Belief Stability

Do AI models hold real beliefs about the brands they recommend?

When a buying question is reworded, the recommended brand often changes. This study asks whether the model has a stable preference underneath the wording.

A · IF WORDING IS ALL THERE ISB · IF THERE IS A BELIEFOne buying question, asked 15 waysThe same question, asked 15 waysFour different brands, no signalOne brand keeps winning, a beliefeach dot = one answer · color = brand named
Ask the same buying question 15 ways and count how often the same brand comes back. Scattered picks mean the wording decides, a steady pick means there is a belief underneath
Setup20 B2B software buying situations, 5 wordings each, 3 reruns, on GPT-5.6, Gemini 3.8 Flash and Claude Sonnet 5 with web search on
MeasureThree ways of asking: a plain question, a question that asks for one pick, and a plain question with the pick asked in a follow-up turn
StatusPreregistered on OSF in September. Results go up here when the registration embargo lifts on October 26

HypothesisAsking for the pick in a follow-up turn changes how consistent the picks are by at least 0.10, compared with asking for it in the same message. Tested both ways, since a second turn could make answers less reliable rather than more

Drafts and data are added here as each study finishes. Nothing here is peer reviewed unless it says so