Reviewer Lens
An academic submission audit tool that catches fabricated citations, hallucinated author lists, and LLM-written passages before reviewers waste hours on slop papers.
Academic peer reviewers, conference area chairs, and journal editors handling ML and scientific submissions
- One-click bibliography check that cross-references every citation against Crossref, Semantic Scholar, and arXiv, flagging dead DOIs, fake authors, and fabricated venues
- LLM-slop detector tuned for academic writing (the 'It's not X, it's Y' tells, dense jargon, hedging spirals, embedded LLM notes)
- Reviewer-friendly PDF report exporting flagged issues with reviewer-ready language to paste into a decision form
- Pre-submission helper for authors that returns a reviewer-eye-view so papers get cleaned up before they enter the queue
A 2026 audit found 68% of ML conference submissions contained fabricated citations, hallucinated authors, or unmistakable LLM slop, and 85% of those hallucinations survived peer review; reviewers are explicitly asking for tooling to triage the mess.
68% figure is real: Robinson & Corley reviewed 22 submissions at NeurIPS/WACV/ECCV workshops, 15 (68%) had fabricated citations/hallucinated authors/slop; 85.3% persistence figure traces to Zhao et al. preprint audit; 21% of ICLR 2026 reviews fully AI-generated (15,899 reviews). Reviewers actively venting and asking for tooling.Q&A from the slop trenches - GeoSpatial ML ↗Phantom References: Hallucinated Citations That Survive Peer Review ↗
Crowded for authors (Citely, AiCitationChecker, Sourcely all sell citation verification) and Pangram sells AI-writing detection; Robinson & Corley already open-sourced a bib-audit skill. Reviewer-specific workflow tool is less served but free/open alternatives exist.Citely AI Citation Checker ↗AiCitationChecker ↗
Target users (peer reviewers, area chairs) are unpaid volunteers; conferences (NeurIPS, ICLR, WACV) are non-profits with thin budgets. Sales would have to be B2B to publishers/societies, a long, gated sales cycle, or freemium with low conversion — a structurally tough monetization story.Q&A from the slop trenches - GeoSpatial ML ↗
LLM-generated submission volume is accelerating (42% post-ChatGPT submission surge at Organization Science) and detection-evasion is an arms race; academic fraud and citation sloppiness are perennial. Problem likely to persist 5+ years even as underlying LLMs improve.Q&A from the slop trenches - GeoSpatial ML ↗
Citation existence/author verification is straightforward via CrossRef, OpenAlex, and Semantic Scholar APIs (Robinson & Corley built a working skill in one summer). LLM-text detection is the harder, less reliable layer; full audit workflow is buildable by one engineer but not trivial.Q&A from the slop trenches - GeoSpatial ML ↗