the other hand, map explicit relationships between entities.
- Knowledge Graph ≠ Knowledge Panel: The Google Knowledge Graph is a knowledge base. A Knowledge Panel is merely one possible visual representation of such information in Google search results.
The terms are partly related, but fulfil different functions.
What is the Google Knowledge Graph?
The Google Knowledge Graph is Google’s knowledge base for entities and their relationships. It helps the search engine not only to recognise people, places, companies, organisations and other real-world entities as search terms, but also to categorise them by context and link them together.
Where does the data come from?
Google draws on various sources for this. These include publicly available information, licensed data, and details provided by rights holders and other trusted sources. The aim is to consolidate information about entities and assign it as unambiguously as possible.
What does Google use the Knowledge Graph for?
Among other things, the Knowledge Graph helps Google to
- recognise entities and distinguish them from one another,
- better understand ambiguous search queries,
- establish relationships between people, places, organisations or topics,
- and provide relevant facts directly in the search results.
In this way, it supports Google’s evolution from a purely keyword-based search towards a more semantic understanding of search queries.
Why is the Knowledge Graph relevant for SEO and GEO?
Structured data, consistent company information and clear entity signals can help search engines better categorise content, brands, people or services.
This is relevant for SEO and GEO because modern search and AI systems increasingly process information in the context of entities and relationships. However, a well-structured understanding of entities alone does not guarantee a Knowledge Panel, visibility in AI overviews or generative AI responses. Rather, it creates a better semantic foundation upon which further signals – such as content quality, authority and external mentions – can build.
Knowledge Graph and AI: What do RAG and GraphRAG have to do with it?
In Retrieval-Augmented Generation (RAG), relevant information is first retrieved from a knowledge source and then provided to a language model as additional context. Knowledge graphs can supplement this knowledge base. They do not merely find similar documents, but map out explicit relationships – for example:
GEO → enhances → SEO
GEO → pursues the goal of → AI visibility
AI visibility → measurable via → brand mention rate
With GraphRAG, such graph-based relationships are specifically incorporated into the retrieval process.
Vector search identifies similarities. A knowledge graph provides relationships. Both approaches can be useful in combination, particularly when dealing with complex organisational knowledge.
What are some typical applications for knowledge graphs?
Concepts and entities are considered in context.
Knowledge spread across several systems is linked together.
Relationships between products, users or interests are taken into account in recommendations.
Topics, target audiences, services and search intent can be linked semantically.
Knowledge graphs can provide RAG and GraphRAG systems with structured context.
Relationships between accounts, individuals, transactions, devices or organisations can reveal suspicious patterns.
Practical examples
E-commerce: An online shop can link products, categories, brands and accessories together:
- Camera → compatible with → lens → belongs to → brand.
This enables search and recommendation systems not only to find similar products, but also to take specific relationships – such as compatibility or affiliation – into account.
Corporate knowledge: Corporate information is often spread across various systems. A knowledge graph can, for example, establish the following relationships:
- Employee → works on → project → belongs to → client.
This makes it possible to find information across systems and answer questions that would be difficult to resolve using a traditional keyword search.
When is it worth having your own knowledge graph?
A knowledge graph is particularly useful when information is spread across many systems and the relationships between these pieces of information generate business value.
A marketing team could, for example, link keywords, content, services, target audiences and search intentions. This reveals which content addresses which search intention, which service it relates to, and which target audience it is relevant for.
For simple, clearly structured datasets, however, a knowledge graph is not automatically the best solution. The crucial question is therefore not: “Can we represent our data as a graph?” but rather: “Does linking this data help us make better use of our knowledge?”
How is a knowledge graph created?
What knowledge should be linked, and to what end?
For example, companies, products, services, people or topics.
For example, content → addresses → search intent.
Identical entities from different sources must be uniquely identified.
New data and relationships are continuously added and verified.
Conclusion: A knowledge graph turns data into connections
A knowledge graph organises knowledge as a network of entities and relationships. This is precisely why the concept is relevant to search engines, corporate knowledge and modern AI systems.
For SEO and GEO, this embodies a key fundamental principle: systems must not only be able to find content, but also classify brands, topics and contexts as unambiguously as possible.
Not every business needs its own knowledge graph. However, where complex information, semantic search or AI applications come together, it can turn isolated data into genuinely usable knowledge.
FAQ: Frequently asked questions about AI Mode
A knowledge graph represents knowledge as a network. Entities such as companies, people, or products are linked to one another through defined relationships.
Not necessarily. It is primarily a knowledge model. A graph database can serve as its technical foundation.
The Knowledge Graph maps knowledge and relationships. The Knowledge Panel is a visible information box in Google Search.
It highlights the importance of entities and relationships. For SEO, therefore, clear entity signals, structured data, and semantically clear content are relevant.
It can provide structured context to AI systems and, in particular, complement RAG or GraphRAG architectures.
