Research

Relevant evidence on AI-mediated communication, perceived effort, authenticity, trust, reciprocity, and human judgment.

This is an emerging research area. The evidence does not show that all AI assistance harms communication. Effects depend on context, disclosure, task type, and how much of the interaction is delegated. The manifesto focuses on relational communication, where human effort and authorship can themselves carry meaning.

Machines in the Middle

Across two studies and a replication, AI-assisted workplace help was perceived as less warm, reduced felt obligation, and reduced subsequent reciprocity.

Dishop et al., Journal of Business and Psychology, 2025. DOI

AI and perceived effort in relationship maintenance

When recipients learned that AI helped produce interpersonal messages, they inferred lower effort, reported lower relationship satisfaction, and greater relational uncertainty.

Liu, Kang & Wei, Journal of Social and Personal Relationships, 2024. DOI

AI-mediated apology in a multilingual work context

Awareness of AI mediation reduced perceived authenticity of workplace apologies and willingness to forgive. Limited AI assistance did not show the same negative effect.

Glikson & Asscher, Computers in Human Behavior, 2023. DOI

Negative perceptions of outsourcing to artificial intelligence

Six preregistered studies found negative character evaluations when people outsourced tasks to AI, with stronger effects for socio-relational tasks. Lower effort signaled lower authenticity and care.

Claessens, Veitch & Everett, Computers in Human Behavior, 2026. DOI

Trust and reliance on AI

Participants sometimes over-relied on AI advice even when it conflicted with available contextual information, demonstrating why AI-supported claims still require independent human judgment.

Klingbeil, Grutzner & Schreck, Computers in Human Behavior, 2024. DOI

Writing with AI boosts trust-building efficiency

An important counterpoint: in two preregistered trust-game experiments, AI-assisted writing achieved similar trust while taking less time. The authors emphasize transactional trust-building, showing that AI assistance is not uniformly harmful.

Purcell et al., iScience, 2025. DOI