How AI Screening Works

When you run AI screening, Zynthia makes a single AI call per paper that returns everything you need to make a fast, well-informed decision, not a separate title/abstract pass followed by a full-text pass, one combined analysis per paper.

What comes back per paper

• An include/exclude recommendation with a plain-language reason.
• A Relativity Index (RI) score across four dimensions, see the RI scoring guide for the full breakdown.
• Evidence quotes, the exact phrases in the abstract that support the recommendation, highlighted so you can verify it at a glance instead of re-reading the whole abstract.
• Analytics metadata: methodology type, theory tags, and industry context, used to power your analytics dashboard and QDA Map without any extra manual tagging.

How exclusions get attributed to your criteria

If a paper is excluded, Zynthia tries to attribute the exclusion to the specific inclusion or exclusion criterion that drove it, rather than a generic "out of scope" reason. This attribution only exists for criteria evaluated after this feature shipped; criteria checked earlier don't retroactively grow it, that requires screening the affected papers again.

Reviewer, not replacement

Every AI decision is a suggestion. You'll see it clearly labeled and separate from your own decision on the screening page, and your final call, include, exclude, or uncertain, is what counts toward your PRISMA numbers and audit trail. The manual screening workflow covers how to record it, including bulk-accepting the AI's recommendations when you trust them for a batch of papers.

Re-running AI screening

If you change your criteria, or the prompt/scoring logic has since improved, you can re-run AI screening on papers that don't yet have a confirmed human decision. Papers you've already confirmed by hand are never silently overwritten, only unconfirmed papers get refreshed. Re-running consumes your plan's AI usage allowance per paper processed, so it's worth doing deliberately rather than repeatedly.