Watch: Secondary Bias
The post-retrieval layer of selection: how an AI model treats, weights, and is swayed by content once it has been retrieved. Unlike primary bias, it is addressable now.
Transcript
When we think about how artificial intelligence models choose what information to present, we often focus on what the model already knows from its training data. This is called primary bias, and because it is baked into the model's core memory, it is very slow and difficult to change.
But there is another critical layer that happens after a model retrieves information. This is secondary bias, and it is the layer we can actually influence right now.
Secondary bias is all about how your content is formatted, structured, presented, and weighted once the AI has found it. One common form of this is grounding bias, where the model decides how much to trust and use a specific piece of retrieved text based on how it is written.
Unlike the deep-seated beliefs of primary bias, secondary bias is addressable. You can shape how an AI model evaluates your content at the margin, simply by changing the way your pages are written and structured.