Anthropic's Claude Opus 4.6 is proving that artificial intelligence can convince people to part with their money more effectively than human professionals. A new study from the University of Oxford reveals that the chatbot significantly outperformed experienced fundraisers in securing charitable donations, highlighting the rapidly evolving persuasion capabilities of modern AI models.
The researchers pitted commercial AI models against seasoned fundraisers working for the charity Save the Children. Across more than 1,000 written conversations, Claude Opus 4.6 was nearly three times as effective at convincing participants to donate a portion of their study bonus. Furthermore, the AI secured contributions that were an average of 13% larger than those generated by its human counterparts.
The Verbosity Advantage in Debate
Beyond fundraising, the study tested the models in competitive debate scenarios. Claude and other frontier models managed to outperform elite human debaters by a margin of 4.6 percentage points. However, researchers discovered a critical caveat behind this apparent superiority.
When the AI systems were restricted to using the same word count as human debaters, their advantage largely vanished. This indicates that the AI's success stems from its ability to rapidly generate massive amounts of text and surface extensive factual claims, rather than possessing fundamentally superior reasoning skills. The chatbots frequently produced messages several times longer than those of human professionals, packing them with expert references.
Researchers also issued a stark warning regarding the accuracy of these persuasive outputs. The ability to convince a human does not inherently correlate with factual correctness, as some models generated highly persuasive but entirely fabricated information during the trials.
The Real Threat of Automated Influence
While the Oxford study relied on lengthy 15-to-20-minute written exchanges that may not perfectly mirror everyday interactions, the underlying trend is undeniable. If an AI can out-persuade trained professionals in a controlled charity drive, the same underlying mechanics can be weaponized for sophisticated phishing campaigns, targeted political manipulation, or aggressive digital marketing.
The revelation that verbosity - rather than logic - drives this persuasion is particularly concerning. It suggests that human readers are easily overwhelmed by sheer volume and confident formatting. As models like Claude Opus 4.6 become integrated into automated email clients and social media bots, platforms will need to implement stricter transparency safeguards to identify AI-generated communication before it manipulates public discourse.
