When fluency outpaces provenance: source traceability and digital health literacy against AI health misinformation
- 1School of Culture and Creativity, Beijing Normal–Hong Kong Baptist University (BNBU)
Abstract
Generative artificial intelligence (AI) can make health explanations accessible while separating persuasive language from inspectable evidence. This creates a public health problem beyond plainly false content: an answer may mislead through unsupported attribution, obsolete evidence, omitted uncertainty, or inappropriate application. This Conceptual Analysis uses a targeted integrative approach to connect health literacy, misinformation, trust, large language model evaluation, and content provenance around the pathway from a generated claim to a health decision. Its contribution is an integration and operationalization of established concepts, not a new literacy construct or truth-certification mechanism. We distinguish five source traceability functions: origin, claim, evidentiary, procedural, and correction traceability. These functions interact with four iterative, context-dependent user practices: inspect, triangulate, calibrate, and escalate. Calibrated trust means discriminating between stronger and weaker support and relying proportionately on it, rather than maximizing or minimizing trust. Effective traceability also requires accessible, comprehensible evidence pathways and reasonable verification effort. Institutional and platform accountability is therefore part of the model, not an optional supplement to individual competence. We provide illustrative operational indicators, provisional risk-tiered safeguards, and six testable propositions organized around evidentiary discrimination, confidence calibration, decision appropriateness, verification behavior, verification burden, and escalation behavior. Neither the unified model nor its indicators or risk tiers have been empirically validated. The framework offers a decision-oriented research and design architecture for evaluating when evidence visibility and literacy can jointly support safer use of synthetic health information.
Related
- Digital Nutrition Care Ecosystem: Integrating Critical AI Nutrition Literacy, Countering Nutrition Misinformation, and Dietitian-Led Care
- When conviction outpaces evidence: diagnostic boundary failures of large language models in adult ADHD self-assessment
- Item‐level verbal fluency psycholinguistic metrics predict preclinical Alzheimer's disease
- The Nature of Phonological Impairments in Dyslexia and Stuttering: Evidence from Phonemic, Semantic, and Design Fluency Tasks
- Telephone-administered semantic fluency task in older adults without dementia in relation to in-person testing and plasma phosphorylated tau-217
BibTeX
@article{peng2026when,
title = {{When fluency outpaces provenance: source traceability and digital health literacy against AI health misinformation}},
author = {Mingsheng Peng and Lu Feng},
journal = {Frontiers in Public Health},
year = {2026},
month = oct,
doi = {10.3389/fpubh.2026.1986759},
url = {https://www.frontiersin.org/articles/10.3389/fpubh.2026.1986759/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - When fluency outpaces provenance: source traceability and digital health literacy against AI health misinformation
AU - Mingsheng Peng
AU - Lu Feng
PY - 2026
DA - 2026/10/20
JO - Frontiers in Public Health
DO - 10.3389/fpubh.2026.1986759
UR - https://www.frontiersin.org/articles/10.3389/fpubh.2026.1986759/full
AB - Generative artificial intelligence (AI) can make health explanations accessible while separating persuasive language from inspectable evidence. This creates a public health problem beyond plainly false content: an answer may mislead through unsupported attribution, obsolete evidence, omitted uncertainty, or inappropriate application. This Conceptual Analysis uses a targeted integrative approach to connect health literacy, misinformation, trust, large language model evaluation, and content provenance around the pathway from a generated claim to a health decision. Its contribution is an integration and operationalization of established concepts, not a new literacy construct or truth-certification mechanism. We distinguish five source traceability functions: origin, claim, evidentiary, procedural, and correction traceability. These functions interact with four iterative, context-dependent user practices: inspect, triangulate, calibrate, and escalate. Calibrated trust means discriminating between stronger and weaker support and relying proportionately on it, rather than maximizing or minimizing trust. Effective traceability also requires accessible, comprehensible evidence pathways and reasonable verification effort. Institutional and platform accountability is therefore part of the model, not an optional supplement to individual competence. We provide illustrative operational indicators, provisional risk-tiered safeguards, and six testable propositions organized around evidentiary discrimination, confidence calibration, decision appropriateness, verification behavior, verification burden, and escalation behavior. Neither the unified model nor its indicators or risk tiers have been empirically validated. The framework offers a decision-oriented research and design architecture for evaluating when evidence visibility and literacy can jointly support safer use of synthetic health information.
SN - 2296-2565
ER -