● AI Paper Referee

The honest colleague
who's read everything

Skrutini analyzes scientific papers like a senior peer reviewer — scoring every claim, auditing the methods, catching overclaims, and telling you what the authors didn't.

Add to Chrome — it's free See how it works →

10 Free papers/mo

Any Journal, anywhere

30s Analysis time

Referee

Claims

Gaps

Methods

Memory

mRNA-LNP vaccine encoding Ag85B, Hsp70 & ESAT-6 elicits protective immunity against M. tuberculosis

npj Vaccines · 2024 · DOI: 10.1038/s41541...

+ 10 figure(s) analyzed

Read Yes

Cite With caution

Trust Preliminary

Strong murine Th1/CD8 data, well-executed ELISpot. The "protective immunity" title claim is unsupported — no challenge study was performed. Cite for the antigen+adjuvant design; don't inherit the translational confidence.

The title promises protection data but delivers only immunogenicity — a critical distinction reviewers will flag immediately.

How it works

Open a paper. Get a verdict.

Skrutini activates automatically when you open any scientific paper — on PubMed, arXiv, Nature, Cell, or anywhere else.

🔬

Open any paper

Skrutini detects scientific papers automatically — no uploading, no copy-pasting. Works on PDFs, HTML articles, and any journal worldwide.

⚖️

Get an honest verdict

Every claim is scored for evidence quality. Methods are audited for statistical red flags. Knowledge gaps are ranked by fundability. Figures are read visually.

💬

Ask the referee anything

Challenge the verdict. Request a peer review critique. Get a grant citation. Ask why a claim scored 28. The referee knows the paper cold.

What makes it different

Not a summarizer.
A referee.

Every existing tool tells you what a paper says. Skrutini tells you whether to believe it.

🎯

Claim trust scores

Every major claim scored 0–100 with specific reasoning. Cites the actual n values, p-values, and effect sizes. Catches overclaiming language like "demonstrates" when the data only "suggests."

honest referee

🔭

Figure vision

Actually reads the graphs. Checks whether error bars match the claimed statistics. Evaluates histology images. Flags figure manipulation red flags. No other tool does this.

multimodal AI

🧠

Memory & knowledge graph

Remembers every paper you've read. Surfaces contradictions with past papers. Detects patterns across your reading. After 20 papers, Skrutini knows your field better than you do.

personal memory

🔍

Methods audit

Flags underpowered samples, inappropriate statistics, missing controls, and post-hoc analysis dressed as confirmatory. Compares what Methods describes to what Results actually reports.

reproducibility

Real example

What Skrutini actually catches

A real TB vaccine paper. Skrutini scored the core claims — and caught the one that mattered most.

Paper analyzed

mRNA-LNP Vaccine Encoding Ag85B, Hsp70, and ESAT-6 Elicits Protective Immunity Against M. tuberculosis

npj Vaccines · 2024 · Score: 48/100

Score: 91/100 — STRONG

Ag85B+ESAT-6 construct elicits robust IFN-γ Th1 responses in C57BL/6 mice

Well-powered ELISpot (n=8), replicated in 2 independent experiments, appropriate statistics.

Score: 15/100 — CRITICAL OVERCLAIM

"Results demonstrate strong translational potential for human TB prevention"

No NHP data. No aerosol challenge. No safety/tox profiling. Title claims "protective immunity" — no challenge study was performed.

Live context

The title promises protection data but delivers only immunogenicity — this is either a major oversight or misleading framing that will confuse readers about what was actually tested.

The verdict a human reviewer would take 2 hours to write

Skrutini delivered it in 30 seconds — with specific evidence, scored by claim, with a gap analysis that identified 4 fundable research directions.

⏱

30 seconds vs 2 hours

Time to get a structured expert-level analysis of a full paper

🎯

Specific, not vague

Every verdict cites actual numbers, section names, and quoted language from the paper

🔄

Gets smarter over time

After 10 papers, Skrutini knows your field. After 50, it knows your lab.

Pricing

Start free. Pay when it's indispensable.

Every plan includes the full analysis engine. No per-query fees. No usage anxiety.

Free

10 papers/month

Referee verdict + score
Claim trust scores
Knowledge gaps
Methods audit
Figure analysis
Chat with referee
Memory graph

Get started free

Things scientists ask us

Does it work on paywalled papers?

If you have institutional access and can see the paper, Skrutini can analyze it. For papers where only the abstract is visible, use the right-click "Analyze selection" feature to analyze the abstract text directly.

How accurate are the claim scores?

Skrutini scores claims based on the evidence presented in the paper — it doesn't have independent knowledge of whether claims are true. It evaluates whether the data shown supports the claim made. Like a reviewer, it can be wrong, and you should always read critically.

Is my data private?

Paper text is sent to our API for analysis and is not stored beyond your session. Your memory graph (which papers you've read and their verdicts) is stored securely in your personal account and is never shared or used to train models.

What journals does it work on?

Any journal, anywhere. Skrutini uses semantic detection to identify scientific papers rather than a hardcoded list of journal sites. If you can read it in Chrome, Skrutini can analyze it.

Can it analyze preprints?

Yes — arXiv, bioRxiv, medRxiv, and SSRN all work automatically. Preprints are where the most exciting science appears first, and Skrutini is particularly valuable there since preprints haven't been peer reviewed.

How is this different from ChatGPT?

ChatGPT summarizes. Skrutini evaluates. It scores specific claims against the evidence presented, audits statistical methods, reads figures visually, builds a personal memory of your reading history, and gives adversarial verdicts — not polite summaries.

The honest colleaguewho's read everything