Questions we get

No, and it was not built to. Julie does the volume work; the decisions are made by a named person. Six of those decisions are set out on the How it works page, from adjudicating the screening to recording what the panel agreed. The ones that frame the review stay with you throughout: the questions, approval of the criteria and search strategies, running the searches, and reviewing the manuscript.

Julie links extracted information to its sources and brings disagreements and uncertain cases to you. Screening quality is reviewed before the full run. Your checks and scientific judgement remain essential.

That is what it is built for: the PRISMA 2020 checklist has been compared with the chain item by item, the methods report comes from the run's own record, and every paid stage records which model ran, at which stage and under whose approval, so the AI use statement is built from that record. What that does not mean is that the draft goes out unread: the scientific judgement stays with you, where it belongs.

No. Julie is a research instrument for evidence synthesis. She has nothing to say at the bedside, gives no treatment advice, and is not an alternative to the point-of-care references clinicians use during the working day.

Julie helps develop search strategies for PubMed, Embase and CENTRAL in the Cochrane Library, starting from your research questions and the scope you confirm. She prepares the drafts; you review and approve the final strategies, run the searches and bring back the results. The full strategies and search details still belong in your review's reporting.

No. You can start from a Julie systematic review, or from a set of studies you already have. Either way, you sign the analysis plan before anything is pooled, and the values are extracted from each study by two readers, with disagreements coming to you.

Established meta-analysis software, not the AI model. The model reads the studies and drafts the text; the pooled estimates, intervals and heterogeneity statistics are calculated from the extracted values, so the same inputs give the same results and a statistician can recompute them.

Pairwise comparisons of two-arm studies: risk ratios, odds ratios and risk differences for events, mean and standardised mean differences for continuous outcomes, and reported effect estimates such as hazard ratios. Random-effects and common-effect models, methods suited to rare events, and sensitivity analyses come with it. If your question needs a different kind of synthesis, tell us what it is.

Then they are not. Whether studies belong together is your decision, and so is describing a study in the text instead of pooling it. Your reason is recorded under your name and appears in the Methods and the Discussion.

Because a nine stage process is easier to talk about with a name than with an abbreviation. Julie does not play a character, though: she does not think and does not decide. Her name never goes into a field that records who decided something. Only a person's name belongs there.