A researcher prompted ChatGPT to provide exercise recommendations for 26 populations identified in the American College of Sports Medicine’s Guidelines for Exercise Testing and Prescription, which is considered the gold standard within the field. Populations included healthy adults, children and teens, older adults, people with cardiovascular disease, and people with obesity, among others.
Most of the ChatGPT recommendations were factually correct, with a 90.7 percent accuracy rate when compared with “a gold-standard reference source.” But the researchers write that the recommendations weren’t comprehensive enough, covering only 41.2 percent of the guidelines.
The tool also generated misinformation when it came to exercise for people with hypertension, fibromyalgia, cancer and other conditions. The answers were least accurate for people with hypertension, even recommending against the vigorous exercise that’s appropriate for most within that group.
The AI-generated answers also misinformed readers about whether they should exercise at all, prompting them to get medical clearance before exercising 53 percent of the time, even when the population in question didn’t need to ask a doctor before starting a workout regimen. This could discourage people from working out, the researchers warn, triggering undue concern and unnecessary doctor’s visits.
Nor were the recommendations as readable as they should have been, the researchers say: On average, they were considered “difficult to read” and were written at a college level.
Overall, the researchers conclude, health-care providers and patients alike should use caution when relying solely on AI for exercise recommendations, and future studies should focus on measuring “appropriateness, costs, feasibility” and other factors.
Rachel Carter is a health and wellness expert dedicated to helping readers lead healthier lives. With a background in nutrition, she offers evidence-based advice on fitness, nutrition, and mental well-being.