Which AI is most persuasive? New working paper w/ Zhongren Chen & Quan Le, we tested 7 frontier LLMs on 19k people. Ranking: (1) Claude; (2, tied) GPT, Gemini (3) Grok. Consistent across issues and bipartisan stances
Conversation
The bigger finding: all of these models now consistently are more persuasive than actual campaign advertisements. Our earlier work found that chatting with older models was around as persuasive as watching a campaign ad. That's no longer true — frontier LLMs have pulled ahead.
Post your reply
Surprise result that differs from some prior research: information-based prompting doesn't uniformly help. It boosts Claude and Grok but hurts GPT and shows mixed results for Gemini. Prompting effects are model-dependent.
As LLMs get smarter at reasoning and coding, they're also getting more persuasive. That's a real concern for democratic societies and one that needs continuous benchmarking as models improve. Feedback welcome on this working paper! Link:
