Grounding Bias
A form of secondary bias: how much an AI model defers to, or discounts, retrieved sources once they are grounded into the context, rather than leaning on what it already believed.
Grounding bias is a form of secondary bias. It is how an AI model treats retrieved sources once they are grounded into the context: how much it defers to them, and how much it keeps leaning on what it already believed before the search.
It sits opposite primary bias, the model's pre-retrieval instinct. Where primary bias fires before any source is read, grounding bias is what happens after, when the retrieved grounding snippets either shift the answer or fail to. It is produced during grounding. We surface it by running paired ungrounded and grounded probes and comparing the two, the method behind our Tree Walker.