academic-deai

academic-deai

Your paper should sound like you wrote it, not like a model completed a writing task.

Before / After · What It Catches · Install · Usage

---

A Claude Code and Codex skill that evaluates economics manuscripts on two axes: AI contamination and journal quality (benchmarked against AER, QJE, Econometrica, JPE, REStud conventions). It scores every sentence, flags the worst offenders, and tells you exactly what to delete, swap, or split.

Built for economics papers. Useful for any social-science manuscript.

Before / After

Before "Moreover, our findings underscore that trade shocks play a pivotal role in exacerbating food insecurity -- not only through direct price transmission channels but also by undermining the resilience of vulnerable households. These results have important implications for policy design."	After "Trade shocks worsened food insecurity through higher prices for imported staples and by reducing the purchasing power of low-income households."
Before "This paper contributes to a more nuanced understanding of how climate variability shapes agricultural outcomes. Notably, the heterogeneous effects across food groups illuminate the multifaceted nature of dietary vulnerability in developing economies."	After "Climate variability affects agricultural outcomes unevenly across food groups: cereals and perishable crops show larger declines than legumes or dairy in our sample of developing economies."
Before "Interestingly, the coefficient on tariff reduction is positive and significant, lending support to the hypothesis that trade liberalization fosters dietary diversity through enhanced market access."	After "Tariff reductions are associated with higher dietary diversity, consistent with greater access to imported food varieties in the treated regions."

AI traces removed. Journal quality achieved.

---

What It Catches

Two-axis evaluation

Every sentence is scored on two independent axes:

Axis	What it measures	Scale
AI score	Templated rhetoric, inflated phrasing, mechanical balance	0-3
JQ score	Adherence to top-5 economics journal conventions	0-3

A sentence needs revision if either score >= 2. A sentence can be low on AI markers and still fail journal quality (too flat, too memo-like, missing table references).

AI lexical markers

Flags verbs like underscore, foster, leverage, illuminate, dovetail, undergird and adjectives like nuanced, multifaceted, overarching, pivotal that are high-frequency AI fingerprints.

Template sentences

Detects canned patterns: plays a crucial role, has important implications for, fills a gap in the literature, warrants further investigation, a growing body of literature suggests.

Density, not just occurrence

A single Moreover is fine. Three in one section is a pattern. The skill uses per-paragraph and per-1000-word thresholds to avoid false positives.

Journal-quality checks (top-5 standard)

Evaluates against published conventions of AER, QJE, Econometrica, JPE, REStud:

• Prose register: first-person plural "we", active voice, 15-25 word sentences, calibrated hedging
• Results reporting: must include table/column references, coefficient + standard error, magnitude interpretation
• Section structure: canonical introduction formula (hook, question, results preview, mechanism, contribution, roadmap)
• Contribution framing: enumerated, anchored to specific prior papers, specific about nature
• Formatting: standard errors in parentheses, leading zeros, "percent" vs "percentage points"
• Sentence variety: flags uniform sentence/paragraph length as both AI marker and quality failure