Author guidelines

We publish studies of AI in healthcare that matter for patient care. Read the aims and scope first: methods papers with no clinical evaluation, and studies with no AI component, are out of scope.

Article types:

• Original research
• Brief report
• Systematic review
• Perspective

Model development and validation studies, trials, language model evaluations, negative results and replications are all original research. You do not need to pick a design category: we read the design from the paper and apply the matching reporting standard.

• One file. A Word file (.docx) works best. PDF is accepted too; if a PDF is accepted, we ask for the Word or LaTeX source before publication. Up to 30 MB.
• Any clear format. There is no template and no word limit. Use the usual structure for your study (introduction, methods, results, discussion) with an abstract.
• Figures can stay embedded in the file. High-resolution files are asked for only after acceptance.
• Tables should be real tables, not pictures, so readers can search them and the review can check their numbers.
• References in any consistent style, numbered or author-year. Include DOIs where you can: every reference is checked against Crossref and PubMed.
• Equations, footnotes and endnotes made with Word's own tools are read correctly.
• Supplementary files: up to 5 (.pdf, .docx, .xlsx, .csv, .tsv, .txt), each up to 30 MB. Their text is part of the review.

You do not need a separate title page or a blinded copy. The review system works on a blinded copy we make: author names, institutions, countries and acknowledgements are removed before the AI reads the paper.

The submission form asks for these. If they are in your manuscript, the form fills them in for you to check.

• Ethics: the approving committee and reference number, or why approval was not needed, and how consent was obtained or waived. You may mask identifying details for blind review (for example “Ethics Committee of XXX”); masked details never count against the paper.
• Data availability and code availability: where the data, code and model are, how to request them, or why they cannot be shared. “NA” is not accepted for studies that analyse data or train models.
• Trial registration for trials, registered before enrolment.
• Funding and competing interests, including ownership of or payments from companies whose AI products are studied.
• Use of generative AI in writing or analysis, if any. Authors stay responsible for every word, number and reference.

Missing or placeholder statements are flagged before you submit, so they do not hold up the decision.

We check your paper against the reporting guideline that fits its design: TRIPOD+AI for prediction models, STARD-AI for diagnostic accuracy, CONSORT-AI for randomised trials, DECIDE-AI for early live evaluation and TRIPOD-LLM for language model studies. You do not need to upload a checklist: the report lists the items that are missing or partly reported, most important first.

• The corresponding author signs in with ORCID and submits. Name and institution come from the ORCID record.
• Give an e-mail address for every author. Each co-author is asked to confirm their authorship and declare competing interests; all must confirm before the paper is published.
• Authorship follows the ICMJE criteria. AI tools cannot be authors.

1. AI review, in minutes. You receive a reviewer-style report with quoted evidence for every finding and one of five decisions: acceptance candidate, major revision, missing statements, rejection with resubmission allowed, or rejection.
2. Revisions. Upload a revised file with a short response to the report. Revisions are free, up to 3 per paper, and each is checked against the previous report point by point.
3. Expert confirmation, 72-hour target. Two independent experts read every acceptance candidate with the AI report and confirm or overturn it. If either raises a concern, an editor decides.
4. Appeals. You can appeal any decision from your submission page. An editor who was not involved answers within 7 days.

A rejection based on references or scope is checked by an editor before you see it.

Accepted papers are published within 7 days. On your submission page you:

• give any details you masked for blind review, such as the ethics committee and approval number;
• upload high-resolution figures where needed: at least 1800 pixels on the long side, or vector PDF or SVG. Each image in your file is checked and the ones to replace are marked;
• may upload a final Word file (required if you submitted a PDF);
• agree to publish under CC BY 4.0. You keep copyright.

The editor-in-chief checks these and publishes the article, open access, with its review: the AI assessment and the experts' comments, unsigned. See open review.

Peer review is free. Each author can submit 3 new manuscripts a year. There is no article processing charge for manuscripts submitted in our first three months; see Fees for what follows.

Manuscripts are processed by contracted AI services under agreements that forbid using them to train models. Reviews of papers that are not accepted are never published. More in our editorial policies.