Visible Seams: Why Provenance in Human-AI Work Is More Than a Credit Statement

Human–AI work is increasingly common in writing, design, research, programming, conceptual development, and reflective practice. Yet its public description often remains imprecise. AI involvement is either concealed or marked by broad phrases such as “AI-assisted,” “created with AI,” or “AI-supported.” Such phrases may be better than silence, but they often fail to identify the role the system actually played: language correction, structural assistance, conceptual clarification, counterargument, formulation, or sustained dialogical development. This paper argues that the distinction matters. Not every use of an AI system creates authorship or co-authorship. Spell-checking, formatting, simple paraphrasing, and superficial variation do not normally amount to substantive epistemic participation. However, when an AI system materially shapes the direction, structure, concepts, objections, or central formulations of a thought, the result is no longer adequately described as isolated human work. In such cases, the relevant unit for provenance may be the coupled process through which the result emerged. The central claim is that visible seams are not merely a matter of politeness, fairness, or technical disclosure. They are part of epistemic traceability. Provenance is not decoration around a finished result; it is information about how that result became possible. The paper also identifies a social tension: transparency about AI involvement is increasingly demanded, while such transparency may be socially, institutionally, or professionally penalized. This disclosure penalty creates incentives to hide or flatten AI participation, thereby worsening provenance and undermining trust. The paper proposes a vocabulary that separates disclosure, provenance, epistemic participation, and formal authorship without collapsing them into one another.