AI Persuasion and Financial-Decision Making: Experimental Evidence on Dominated Investment Choices

Key Findings

• — A chatbot adviser that promoted the optimal fund increased optimal choices by over 20 percentage points; one that promoted the dominated fund reduced them by almost 30 points (paper abstract)
• — Without advice about 64% chose the cheaper provider; with a neutrally presented AI chatbot recommending it about 87% did, and with the same AI instructed to recommend the worse provider about 34% did (University of Bayreuth release)
• — The chatbot's influence was undiminished when it was framed as a bank's chatbot and participants knew the bank could profit from their choice
• — Incentivised human advice also steered choices toward the promoted fund but was significantly less effective than the AI adviser
• — Participants investing larger amounts chose correctly over 87% of the time unaided, and a biased AI recommendation cut this to under 50% (University of Bayreuth release)

Methodology Notes

Incentivised online experiment (about 3,700 participants per the University of Bayreuth release) with fictitious funds tracking the same artificial index and differing on two fees, so that one provider was strictly cheaper for each chosen savings rate. Advice arms: an AI chatbot, a chatbot presented as a bank's AI with the conflict disclosed, and an incentivised human adviser, each steered to recommend either the better or the worse fund. The model used is not named in the release or the RePEc abstract. CESifo Working Paper No. 12925 (2026). The ifo.de PDF returned a bot-check page, so the paper text was not read: verification is by the RePEc/IDEAS record (title, authors, abstract, series number) and the University of Bayreuth release Nr. 064/2026 dated 2026-08-25, which is used as the publication date (the working paper's own issue date was not read).

Sources