The feedback a good senior colleague would give. Before the reviewers see it.
PaperPrep reads your draft against the reporting guideline for your study design, explains what it finds, and asks you the questions a careful reader would ask. It never writes your paper. It's free.
These are invented studies, written for this page so no one's real paper is on it.
Variables
¶3Maternal education was recorded as the highest level completed. Parity was the number of previous live births. Household wealth was derived from an asset index and divided into quintiles.
¶4Distance to the nearest facility offering ANC was categorised as under or over 5 km.
Where your paper goes
Your manuscript is analysed by OpenAI models hosted by Microsoft inside Azure, processed within Microsoft's EU Data Boundary (EU member states; Microsoft says this may include Norway and Switzerland). Microsoft states that prompts and completions are not available to OpenAI and are not used to train any model.
We send every request with storage switched off, and we never upload your file to the model provider's file store: the document is converted to text on our server and passed inline. Your original file is deleted as soon as we have read it, whether or not conversion succeeds; a storage rule deletes anything left within two days as a backstop.
While your paper is open, an encrypted copy of the text, the findings, and your answers is kept under a key that exists only for this manuscript. When you press delete, or when the paper closes, that key is deleted at once and the copy cannot be read by anyone, including us.
What remains is your record: dates, hashes, which checks ran, item statuses, and how you responded. No manuscript text and no results numbers are stored in it.
De-identified manuscripts only. You confirm it when you upload, and we scan for identifiers before any model call. If the scan finds them, the file is deleted and nothing is sent.
What it does
You tell it your research question, your design, and the one claim you want a reader to believe. Then you upload the draft.
- It finds the reporting guideline that fits your design: CONSORT for a randomised trial, STROBE for an observational study, COREQ for interviews, and the others in the EQUATOR family. Then it checks the paper against that guideline item by item and quotes the passage each finding rests on.
- It checks that the numbers in your abstract, text, and tables agree, and that each figure's caption matches the text that cites it. The arithmetic is computed, not guessed.
- It checks that the declarations a journal needs are there: ethics, consent, competing interests, data availability.
- It talks through whether your design supports your conclusion.
- You revise. It shows you what changed.
It works on a phone and over a slow connection, you can answer in any language, and a session picks up where you left it if the connection drops.
What it will not do
Write, rewrite, or polish your text.
Not a sentence, not a word. That keeps the paper in your voice, and it means your disclosure can say that PaperPrep wrote none of it. If you want language editing, that is a different service.
It does not grade your paper, and it does not certify anything. Every status is a suggestion you verify.
When the tool is wrong
Akinyi is writing up a household survey of antenatal care in western Kenya. Her design is cross-sectional, so her guideline is STROBE. On item 8, the first pass missed something. Here is the exchange, exactly as it appears in the app.
No score. A status only changes when the check itself agrees.
Why free, and who it's for
The models that can read a manuscript carefully sit behind subscriptions most departments don't buy. PaperPrep has no ads, sells no data, and needs no institutional licence. It's a research project, and the limits page, published with uploads, says who is paying for the model calls and what the monthly ceiling is.
Anyone with a draft health research paper and a design that has a reporting guideline.
- Students
- Postdocs
- Clinicians writing up a service evaluation
- Faculty who haven't had a second reader in months
Prefer your own chatbot?
The PaperPrep prompt pack
The reporting checklist for your design and instructions that tell a chatbot to go item by item, quote the evidence, and never rewrite. Free.
- Which free chatbot tiers we tested it in on a full manuscript, and when.
- How to turn off training and chat history in each one.
- What to put in your disclosure: name the chatbot as the tool, and treat any sentence it rewrites as AI text to disclose.
Pasting a draft into a chatbot sends it to that company under its terms.
Who built it
PaperPrep is built by Eric Green at Duke University. This is a research project: we are measuring whether the tool's checks are accurate and whether the teaching sticks. What we record is listed on the privacy page, and none of it is your text.
PaperPrep gives feedback on reporting. It does not assess whether a study was conducted as described, and neither its authors nor its operator vouch for any manuscript.
Hear when it opens
PaperPrep isn't taking uploads yet. Leave your email and we'll write once, when it opens. That's the only email you'll get, and we don't share the list.