Generative AI isn’t just one more pipe that health misinformation flows through. It shapes every stage of the problem, from cooking up false claims to catching, correcting and amplifying their fallout. That’s the conclusion of a new review by University of Maryland communication researcher Xiaoli Nan, as Ophelia Keating reports for Bioengineer.
Nan’s review, published in Frontiers in Medicine, maps how large language models operate across what she calls the health misinformation lifecycle. “A central paradox emerges: generative AI functions both as a driver of misinformation and as a potential tool for mitigation,” she writes.
On the production side, the trouble starts with fluency. Chatbots can churn out polished, expert-sounding health content at scale, complete with hallucinated facts, outdated advice and fabricated references. Medical models can be manipulated through data poisoning and adversarial prompts, while deepfakes can put convincing words in the mouths of health officials.
Detection is a mixed bag. AI can flag dubious claims, cite evidence and sort narratives, but it errs in both directions, performs unevenly across languages and can be dodged with simple paraphrasing. Nan argues it should support human fact-checkers rather than act as the final judge.
Correction shows promise. Tailored, conversational rebuttals can reduce belief in falsehoods under some conditions and may boost vaccine confidence. But they also risk users trusting a chatbot over their doctor, or debated science being presented as settled.
Related on the SLP
The stakes are real. AI-generated misinformation has been linked to lower adherence to health guidelines and higher vaccine hesitancy, and labeling content as AI-made doesn’t reliably dull its persuasive pull. Vulnerable communities that lean heavily on conversational AI may be hit hardest, and accountability for misleading outputs from proprietary models remains murky.
Nan calls for coordinated, multilevel strategies that address AI’s simultaneous roles “while preserving human oversight, institutional accountability, and public trust.”
