Evaluation format, not model capability, drives triage failure in the assessment of consumer health AI

TL;DR

This study shows that the way consumer health AI is evaluated, especially the use of forced-choice formats, significantly affects perceived triage accuracy, highlighting the need for more realistic testing conditions.

Contribution

The paper demonstrates that evaluation format biases triage performance metrics and advocates for testing AI in naturalistic, user-like interactions to better assess real-world safety.

Findings

Naturalistic interactions improve triage accuracy by 6.4 percentage points.

Forced-choice evaluation underestimates true triage performance.

Evaluation format critically influences perceived AI safety and effectiveness.

Abstract

Ramaswamy et al. reported in Nature Medicine that ChatGPT Health under-triages 51.6% of emergencies, concluding that consumer-facing AI triage poses safety risks. However, their evaluation used an exam-style protocol -- forced A/B/C/D output, knowledge suppression, and suppression of clarifying questions -- that differs fundamentally from how consumers use health chatbots. We tested five frontier LLMs (GPT-5.2, Claude Sonnet 4.6, Claude Opus 4.6, Gemini 3 Flash, Gemini 3.1 Pro) on a 17-scenario partial replication bank under constrained (exam-style, 1,275 trials) and naturalistic (patient-style messages, 850 trials) conditions, with targeted ablations and prompt-faithful checks using the authors' released prompts. Naturalistic interaction improved triage accuracy by 6.4 percentage points ($p = 0.015$). Diabetic ketoacidosis was correctly triaged in 100% of trials across all models and conditions. Asthma triage improved from 48% to 80%. The forced A/B/C/D format was the dominant failure mechanism: three models scored 0--24% with forced choice but 100% with free text (all $p < 10^{-8}$), consistently recommending emergency care in their own words while the forced-choice format registered under-triage. Prompt-faithful checks on the authors' exact released prompts confirmed the scaffold produces model-dependent, case-dependent results. Our results suggest that the headline under-triage rate is highly contingent on evaluation format and may not generalize as a stable estimate of deployed triage behavior. Valid evaluation of consumer health AI requires testing under conditions that reflect actual use.