How do readers project author identity from textual cues during interpretation?
This explores how readers build a mental picture of *who wrote this* — confidence, status, character, expertise — from cues on the page, and how fragile and projection-driven that act turns out to be.
This explores how readers construct an author's identity from textual cues — and the corpus suggests that what feels like "reading the writer off the page" is really an act of projection, easily steered by surface signals and easily distorted by who's doing the reading. Start with the clearest demonstration that the projection is real and movable: a large study of writers and readers found that AI writing assistance shifted *every* measured dimension of perceived persona — confidence, agreeableness, quality, even perceived social privilege — all in the same direction, without the underlying person changing at all Does AI writing assistance change how readers perceive the writer?. The identity readers infer lives in the text's stylistic fingerprints, not in the human behind it.
That raises the question of *which* cues carry the signal. Style is one channel, and it's a shallow one: a model can identify authorship from style patterns with 95% accuracy, yet "detection without interpretation remains cataloguing, not criticism" — pattern-matching the hand without grasping why the choices mean anything Can language models truly understand literary style?. Deeper identity cues hide above the sentence, in discourse-level choices like how agency and chronology are arranged; these resist mimicry precisely because they encode a writer's structural habits rather than their word choices Can AI stories be detected without analyzing writing style?. Reading an author, then, means simultaneously tracking segments, intentions, and what's salient — three layers that constrain each other during comprehension How do readers track segments, purposes, and salience together?.
Here's the thing readers may not realize they're doing: a large part of the "author" they project is *authority* they supply themselves. The force of an argument depends on the standing of the thinker — reputation, track record — and that standing is built in the social world, not carried in the words Can language models distinguish expert arguments from common assumptions?. When the social context is stripped away, the cues that remain become exploitable: LLM judges (and, by implication, hurried human ones) fall for fake credentials and rich formatting, treating authority *signals* as authority itself Can LLM judges be fooled by fake credentials and formatting?. The projected author is partly a costume the reader is willing to believe.
And the reader is never neutral. The same sentence yields genuinely different readings depending on a reader's social position — disagreement here is meaningful information, not noise Why do readers interpret the same sentence so differently?. In persuasion, a reader's prior ideology predicts the outcome better than anything the writer actually said Does what readers believe matter more than what debaters say?. So projecting author identity is a two-sided act: the text offers cues, but the reader's own standpoint decides how those cues resolve into a person.
The quietly unsettling payoff comes from the machine mirror. When an LLM generates text, there's no fixed author behind it at all — it holds a superposition of possible characters and samples one at generation time, so regenerating the same prompt produces a different "writer" each time Do large language models actually commit to a single character?. Readers will confidently project a stable identity onto something that never committed to one — which suggests the author we "find" in any text was always partly authored by us.
Sources 9 notes
A study of 2,939 writers and 11,091 readers found AI assistance shifted every tested dimension—29 total—toward extremism, confidence, quality, agreeableness, and perceived privilege. Distortions were statistically significant and directional, not random noise.
GPT-2 achieves 95% accuracy identifying authorship through style patterns alone, but lacks the evaluative framework to explain why those stylistic choices carry meaning. Detection without interpretation remains cataloguing, not criticism.
StoryScope achieved 93.2% accuracy separating AI from human fiction using only discourse-level features like character agency and chronological structure, retaining 97% of performance while eliminating stylistic cues. These structural choices resist humanization because they require rewrites, not surface edits.
Discourse processing demands parallel recognition of linguistic segments, intentional structure, and attentional salience—not sequential processing. These three layers constrain each other during comprehension, and failures in any single layer disrupt overall understanding.
LLMs lose the social context that gives expert claims their force—reputation, track record, and standing—because they process only text, not the social world where expertise is built and evaluated.
Research identified four evaluation biases in LLM judges, with authority and beauty biases being semantics-agnostic and trivially exploitable through fake references and formatting—zero-shot attacks requiring no model access or optimization.
Interpretation Modeling research shows that disagreement on socially embedded sentences reflects valid differences in reader perspective, not annotation failure. Structured human disagreement in NLI benchmarks confirms that interpretation distributions carry meaningful information.
Analysis of debate corpora shows that political and religious ideology labels of voters outpredict linguistic features when modeling debate outcomes. Language effects observed without reader controls are confounded by audience composition correlated with debate topics.
Shanahan's 20-questions test shows LLMs maintain a superposition of consistent objects or characters and sample from that distribution at generation time. Regenerating the same response yields different outputs, each consistent with prior context, proving no fixed commitment exists.