Claims: small building blocks, powerful possibilities

Claims are one of the things I find most powerful about Geo. The idea is simple: take a statement, make it clear enough to stand on its own, and connect it to where it came from. Once you start doing that, you begin to see how much you can build with it.

An essay, a news story, a podcast, even a conversation with a buddy can contain dozens of claims. Someone says something happened. Someone explains why they think it happened. Someone else disagrees. Within that conversation are individual statements we can pull out, examine and connect to other things we know.

We call these atomic claims. Each one expresses a single idea, with enough context for someone to understand what is being said. That makes it easier to check a particular detail or challenge a particular argument.

Say your buddy tells you, “I had a burger for breakfast this morning.” That is a claim. A video of them eating it could help support it. You would still want to know when the video was taken and whether it shows what they say it does. If the claim is about a scientific breakthrough, you can look at the research paper, the experiment and its results.

The useful part is being able to follow the statement back to the evidence. That is provenance: knowing where something came from and how it reached you.

A source gives you something to check. You still have to assess whether it supports the claim. Some claims can be shown to be true or false. Others remain uncertain, and opinions need to be considered through the reasons and evidence behind them. Keeping those differences clear is part of making the knowledge useful.

Now imagine these claims as Lego pieces.

A claim can be connected through its sources to the people who made it, and linked to the topic it is about, an event, a location or a particular time. It can support another claim or be challenged by one. Each relationship adds context and gives you another way to explore the information.

You can reuse the same claim in a news story, a topic page, a debate or a research collection, while keeping the connection to its source. As more of these pieces are connected, you have a knowledge graph that people can explore and build with.

This is part of what we do on Geo. We break news stories down into claims and organise them into collections that help tell the story. It gives readers a way to examine the individual statements within it.

Take this story on Geo about a winning video losing its prize over AI. Nikon runs a competition for videos filmed through microscopes. After announcing a winner, it reviewed the entry and disqualified it for breaking its rules on generative AI.

You can already imagine the conversation: how much AI is too much? Does using it to improve a video count as cheating? Should the judges have spotted the problem before awarding the prize?

On Geo, the story is broken into separate claims about the disqualification, the competition rules and the entrant’s response. The entrant said he used AI for processing, but denied generating the scene or its movement. There are also opinions about whether Nikon made the right decision and who bears responsibility.

Each part gives you something specific to examine. You can check what the rules say, look at the evidence and consider the entrant’s explanation. You can accept that someone was disqualified while disagreeing about whether the decision was fair. That is much more useful than treating the whole story as one thing to agree or disagree with.

We also process podcast transcripts into claims, keeping the connection to the original conversation, the people speaking and the topics they discuss. The same happens with Geo debates. Arguments from the conversation become claims that others can explore, question and build on. The original conversation remains important because it helps you understand the context in which something was said.

This becomes even more interesting when you think about AI agents.

If you are building a knowledge base for an agent, the relationships between pieces of information matter. Who said this? When? About which event? What supports it? Is there another claim that contradicts it?

A knowledge graph makes those connections explicit. An agent can use them to find relevant information and follow it back to its source. This can help with questions that require connecting information from several documents, especially when the answer is spread across different sources.

It also gives us a way to be more deliberate about what we put into an agent’s context. We can retrieve the relevant claims, relationships and supporting passages. Whether that saves tokens depends on how the system is built, but making knowledge easier to find and check is already valuable.

For anyone building with agents, I think this deserves attention now. The quality of the knowledge we give them, and how we organise it, will shape how useful they can be.

That is why I find these small building blocks so exciting. A claim from a podcast can become part of a debate. A claim from a news story can help someone research a topic. Connections across those sources can lead to new questions and things worth investigating.

We can keep reusing, checking and organising what we know, while leaving a path for the next person to see where it came from.

If you are curious about Geo and what I work on, come and explore. We are putting together some interesting content, and there is plenty to get into. Follow a claim, check its source, explore the connections and add what you know.

There is a lot we can build from there.