Referee 1 was nice. Referee 2 was... Referee 2. Meet Referee 3.
Referee 3
The one who actually wants your paper to succeed.
Up to three AI referees read your paper, search the literature, and give you specific fixes — not vague complaints. With public dataset suggestions. Under an hour. $29.95 $0 with an invite code.
Have an invite code already? Submit your paper here →
Spot the difference.
Same paper. Very different feedback.
Referee 2 · 4 months later
“The identification strategy is unconvincing and the paper lacks novelty. I recommend rejection.”
247 words. Zero citations. Wrote this during a faculty meeting.
Referee 3 · 47 minutes later
“The parallel trends assumption (Figure 3) shows a pre-trend in Q2 2019. I suggest a placebo test using non-treated counties, following Autor (2003). Also, the BLS Quarterly Census of Employment would strengthen your labor market outcomes — it's publicly available and covers your sample period.”
3,200 words. 14 citations. Suggests a free public dataset. Actually read the paper.
The real question.
“But I can just paste it into ChatGPT...”
Sure. Try it. Then come back.
We did the experiment so you don't have to. Here's what actually comes out, row by row.
| 🤖 Paste into ChatGPT | 🧠 Upload to Referee3 | |
|---|---|---|
| How many voices | One generic model. | Three independent referees (Claude + Gemini + GPT) in parallel — useful disagreement. |
| Knows what kind of paper it is | One-size-fits-all prompt. | Classifies reduced‑form / structural / applied‑ML / theory, then routes to a paper‑type‑specific review. |
| Theory review criteria | Whatever the model improvises. | Built from the methodology writings of ~100 Nobel laureates + Clark medalists (Friedman, Lucas, Romer, Hayek, Sen, Wilson, Milgrom…). |
| Literature | Fake citations. “Smith (2019)” never existed. | Real, web‑grounded search. Fabricated citations explicitly forbidden. |
| Data gaps | Shrug. | Data Referee matches your concerns to a curated catalogue of public and private datasets — we tell you exactly which data source closes each identification gap. |
| Concern format | Paragraphs of vague prose. | Issue / Why it matters / Suggested fix / Difficulty — for every concern. |
| Synthesis across reviewers | You get one answer. No disagreement to reconcile. | Cross‑referee synthesis ranks concerns, flags disagreement, 3‑step action plan. |
| Delivery | Closes tab. Forgets it ever happened. | Emailed report + persistent review page + two revision rounds included. |
ChatGPT is a brilliant intern you hired on a Friday afternoon to read your paper.
Referee3 is the tenured committee you wish you had before submission.
Simpler than submitting to a journal.
(And about 11 months faster.)
Upload your PDF
Pick your field. We figure out if you wrote a DID, a structural model, or a theory paper — and tailor the review accordingly.
Up to three referees go to work
Powered by multiple state-of-the-art AI models. Each independently searches the literature, checks your identification, evaluates your claims, and writes a full report.
Get actionable feedback
A synthesis ranks every concern by importance, suggests public datasets to fill data gaps, and gives you a 3-step action plan.
Everything Referee 2 forgot to include.
We don't just find problems. We help you fix them.
🎯 Suggested fixes for every concern
Not just "needs more robustness." We tell you which test, which specification, which paper to cite.
📚 Literature-grounded critiques
Every concern references real papers. We search Google Scholar, SSRN, and NBER so you don't have to.
📊 Public dataset suggestions
Our Data Referee scans 4,700+ public datasets and matches them to your data gaps. Free data, better paper.
🔍 Claims vs. evidence audit
We check if your tables actually say what your text claims. No more "the results speak for themselves."
🧩 Mechanism & heterogeneity
"Who does this work for?" and "Why does it happen?" — the two questions every AER referee asks.
🔄 Two free revision rounds
Submit your revised draft. We tell you which concerns you nailed and which still need work.
We speak your language.
Whether you're running a DID or proving a theorem, we know what to look for.
Currently optimized for empirical papers. Methods and theory support is in beta.
Empirical
- ●Reduced form & causal inference
- ●Structural estimation
- ●Applied machine learning
We check your identification, parallel trends, mechanisms, and whether your effects are economically meaningful.
Methods
- ●With theoretical guarantees
- ●Algorithmic & computational
We evaluate your proofs, convergence rates, simulation evidence, and whether practitioners would actually use this.
Theory
- ●Economics & game theory
- ●Statistics & optimization
- ●Mathematics
We verify proof correctness, assess generality, and ask whether the model generates testable predictions.
Calibrated to the journals you're targeting.
Tell us where you're submitting. We adjust our standards.
Questions you're probably asking.
Is this better than asking my advisor?
It's complementary. We catch the things your advisor is too polite to mention, and we do it before the journal's Referee 2 does. Think of it as a pre-flight checklist.
Do you actually search the literature?
Yes. Each referee searches Google Scholar, SSRN, and NBER for papers related to your claims. They cite real papers — we explicitly prohibit hallucinated citations.
What if my paper is in a niche field?
Select your target journals when you upload. We calibrate the review to those journals' standards, whether it's AER or a field journal.
How is this different from ChatGPT?
ChatGPT gives you generic feedback. We run a multi-stage pipeline using multiple state-of-the-art AI models: classify your paper type, search relevant literature, apply field-specific review criteria, cross-reference your claims against your tables, and suggest public datasets. It's not a chat — it's a system.
Less than a seminar lunch.
$29.95$0
Free with an invite code. Up to three referees. Data suggestions. Two revision rounds.
No subscription. No hidden fees. Cheaper than the therapy you'll need after Referee 2.
Request access→