Listen: Primary Bias

When an AI search engine chooses which sources to show you, its decision isn't just based on the search results it finds. It is heavily influenced by something called primary bias.

Primary bias is an AI model's ungrounded confidence in a specific source or entity. This bias is formed during the model's initial training, long before any search or retrieval ever takes place. It is a pre-judgment baked directly into the model's weights. Because this bias lives deep within the training data, it is incredibly slow to change.

In the world of AI search, primary bias is actually the single largest factor in whether a source gets selected and shown to a user. We measure this influence by looking at its direct effect on the selection rate of different sources.

Once a search is performed, other biases come into play. These are known as secondary biases, which include grounding bias. But it is primary bias—that initial, built-in preference—that sets the stage for everything else.