Skip to content
Research · Nov 9, 2026

Study finds real-world AI identification practices rely on context-dependent norms rather than fixed textual features

Qualitative and computational analysis of Wikipedia editors and students shows AI-likeness judgments are contradictory, dynamic, and tied to sociolinguistic appropriateness.

Trust79
HypeLow hype

1 source · cross-referenced

ShareXLinkedInEmail
TL;DR
  • AI detection practices among Wikipedia editors and students are context-dependent and contradictory, often relying on pragmatic infelicity rather than fixed textual features.

A new arXiv preprint examines how people identify AI-generated writing in real-world settings, analyzing practices among English Wikipedia editors and students. The study combines qualitative coding with computational analysis of lexical change, grounded in linguistic anthropology, to show that AI identifications are not based on consistent textual markers but on perceived mismatches between content and context—what the authors term pragmatic infelicity.

The authors report that judgments of AI-likeness draw on diverse and often contradictory evidentiary types, including discrete textual features, content judgments, circular reasoning, and assumptions about authors’ moral character. These identifications are dynamic and contextually dependent, meaning what counts as "AI-like" shifts with the norms of a given setting.

The paper argues that tools treating AI-likeness as a context-free property risk inheriting and scaling the norms of the contexts they were trained on, potentially enforcing those norms at scale. This raises concerns about how AI detection systems might embed and propagate sociolinguistic biases, shaping not just what is detected as AI-generated but what is deemed appropriate or human in different discourse communities.

Sources
  1. 01arXiv cs.CL — Contextuality and Contradictions: Exploring Real-World Identifications of AI-Generated Writing
Also on Research

Stories may contain errors. Dispatch is assembled with AI assistance and curated by human editors; despite the trust-score filter, mistakes happen. We correct publicly — every article links to its revision history. Nothing here is financial, legal, or medical advice. Verify before relying on any claim.

© 2026 Dispatch. No ads. No sponsorships. No paid placement. Reader-supported via Ko-fi.

Built by a person who cares about honest AI news.