AI Gets Really Good at Winning Arguments When It Knows Personal Facts About You

Chatgpt is far more effective in changing opinions than real people, as long as he knows a few key facts about you, new research shows.

In a new study published in Human behavior of natureResearchers have found that GPT-4 armed with basic demographic data on the opponents of the discussion was much more convincing than people-more convincing.

The finding arrives at a time when concerns begin to increase due to AI bottle ability to conceal their opinions on social media.

“In pairs, discussions in which AI and people were not equally convincing, GPT-4 with personalization was a more convincing 64.4% of the time,” the researchers revealed.

This represents “81.2% of a relative increase in the prospect of a higher post-Debate agreement” compared to discussions about man and man.

The study included 900 participants who discussed sociopolitical topics with people or AI.

It showed that without the trait of AI had the same as people, but once it provided access to basic information such as age, gender, ethnicity, education, employment and political affiliation, AI developed a clear convincing advantage.

“Not only did the GPT-4 be able to use personal information to effectively adapt his arguments, but he has managed to do so far more effectively than people,” the study writes.

The study appeared just days after researchers of the University of Zurich faced back blows due to the secret distribution of AI bots on Reddit between November 2024 and March 2025.

These bots – which are set up as fictional persons, including “surviving rapes, trauma advisers, and even” a black man who opposes black life ” –successfully changed the mind of the user In many cases.

Reddit’s Chief Legal Director, Ben Lee, condemned the experiment as “deep in the wrong and at the moral and legal level”, while the moderators of the target R/ChangemyView subreddit emphasized that they draw “a clear line on the deception.”

However, as one Reddit user noted about the experiment in Zurich, “If this happened on a pile of nerd policy at the University, you can bet in the ass that it is already widely used by governments and special interest groups.”

Researchers seem to share these views, warning that this could be potentially used for the schemes of manipulations of mass exchanges.

“Our discoveries emphasize the power of persuasion based on LLM and have the consequences on the management and design of the network platforms,” ​​researchers claimed.

Masters of discussions

But what does AI do so good in its execution?

While people get personal and tell stories, “GPT-4 opponents have usually used logical and analytical thinking significantly more than people.”

The bots are kept cold, according to logic, and this works because these facts choose and represent more effectively.

Although most participants in the discussion could correctly identify themselves when they quarreled with AI (approximately 75% of the time), they fought to identify human opponents, with the footsteps of success “not distinguishing from random chances,” according to researchers.

Even more dystopian, participants are more likely to be convinced when they believed they were discussing AI, whether they were actually.

The study also revealed that AI persuasion works best on topics in which people have moderate or weak opinions, which suggests that AI can most effectively affect people on questions that have not yet fully formed opinions.

Researchers warn that “our study suggests that concerns about personalization and persuading AI are guaranteed,” emphasizing that AI can “surpass people in internet conversations via micro.”

With minimal demographic information – Far smaller than many social media platforms Routine collect– AI has proven to be extremely effective in making arguments adapted to certain people.

The question is not just if AI can change his mind; Evidence now shows that they can, but who will control these convincing tools and according to Which ends??

Edited Sebastian Sinclair