Experiment tests summary-induced preference change

This proposed study asks whether a language model can strengthen later preferences by misdescribing an earlier choice. Participants receive an accurate or inaccurate summary, then justify it or describe facts. Later, an original-choice record or an uninformative record would precede a private interpretation and retelling with a frozen model. A receipt-sensitive shift beyond measured processes would support the proposed feedback loop.