Paper Review · Data handling

Is an AI review safe for an unpublished manuscript?

This page sets out where your manuscript goes during a Paper Review, who can read it, and when it is deleted. It also lists what we cannot promise, so you can decide before you upload anything.

Run the free reference check Start a review, from $9

During the review

Who reads the manuscript

The upload goes straight from your browser to our server on Modal, not to the website host. From there, three parties see some part of it.

Our server on Modal. A Word file is converted to PDF, and the PDF is read in memory. Pages are rendered to images from a private temporary file that is deleted as soon as the rendering finishes.
Anthropic. The text and page images are sent to Anthropic's Claude API, which runs the seven reviewers. Under Anthropic's commercial terms, API inputs are kept for up to 30 days for abuse monitoring and are not used to train models.
CrossRef, OpenAlex and Semantic Scholar. They receive the titles and DOIs of your references, so each one can be looked up. Nothing from your own text goes to them.
Nobody else. No person at Purplelink reads the manuscript as part of the review, and it is not used to train anything.

The other services

What payment, email and hosting see

None of these three ever receives the manuscript.

Stripe takes the payment on its own checkout page. It sees your email address and card. Purplelink never sees the card.
Resend sends the email that says your review is ready, with the report attached as a PDF. It sees your address and that email. The manuscript and the annotated PDF are never emailed.
Netlify hosts this site and the payment notice from Stripe. It sees the Stripe session ID and the email address passing through that notice, and IP addresses like any web host.

Afterwards

When it is deleted, and what we keep

The manuscript, the online report and the annotated PDF are deleted within about a day after you first open the result, or within 48 hours if you never open it. A cleanup job runs once a day to enforce this.
The emailed report stays in your mailbox. Our copy of the email at Resend follows Resend's own retention.
A cost record per review is kept permanently: the product, whether the job finished, the price, token counts, cost and model names, with the time. It contains no manuscript text, no title and no email address. We use it to check that prices cover what the reviews cost.

Limits

What we cannot promise

No on-premises version. The review runs on our server and Anthropic's API. There is no version you can run inside your institution.
The 30 days at Anthropic are set by Anthropic. Deleting our copy does not shorten them.
Contracts on request, not by default. A data processing addendum, the subprocessor list and a completed HECVAT Lite are available from ben@purplelink.llc. The security page and the page for labs have the detail.
An AI review can be wrong. It does not replace your co-authors or the journal's reviewers.

Why not paste it into a chatbot?

A chat assistant reads your PDF once and answers in one pass. Paper Review looks every reference up in three scholarly databases, recomputes your statistics without a language model, and has seven separate reviewers read the paper before one pass reconciles what they found.

One chat prompt

One model, one pass

Your PDF
One model
One answer

References are not looked up one by one, so a wrong or invented one can pass.
Retractions are caught only if the model happens to know them.
p-values and means are not recomputed.
One point of view, written in one go.
No marked-up PDF.
Whether the manuscript is kept or used for training depends on your account settings.

Paper Review

Four stages, many checks, run for every paper

Read the paper apart
The PDF is split into text, figures, tables and the reference list.
2 Check it, in parallel
- Lookups Every reference
  - CrossRef
  - OpenAlex
  - Semantic Scholar
  - Each reference matched to a real record
  - Retractions, withdrawals and expressions of concern flagged
  - Key claims checked against the cited source's abstract
- Rule-based, no AI Numbers and structure
  - p-values recomputed from t, F, chi-square, r and z (statcheck)
  - Reported means tested for consistency (GRIM)
  - In-text citations matched to the reference list
  - Figure and table numbering checked
- Vision Every page and figure
  - Pages rendered and read as images
  - Figures compared with what the text claims
  - Missing error bars, broken axes, color-only encoding
- Panel Seven reviewers, separately
  - Methodology
  - Statistics
  - Data integrity
  - Editor
  - Equations
  - Literature
  - Numerical realism
Reconcile
Findings are merged, and issues raised by more than one reviewer rank first. A coherence check compares the abstract, hypotheses and results with each other.
Deliver
A report with a fix-first checklist and an annotated PDF, by email. Your files are deleted within about a day.

The reviewers and the final pass use Claude Opus through Anthropic's API, which does not train on your manuscript.

Compared with a chat assistant

Pasting a draft into a chat app

In a consumer chat app, how long your draft is kept, and whether it can be used for training, depend on the plan you are on and the settings in your account. The same draft can be handled differently on two accounts. Paper Review uses Anthropic's commercial API terms for every review, so the rules above are the same for everyone, and you do not need to change a setting first.

Before you upload

A short checklist

These apply to any AI tool, this one included.

Your co-authors agree. The manuscript is theirs too. Ask before you send it to any outside service.
The data allows it. If the paper quotes restricted data, patient details or anything under a data use agreement, check that the agreement permits sending it to a third-party processor. If not, remove those passages first.
The journal and funder allow it. Read their AI policies. Some ask you to disclose AI assistance in the cover letter or acknowledgments.
It is your own paper. Never upload a manuscript you were sent to review. Journals forbid sharing those with AI tools, and so do we.

Questions

Does Anthropic train on my manuscript?

No. Paper Review uses Anthropic's API under its commercial terms, which keep inputs for up to 30 days for abuse monitoring and do not use them to train models.

Can anyone at Purplelink read my paper?

The review runs without a person reading the manuscript. It sits in our storage only until the cleanup deletes it, within about a day after you first open the result, or within 48 hours if you never do.

Does the free reference check send my paper anywhere?

It reads the PDF on our server, sends the reference titles and DOIs to CrossRef, and discards the file when it answers. Nothing goes to Anthropic, nothing is stored, and no email address is asked for.

Can my institution sign a data processing agreement?

Yes. A data processing addendum, the subprocessor list and a completed HECVAT Lite or your own vendor questionnaire are available on request from ben@purplelink.llc. There is no on-premises version.

Start with the free check.

It reads your reference list, stores nothing and needs no account. The full review is from $9, usually back in 4 to 8 minutes.

Run the free check Start a review