Data Extraction Forms and AI Extraction

Build your extraction form

Click Data extraction in the sidebar (Pro, on a cloud-saved review). The panel has an Extraction grid tab where you enter data and an Edit form tab where you design the form. Five field types are available: Short text, Long text, Number, Choice (a dropdown with options you define), and Yes / No.

You do not have to start from scratch. Under Start from a template on the Edit form tab, pick one of the prebuilt forms and click Load template: Intervention randomized trial (registration, design, arms, outcomes, results, funding), Prevalence study, Diagnostic accuracy study (index test, reference standard, two-by-two counts), Qualitative study, or Blank to build your own. Loading a template copies its fields into your form, and every field stays editable, renamable, and removable afterwards.

Extract in the grid

The Extraction grid shows one row per included study and one column per field. Click into any cell to enter data; everything saves automatically. When you are done, click Export CSV for a spreadsheet-ready file of the whole dataset.

Dual extraction and the Consensus tab

On the Team plan, dual extraction lets two reviewers extract independently from the same studies without seeing each other's values; the grid tab is then labeled My extraction grid, because each reviewer sees only their own work. The owner or an arbiter opens the Consensus tab, compares the two extractions field by field, and records the final consensus value per record. Consensus is the authoritative dataset used for export and synthesis, and any record's consensus can be reopened later if new information appears.

AI extraction with evidence quotes

The AI extract button fills your form automatically from each study's full text, or from the abstract when no PDF is attached. Every filled field comes with an evidence quote, the sentence from the paper supporting the value, so verification is a read-and-compare rather than a re-extraction.

Nothing is applied automatically. You Accept or Reject each field individually, and an Accept all high confidence button applies just the fields the AI marked as high confidence in one click. AI extraction costs 2 credits per record, twice the screening rate, because it reads the entire full text. Records that cannot be extracted are refunded.

1. Attach PDFs in full-text screening first, so extraction reads the complete paper rather than the abstract.
2. Open Data extraction and finish your form on the Edit form tab.
3. Click AI extract, choose the scope, and confirm the credit cost.
4. Review each field against its evidence quote and click Accept or Reject.
5. Use Accept all high confidence for the remainder, then export with Export CSV.