The 'Missing Middle' Editors Are Finding in AI-Written Reviews

Journal editors in applied linguistics and writing studies say they are being flooded with machine-written submissions. At recent academic conferences, editors reported that submission rates have doubled or even tripled in the past year. The extra volume, they argue, pulls attention away from papers with authentic data and real human insight, because every submission still has to be read and discussed.

Many AI-generated papers reveal a distinctive flaw: a literature review that cites three or four foundational works from the 1980s and early 1990s, then jumps to fifteen to twenty recent papers from 2023 to 2026. Almost nothing from the mid-1990s through the early 2020s appears. The authors call this the missing middle. It happens because generative AI cannot get behind publishers' paywalls, so it combines older seminal papers that have been uploaded online with recent open-access and preprint articles, while skipping the deep peer-reviewed literature in between.

The problem then travels downstream. In the Discussion section, experienced scholars return to works cited earlier and place their findings in conversation with them. AI-generated papers instead repeat the findings, list obvious limitations such as sample size, and move on to future research. The result is a polished but hollow survey that asserts a recent scholarly gap without demonstrating how the field progressed from its founding studies to the present.

The journal editors are not proposing to police every use of AI. They want editors to desk-reject manuscripts with obviously missing-middle reviews, ask reviewers to read with this pattern in mind, and insist that literature reviewing remain a human process of connecting past and present work.

Advertisement

Why Missing-Middle Reviews Rewrite the Rules of Academic Publishing

How the 'Missing Middle' Actually Forms

The mechanism is structural, not stylistic. Generative AI models retrieve older canonical articles that have been widely posted online and recent open-access or preprint material. Because most high-quality peer-reviewed research from the mid-1990s to the early 2020s sits behind subscription firewalls, the AI's reference base has a blind spot. The output therefore looks comprehensive but skips the empirical work that tested, refined, and often corrected the foundational studies.

What This Means for Journals and Editors

Editors face a triage problem. Submission growth is not neutral: it consumes reviewer time and crowds out papers with genuine datasets. Missing-middle reviews also undermine the credentialing function of a literature review, which is meant to show that authors understand their field deeply enough to ask the right research questions. When that signal disappears, editors lose a fast, fair way to separate substantive work from generated text.

The Career Penalty for Early and Mid-Career Researchers

AI-generated reviews tend to reamplify a few already prominent sources. Researchers whose work does not use keywords recognized by the model are less likely to be cited, which hurts early and mid-career scholars who depend on citations for advancement. The same dynamic can elevate predatory journals that publish open-access content easily accessible to AI systems, even when their editorial oversight is weak.

The Limits of Detection

The authors are explicit that trying to police researchers' use of AI is likely hopeless. Instead, they see the missing middle as a reason to change editorial and review practices. The goal is not to catch every machine text, but to protect the process of scholarly knowledge building by insisting that authors demonstrate real engagement with the literature.

What Editors, Reviewers and Departments Can Do Now

  • For journal editors: treat an unexplained gap between foundational works and recent 2023–2026 citations as a desk-reject criterion, and ask for page-specific references or direct quotations rather than citation lists alone.
  • For peer reviewers: check whether the Discussion section returns to the literature cited in the review. If it only repeats findings and lists limitations, flag the submission for revision or rejection.
  • For department and graduate program heads: require literature reviews to be developed through institutional library databases and evaluated for their narrative connection between foundational studies and mid-period research.
  • For early-career researchers: use paid database access to trace the post-1990s empirical record; generative AI summaries may skip your subfield, so verify citations before submitting.