Reviewer3Reviewer3

Research integrity checks for the AI era

Reviewer3 runs objective, verifiable checks on submitted manuscripts. It flags invalid references, unfinished drafts, and unsupported claims. A full panel of integrity and quality issues comes back in minutes.

Talk to usSee the benchmarks

Quality signals are changing

AI-generated work carries none of the conventional signs of bad research, like tortured phrases, plagiarism, or image manipulation. Reviewer3 looks for the signs that still hold. Every check below is configurable per journal.

Intake screening

Is it complete, original research?

Submission Screening

Flags unfinished or unsuitable submissions, including non-research documents, partial drafts, leftover AI artifacts, and hidden prompts.

Duplicate Papers

Catches when the same paper has already been published elsewhere, including semantically similar work such as AI rewrites, which evade plagiarism detectors.

AI-Generated Text

Detects AI-assisted or AI-generated text segments. Powered by Pangram.

Reference integrity

Every citation is checked against the literature. Integrity issues are flagged.

Hallucinated References

The citation does not resolve to any real source.

Retracted References

The cited work is real but has been retracted.

Irrelevant References

The source is real, but it does not support the claim it is cited for.

Reference Padding

References sit in the list but are never cited anywhere in the text.

Self-Citations

How much of the reference list points back at the paper's own authors.

Overcited Authors

How much of the reference list points at any one other author.

97.2% of fabricated references caught across a 476-citation benchmark, at 98.5% overall accuracy. See the benchmark

Evidence verification

AI can produce polished text with none of the leftovers above. Reviewer3 goes past the text and into the evidence.

Claim Review

Checks the evidence behind each claim, plus the methods, statistics, derivations, and internal consistency.

Fatal Flaws

Surfaces unsupported conclusions, impossible numbers, and circular reasoning.

Code Replication

Runs the code and compares its output with the paper.

90% of Reviewer3 findings could undermine a major claim, and it ranks first among AI review tools in agreement with human reviewers, measured across 145,000 human and AI review comments. See the benchmark

How it fits your workflow

Reviewer3 does not replace the system your editors already use.

Batch Pilot

Send a small batch of manuscripts and see what comes back. Nothing to install or build.

Contact us

API Access

Post a manuscript and get a structured report back as JSON.

API access

Submission System

We build the connection with your submission provider, so editors never leave the tool they already use.

Inquire about an integration

Built for Sensitive Research

Model Providers

Choose between open-weights models and frontier AI providers under zero data retention.

Never Used for Training

Neither Reviewer3 nor the model providers will use manuscripts to train or improve models.

Data Privacy

Manuscripts are encrypted in transit and at rest, and never made public.

Security details

Try Reviewer3

Run a pilot on a small batch of manuscripts.

Contact us