Scoping review maps trust across firsthand and secondhand human-AI communication

A scoping review of 124 publications found that research on trust in communication involving humans and artificial intelligence is concentrated in firsthand human-human, firsthand human-AI, and secondhand human-human routes, while AI-mediated secondhand routes remain comparatively sparse. The review organized 21 trust factors into four dimensions and identified four recurring route-level patterns: distorted trust calibration, trust transfer, extremity amplification, and accountability diffusion.

The analysis focused on communications in which a human decision maker evaluates a claim. In firsthand communication, a human or AI source sends the claim directly to the decision maker. In secondhand communication, an intermediary passes or changes it, creating a triad in which the source and intermediary may each be human or AI. The authors described six role arrangements as minimal motifs for comparison rather than models of complete communication networks.

Published in 2026, the review reported screening 3,810 records, including 3,765 keyword-search records returned by Google Scholar and 45 records identified through exploratory searches. Searches were conducted on February 1 and 2, 2025, and the final corpus covered publications from 1987 through 2024. The authors used the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews as a guide for reporting study selection.

Eligible material included English-language journal articles, conference papers, institutional publications, and books addressing trust, communication, and human or AI interaction. Preprints, studies of AI performance alone, and embodied or non-text multimodal interactions such as work with physical robots were excluded. The resulting evidence base mainly covered text-based and other largely unimodal settings in which people could question, accept, reject, or rely on information from a messenger.

The first author led screening, study-level extraction, and initial coding, while all three authors discussed unclear cases and integrated the resulting categories. Initial coding identified 31 provisional factors across five provisional dimensions. Collaborative review merged seven overlapping sets, reducing the list to 21 factors, and reassignment of factors in a provisional cognitive-mechanism category reduced the framework to four dimensions: Decision Maker, Messenger, Message Property, and Environment.

Decision Maker covers characteristics of the person evaluating the claim, including disposition, ability, habit, reciprocity, physiological basis, and information processing. Messenger covers cues associated with actors who create or deliver the claim, including ability, benevolence, integrity, source, and temporal dynamics. Message Property addresses the claim and its presentation through quality, quantity, process, framing, and modality, while Environment encompasses context, responsibility, task, and trust distance. The dimensions can overlap, and a single study could receive several codes.

Qualitative synthesis of studies assigned to secondhand routes produced four descriptive patterns. Distorted trust calibration, the most common Decision Maker category in 12 studies, describes situations in which available evidence about a source or changes to a claim is difficult to see, potentially leading people to assign more trust than the evidence supports. Trust transfer, the leading Messenger category in 10 studies, refers to trust in one actor extending to another along a communication route, although studies differed in the direction and strength of that transfer.

Extremity amplification, the most frequent Message Property category in nine studies, describes judgments becoming stronger or more polarized when a claim is summarized, edited, or reframed, even as important detail is lost. Accountability diffusion, the leading Environment category in six studies, describes responsibility or blame becoming harder to locate when multiple actors create, change, deliver, or deploy a claim. These labels summarize recurring ideas in the included literature rather than pooled effects, and the review did not test causal relationships among the patterns.

Based on these patterns, the authors proposed four preliminary topics for direct testing in future AI research: role inversion, linked to distorted trust calibration; cognitive offloading, linked to trust transfer; collective bias, linked to extremity amplification; and psychological accountability, linked to accountability diffusion. They also proposed route legibility as a shared starting point for design, governance, and auditing, involving recording and displaying a claim’s origin, changes, supporting evidence, and handoffs. These proposals and their accompanying hypotheses were not findings of the review and require empirical testing.

Several limitations constrain the synthesis. The six motifs do not represent larger networks with multiple sources, cycles, parallel paths, or changing roles, and the corpus did not distinguish among AI model types or systematically compare user groups and domains. The coverage map reflects study counts without considering sample size, study quality, or importance, while differences in methods prevented the authors from combining effect sizes or testing causality. Screening and coding were led by one author and were not independently repeated, and English-language searches, fixed Google Scholar result limits, and the February 2025 cutoff may have excluded relevant evidence. A sensitivity analysis retaining 115 of the 124 studies preserved all six topologies, four dimensions, and upper-category codes, but it could test only stability within the retrieved corpus rather than the completeness of the search.

Subject of Research: Scoping review of trust factors in human-AI communication routes

Article Title: A scoping review of trust in firsthand and secondhand communication involving humans and artificial intelligence

Article References: Yamamoto, K., Tanikawa, T., & Kuzuoka, H. (2026). A scoping review of trust in firsthand and secondhand communication involving humans and artificial intelligence. Discover Artificial Intelligence, 6(1), Article 1445. https://doi.org/10.1007/s44163-026-02469-5

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02469-5

Keywords: scoping review, human-AI communication, trust calibration, secondhand communication, trust transfer, message framing, accountability, AI mediation, literature mapping, scoping, review, trust