What are the benefits and limitations of website knowledge graphs? Distinguishing search visibility from organizing website knowledge
The main purpose of a website knowledge graph is to make the site’s entities, concepts, and relationships explicit. It can help organize knowledge and may also be read by AI agents, but that does not mean it will improve search rankings or make language models “remember a brand.” Before evaluating it, distinguish between two goals: influencing Google’s Knowledge Graph and building a website’s own knowledge package. A website can only influence the former indirectly. For the latter, you need to consider whether agents will use it and whether its content can be kept up to date as the website changes.
The same name, but different graphs to work with
In a July 2026 article, Duane Forrester distinguished between Google’s Knowledge Graph and an entity-relationship graph maintained by a website itself. He explained that signals such as structured data and consistent third-party corroboration can indirectly influence Google’s understanding of entities. This does not mean a website can directly rewrite Google’s graph.
Another approach is to organize website knowledge into files that explicitly record its relationships. That same month, Slobodan Manic introduced a Markdown-based OKF knowledge package, noting that its original purpose was to organize internal enterprise knowledge. At the time, the version was v0.1 and the standard had not been finalized. Using it on a public website was the author’s experiment with extending its use, not an official requirement.
The benefit lies in clarifying relationships, not merely changing file formats
Providing a concise Markdown copy of each page is still not the same as building a knowledge graph. Even if two pages mention the same concept, they may not explain how their subjects relate: does one describe an overarching framework and the other a subordinate concept, or do they describe parallel goals? Slobodan Manic emphasizes that a graph’s value lies in making these relationships explicit, rather than simply making individual pages easier to read.
- Organizing website knowledge: Slobodan Manic says that writing the knowledge package forced him to clarify relationships between concepts and revealed gaps in the existing content. This is one person’s experience, not a guarantee of results.
- Helping agents understand: If agents read the knowledge package and follow its relationships when querying it, they may gain a fuller understanding of how concepts connect than they would by retrieving pages individually. The benefit depends on whether agents use this approach.
Machine-readable does not mean it will be used or directly update a model’s memory
Slobodan Manic explicitly describes agents querying a website’s graph as a direction to explore, not a prediction. According to his July 2026 account, no agents were reading websites this way at the time. It should therefore not be presented as a widely used way to access websites today.
Duane Forrester also notes that a language model’s parametric memory differs from the information it looks up in real time when answering: there is no brand node in the model’s weights that a website can directly edit. Providing entity-relationship data therefore does not directly update a model’s memory.
The maintenance cost is concrete: if a knowledge package is another copy of the website, it must be updated whenever the website changes, or it may retain outdated content. As Slobodan Manic cautions, the reasons for investing in this work should remain valid even if agents never visit—for example, clarifying the site’s concepts—rather than relying solely on unproven visibility gains.
Website knowledge graphs: perspectives from across the web
Does building a knowledge graph guarantee visibility in a Google knowledge panel?
No. A 2015 article used a case study to argue that certain practices could help secure knowledge panel visibility and targeted organic traffic. However, the supplied excerpt does not list the specific steps and cannot be treated as current rules. Another article argues that most people will struggle to gain Google’s recognition of their content and recommends abandoning the pursuit. These reflect different judgments and do not provide a basis for promising knowledge panel visibility.
Does adoption of related formats by different Google teams mean they offer ranking benefits?
No. At the time described in Slobodan Manic’s article, Google’s Search team said that any ranking impact of llms.txt was purely speculative. The Chrome team added an llms.txt check to an audit assessing whether websites were ready for agents to use, while the data team separately released OKF. These teams were working toward different purposes. Including a format in an audit or releasing it does not prove that it improves search rankings.
When a user asks a question, do chat products turn website markup into a knowledge graph in real time?
One author rejects this claim. The argument is that building a graph requires identifiers, relationship rules, and alignment with existing entities. Chat products have not publicly exposed infrastructure that could complete this work between receiving a question and producing an answer. This is the author’s assessment of a specific claim, not a conclusion that no system could ever build or query a graph under any circumstances.
References
- Entity Mapping Works On Google. Does Any Of It Reach ChatGPT?www.searchenginejournal.com · 2026-07-30T12:00:50+00:00 — Source used in the article above.
- The Beginner’s Guide to Google’s Knowledge Graphneilpatel.com · 2015-06-30T14:00:31+00:00 — Discusses: Does building a knowledge graph guarantee visibility in a Google knowledge panel?
- SEO 教學:45 個Website新手處理 SEO 要注意的小技巧和清單ithelp.ithome.com.tw — Discusses: Does building a knowledge graph guarantee visibility in a Google knowledge panel?
- Google’s Open Knowledge Format Could Work For Websites, Toowww.searchenginejournal.com · 2026-07-06T09:00:43+00:00 — Discusses: Does adoption of related formats by different Google teams mean they offer ranking benefits?
- Schema, LLMs & The Low Bar For ‘Evidence’ In GEOwww.searchenginejournal.com · 2026-06-16T11:00:57+00:00 — Discusses: When a user asks a question, do chat products turn website markup into a knowledge graph in real time?


Join the discussion 0 comments
What are you working through? Share a question or experience and let’s find the next step.
No comments yet. Start the conversation.