“The cruelest feature of the nocebo effect is that it can manufacture the very harm that it was only predicting… The belief does not sit passively waiting to be confirmed or denied. People change their behavior now.” — Dr. Sanjay Gupta, CNN Chief Medical Correspondent
That observation from neurosurgeon and journalist Dr. Sanjay Gupta in our latest “Learning Machines” podcast is usually applied to medicine. Today we explore how the “nocebo effect” may also become one of the defining leadership challenges of the AI era.
In our conversation, Dr. Gupta shared the tale of two of his patients who shared a name and operative experience. Joanna 1 had been very anxious about the intervention, focused on potential risks. Joanna 2 had come in optimistic about a good outcome. On his rounds the day after, Joanna 1 was still in bed with morphine; Joanna 2 was sitting up and had applied some makeup.
Leaders, politicians, and many of us might be unintentionally creating nocebos all the time, especially around AI. The AI debate should therefore not be limited to whether technology will replace or augment human work. It should also examine how our predictions about AI influence people’s willingness to learn, experiment, adapt, and remain engaged.
“I shall harm”
In medicine, a placebo—from the Latin word placebo—is an inert or sham treatment with no therapeutic effect on the condition being treated, but is administered as though it were an active treatment. The effect refers to the beneficial changes that occur as a result of patient expectations, their interaction with the clinician, and the therapeutic context. Expectations alone can influence mood, pain, and even some physiological responses. The term placebo has long since migrated from the clinical understanding that emerged in the 1940s and 1950s into the vernacular.
Fewer people know about its darker twin introduced to the world in the early 1960s: the nocebo effect, derived from the Latin word nocēbō meaning “I shall harm” or “I will harm,” in which the expectation of harm can produce harm itself. In other words, expectations can become part of both the treatment and the illness. In this respect, prediction can simulate the placebo effect by alleviating the anxiety of anticipation and fostering confidence in healing, but it can also have the opposite effect.
Because we are increasingly turning to AI for advice on everything from relationships and workplace stress to physical ailments, it is important to examine how GenAI tools shape their responses in such instances. How it formulates its prediction and catastrophizes could significantly affect what happens next.
READ: Sanjay Gupta on Learning Machines podcast: In the age of AI, humans still win the day (July 9, 2026)
AI as a nocebo amplifier
When given only a few words of context, AI may reasonably generate responses that err on the side of caution. That can be appropriate in some situations. But it also illustrates a broader challenge for both AI and humans: predictions made from incomplete information can shape expectations, emotions, and subsequent behavior.
Below are two examples of how AI could function as potential nocebo from our respective disciplines, medicine and performance at work and healthcare and how. A brief prompt about our toxic boss and chronic exhaustion at work might yield a ChatGPT response such as: “Your symptoms are consistent with burnout. Toxic managers can seriously damage the mental health of those in their care. This pattern often gets worse over time if left unchecked. Consider whether this job is harming you and whether you should begin looking for another position.”
Nothing in that response is necessarily false. But it nudges the user toward one interpretation: your situation is dangerous, your boss is toxic, and the likely trajectory is decline. It strengthens one prediction. It could also have asked what happened at work, about missed deadlines or unusual circumstances. This might have surfaced that the user had started on an important project too late. Additional context change interpretation.
Similarly, a brief query about strong neck pain and a headache on July 5 might provoke caution that “pain between the shoulders and headaches can sometimes be signs of a serious underlying condition, including cardiovascular, neurological, or musculoskeletal disorders. Chronic stress can also cause lasting physical damage. If your symptoms persist, they could worsen or indicate a more significant problem.”
This response emphasizes serious possibilities, primes the questioner to monitor symptoms more intensely, and could increase anxiety, muscle tension, and vigilance—all of which may worsen pain. Had we been prompted to disclose that we had spent July 4th BBQ-ing and playing volleyball in the pool, the answer might have been different. The response given is not necessarily inaccurate, but it foregrounds threat.
Importantly, when properly prompted, AI can also ask clarifying questions or provide balanced, efficacy-promoting responses. The point is not that AI is inherently pessimistic, but that how models are prompted and the predictions they elicit influence what people think, feel, and do next.
Learning Machines Podcast: Watch the first episode with Vivienne Ming (May 22, 2026)
The management question
Organizations are prediction machines that gamble on future outcomes. Leaders spend all day forecasting and interpreting the future, and employees spend all day interpreting those forecasts. Managers often make predictions about layoffs, mergers, AI, restructuring, return-to-office, and performance reviews, and those predictions change behavior. The language becomes “we’re probably going to restructure,” “we’re behind our competitors” to “everybody needs to work harder because AI is coming for our industry.” And countless other messages suggesting that change has never been faster, competition has never been fiercer, and everyone is in survival mode
On AI specifically, Gupta noted that when people hear “AI will replace you,” they disengage, stop learning how to use AI, and they become less employable. “The prediction manufactures the outcome,” he said. So, the discussion is not just about whether AI will replace jobs, but about what happens when people believe it will. That is a nocebo. The nocebo effect feeds the “Tabloid in Your Head” effect that we wrote about in July 2025, leading to cognitive distortions and rumination.
For business and civic leaders, the question is whether those predictions become placebo—or nocebo—effects for the people who hear them. Stress and burnout are real, and many people face genuinely demanding work environments. At the same time, the way organizations talk about these experiences matters. If every challenge is framed as overwhelming, setback as trauma, and period of fatigue as burnout, employees may begin to interpret normal fluctuations through a more pathological lens. That does not create all stress, but it can amplify how stress is perceived, experienced, and communicated.
Dramatic and hyperbolic language may work on social media or to generate clicks. It is far less clear that it produces the behaviors that organizations—and perhaps civic and global leaders, too—actually want.
READ: Sreedhar Potarazu and Carin Isabel Knoop | Friction vs fiction: AI automators, validators, and cyborgs (May 22, 2026)
No nocebo without placebo
Medicine reminds us that expectations are part of the cure and the illness. And in that respect, as in the tale of the two Joannas, medicine is not just about diagnosis; it is also about prognosis and recovery. An experienced physician does not end every appointment reminding a patient of everything that could go wrong. They diagnose honestly, explain the risks, but also reinforce the body’s capacity to heal.
The more we understand that predictions are behavioral interventions, the more we need to turn to neuroscience and thinkers like Dr. Gupta to help us communicate our predictions in ways that encourage others to adapt rather than surrender.
As Dr. Gupta put it, “With regard to AI, young people […] may start to disengage from learning the tool that may have otherwise helped them.” For managers, educators, parents, and civic leaders, the caution is therefore to communicate in a way that focuses on agency and curiosity rather than despondency and hyperbole.
Listen to the full episode on Spotify: Learning Machines Podcast with Dr. Sanjay Gupta
Watch the podcast on YouTube: