Can a chatbot change someone's mind about trans rights? A new study says yes—at least for a little while.

Researchers at universities in the United States ran an experiment with more than 2,800 American adults. Half of them chatted with an AI chatbot designed to understand their personal values—like fairness, compassion, or loyalty. The other half didn't have that conversation. The researchers wanted to see if a thoughtful, personalized chat could shift people's attitudes.

It did. Participants who talked with the AI showed measurable increases in support for transgender rights right after the conversation. On a 0-to-100 scale used to measure feelings, support jumped by nearly 4 to 5 points compared to the control group. The chatbot-talking group also showed more openness to supportive actions and broader positive attitudes.

The secret, the researchers found, was moral matching. Before the chat, participants took a questionnaire that revealed their core values. The AI then tailored its messages to match. Someone who cared deeply about fairness got arguments focused on fairness. Someone guided by compassion heard about human dignity and empathy. This made the conversations feel personal rather than preachy.

But here's the catch: the effects faded over time. One week later, most of the gains had shrunk and were no longer statistically significant—in other words, the researchers couldn't be sure the change was real anymore. Still, that brief shift matters, the study's authors argue. Even temporary movement can be a starting point, especially when combined with other efforts in communities and schools.

The study was published in the journal PNAS Nexus by John Holbein and colleagues. They used a version of GPT-4o, a powerful AI language model, and gave it careful instructions to be empathetic, take the person's perspective seriously, and stay aligned with their values.

The researchers aren't claiming AI is a magic fix for prejudice. Instead, they see it as one tool among many—one that can start conversations at scale, reaching thousands of people who might never sit down for a long talk with someone different from them. "AI may serve as a scalable complement to community-based and structural approaches," they wrote.

So while a single chatbot chat won't solve deep-seated bias overnight, this study suggests it can crack the door open—and that opening, however small and temporary, might be enough to let something better through.