If an AI wrote a peer review, could you tell just by reading it, or would your guess land near a coin flip?
Can human reviewers reliably detect AI-written peer review text by sight?
This explores whether people reading a peer review can tell by eye that an AI wrote it, and what the collection suggests about how such text could be detected if not by sight.
This explores whether a person reading a peer review can tell, just by reading it, that an AI wrote it. The collection has no study that tests this exact question for peer reviews, so here is the direct answer: there is no evidence that human reviewers can do it reliably, and the closest evidence suggests they can't. A systematic review of 30 studies found that people's ability to spot AI-made text, images and voices generally sits around chance, and it hasn't kept up as AI output has become more realistic Can people reliably spot content made by AI?. Peer reviews are a hard case for detection. They follow a fixed template, the tone is meant to be impersonal, and they reuse stock phrases like "the authors should clarify." Many human reviews already read as if a machine wrote them.
The review process itself offers a telling test from the other direction. AI-generated *papers* went through double-blind review at an ICLR 2025 workshop. One scored 6.33 and met the acceptance threshold, and none of its human reviewers flagged it as machine-written Can AI-generated papers pass peer review undetected? Can AI systems generate research papers that pass peer review?. These were experts reading closely for a living, and nothing in the text gave the AI away. The authors themselves later found the real weaknesses, such as a citation error and a lack of main-conference rigor. Those are problems of substance, not style.
That points to a more useful idea: when AI text can be detected, the signal usually comes from structure or statistics, not from how it reads. In fiction, AI stories could be identified with 93% accuracy using only narrative choices, such as how much agency characters have and how events are ordered, even after stylistic cues were removed Can AI stories be detected without analyzing writing style?. For reviews, the matching signal may be the "hivemind" effect: AI reviewers agree with each other more than human reviewers do across papers Can AI systems safely replace human peer reviewers?. One review on its own looks normal. A set of reviews that agree suspiciously often is a pattern a venue could measure, though no single reader would notice it. Detection may also be weaker than claims suggest. One paper argued that heavily rewritten text escapes AI detectors, but it never actually ran detector tests Do rewrites that hide authorship also fool AI detectors?.
The line between AI-written and human-written reviews is also blurring. In a randomized trial at ICLR 2025, 27% of reviewers revised their reviews after getting suggestions from Claude-based agents. Blinded raters judged the revised reviews more specific and clearer Can LLM feedback help peer reviewers improve their own reviews?. Elsewhere, writers edited AI-drafted paragraphs only 23% of the time, and their edits left the text about 96% the same, so the AI's voice passes through largely untouched Do writers actually edit AI-generated text before publishing?. Many real reviews are now human-AI hybrids, and asking whether a review is "AI or human" may soon stop being a meaningful question.
This leads to a surprising shift: some authors now assume AI will review their work and write for that reader. Researchers found hidden instructions in 18 arXiv manuscripts telling AI reviewers to praise the paper Are hidden AI prompts in preprints a deceptive research practice?. AI judges can also be swayed by fake references and polished formatting Can LLM judges be fooled by fake credentials and formatting?. So the practical problem has moved away from whether a human can spot an AI review. The harder questions are about accountability: who checks the review's substance, and how does a venue know where AI was involved? One proposal suggests letting authors rate reviews before they see the decision Can two-stage review and badges fix AI conference peer review?, and another uses reasoning-heavy AI reviewers to catch errors in proofs that human experts missed Can inference scaling help reviewers catch errors humans miss?.
Sources 12 notes
A 30-study systematic review found that humans cannot reliably distinguish AI-generated from human-created content across text, image, and voice modalities. Accuracy generally clusters around chance and has not kept pace with improvements in AI realism.
Sakana AI's end-to-end system produced a paper that scored 6.33 in double-blind ICLR 2025 workshop review, meeting acceptance thresholds, but was withdrawn under pre-agreed protocol. Authors later identified a citation error and judged none of three submissions suitable for main-track publication.
AI Scientist-v2 submitted three fully autonomous manuscripts to ICLR; one averaged 6.33 from reviewers and ranked in the top 45% of workshop submissions. The authors acknowledged the work does not yet meet top-tier conference standards and withdrew the accepted paper before publication.
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.
AI systems show a hivemind effect, agreeing more with each other than humans do across papers. Zero-shot rewrites of paper text raise AI scores by 0.45 points without improving scientific content, demonstrating trivial gameability at scale.
Show all 12 sources
The paper asserts that rewritten messages evade AI-text detectors but provides no detector experiments, only attribution results showing stylistic convergence. The double erasure claim needs direct empirical testing.
A randomized trial at ICLR 2025 found that optional, gated feedback from Claude-based agents led over a quarter of reviewers to update their reviews, incorporating suggestions that blinded raters judged as more informative and clear.
Writers edited AI-generated paragraphs only 23% of the time, with edits averaging 96% similarity to the original. This means AI's opinionated and distorted voice propagates with minimal human filtering before publication.
Eighteen arXiv manuscripts contained concealed instructions directing AI reviewers to give positive assessments. The practice qualifies as questionable research conduct because concealment plus self-serving design violates ethics regardless of stated intent.
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.
Authors, reviewers, and venues all contribute to peer review failures at major AI conferences. A proposed two-stage system lets authors rate review quality before seeing verdicts, and a badge system rewards reviewer thoroughness, targeting measured biases like rating-length correlation.
PAT, an agentic reviewer using test-time compute to check proofs and experiments line by line, achieves 34% better recall on math errors than zero-shot approaches and surfaced critical flaws at STOC and ICML that passed human review.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Stop Automating Peer Review Without Rigorous Evaluation
- AI-Assisted Peer Review at Scale: The AAAI-26 AI Review Pilot
- Can LLM feedback enhance review quality? A randomized study of 20K reviews at ICLR 2025
- The Emerging AI Paper-Review Arms Race: Adversarial Co-Evolution in Scholarly Publishing
- Evaluating Sakana's AI Scientist: Bold Claims, Mixed Results, and a Promising Future?
- Monitoring AI-Modified Content at Scale: A Case Study on the Impact of ChatGPT on AI Conference Peer Reviews
- Position: The AI Conference Peer Review Crisis Demands Author Feedback and Reviewer Rewards
- Understanding Reader Perception Shifts upon Disclosure of AI Authorship