ChatGPT Tested for Allergy-Safe Product Picks
Researchers put ChatGPT to the test for suggesting products that avoid common skin allergens. The AI gave valid recommendations nearly 80 percent of the time across 30 allergen groups. Dermatologists reviewed every answer and sorted them into five categories. Valid picks led the pack at 1,782 out of 2,250 total responses. Invalid answers were rare at just 40 cases. Cross-reactions showed up 45 times. Unknown and unresponsive categories each accounted for roughly 8 percent of results.
The study ran from September 2023 through January 2024 using GPT-4. For each allergen the chatbot was asked to name five safe products. Three different question styles were tried from simple to complex. The simplest phrasing produced the highest success rate at 85 percent valid. More complex prompts lowered accuracy. Cosmetics recommendations shone at 90 percent valid. Medications followed at 82.5 percent. Fragrance mixes and para-phenylenediamine tripped up the model most often. Peru balsam triggered frequent cross-reactive suggestions.
Experts say the tool shows promise as a helper for allergy counseling. It cannot replace a clinician or trusted reference guides. Every AI suggestion needs human verification before a patient acts on it. The technology works best when questions stay straightforward. Complicated wording confuses the model and reduces reliability. Patients should view chatbot output as a starting point not a final answer.
Dermatologists manually classified all responses during the evaluation. They looked for specific product names in each answer. Responses without a named product were excluded. The international standard patch test allergens formed the test set. This covered the most common contact dermatitis triggers. Results suggest large language models could supplement clinical workflows. Oversight remains essential for patient safety.