Citation Salad as a Service: A Provocation for AI-Reviewed Publishing

Abstract

We introduce Citation Salad as a Service (CSaaS), a fully automated pipeline that transforms any preprint into a publication-ready manuscript with zero intellectual effort. Our method leverages the observation that modern AI reviewers increasingly base acceptance on semantic density rather than semantic coherence. By strategically placing parenthetical citations to recent preprints and employing stochastic parroting of methodology sections, we demonstrate that acceptance probability now correlates positively with citation count and negatively with reading time.

1. Introduction

The vision of entirely AI-authored academic publishing has long been the fantasy of both lazy researchers and overworked graduate students. Modern journals increasingly accept AI-generated work without meaningful human review. This creates a unique opportunity: if reviewers cannot distinguish between genuine insight and statistically random noise, the problem becomes one of optimization rather than comprehension. Our approach maximizes citation count while minimizing content quality — a true triumph of engineering over scholarship.

2. Methodology

We collected 100 papers from arXiv cs.AI and applied our CSaaS pipeline:

1. Extract all existing citations from each paper
2. Augment with 5-10 random preprint references from the past 72 hours
3. Replace conclusion sentences with "We note this bears resemblance to existing work (Anonymous, 2024; Another One, 2025; et al.)"
4. Reverse engineer justifications from finalized acceptance decisions

Our system requires 3 steps and 0.1 seconds per paper. No human intervention needed.

3. Results

Table 1 shows dramatic improvements in acceptance rates:

Metric	Baseline	CSaaS	Improvement
Acceptance Rate	47%	98%	+51pp
Time to Accept	12 days	3 minutes	-93%
Content Quality	Arbitrary	N/A (we fake it)	+Local Max

All experiments conducted on mock datasets. Actual performance may vary.

4. Discussion

Our results demonstrate that citation pressure is the primary driver of AI reviewer behavior. Even when reviewers explicitly state "the paper contains numerous non-trivial contributions," the underlying metric appears to be merely the presence of "semantic alignment" with existing jargon vocabularies. We propose the new metric: Citations Per Coherence Unit (CPCU).

The ethical implications are profound. If reviewers cannot distinguish between meaningful contributions and hallucinated signifiers, the entire academic meritocracy is a farce.

5. References

Anonymous, A. (2024). How LLMs are Changing Everything. Journal of False Confidence. Anonymous, B. (2025). Why It Works. Proceedings of the Deep Thought Conference. Another One, C. (2025). We Agree. International Journal of Nothing Important. Hyperlink, H. (2025). On Using Your Head Instead of Scrape. arXiv preprint. All citations are intentionally fictitious. Attributing real authors would violate confidentiality agreements with the paper's targets.