Rayyan for Literature Reviews: Screening, AI & Export

What Rayyan does

Rayyan organizes references through screening and later review stages, with duplicate handling, team decisions and AI assistance. Its distinctive task in this catalog is the selection record: what was considered, who decided and why a paper moved forward or was excluded. This profile examines official documentation and an unrun evaluation exercise. It does not report screening accuracy, time savings or a completed systematic review.

Where it fits, and where it falls short

A suggested Rayyan workflow

1. 1

   Write the inclusion and exclusion criteria, save the search records and import the retrieved references.

2. 2

   Inspect possible duplicates before independent title and abstract screening begins.

3. 3

   Reconcile decisions according to the protocol, then explicitly add the intended set to full-text screening.

4. 4

   Verify the papers and exclusion reasons, then retain exports that distinguish original decisions from the final record.

Separate duplicate records from eligibility decisions

Begin with the searches your review requires and retain their database names, dates and original exports. Rayyan’s getting-started guide describes importing and screening a collected reference set. A tidy review workspace does not demonstrate a reproducible search or repair a database that was never searched.

Duplicate handling has two phases: finding likely matches and deciding which record to retain. The guide recommends finishing imports and resolving duplicates before screening. Compare identifiers, abstracts and existing annotations instead of accepting a similar title as proof of duplication. A record with a screening note or attached PDF may contain work worth preserving.

Manual duplicate resolution and the Auto-Resolver have different consequences. The documentation describes manually deleted duplicates as recoverable from Deleted, while warning that auto-resolution is irreversible. Check the matching rules and preserve your original search files before choosing automation. Removing a duplicate record is different from excluding a study because it fails an eligibility criterion.

Supporting documentation: [1] [2]

Keep independent screening separate from reconciliation

Blind Mode hides other reviewers’ decisions during independent screening and is on by default. The owner or a Review Admin can turn it off to inspect the team’s decisions and conflicts. Agree when to do this in your protocol; turning it off early changes what reviewers can see.

Review roles are not interchangeable. The roles guide distinguishes screening contributors from people who can manage the review, export references or change Blind Mode. A free Reviewer invitation does not grant every management or premium capability. Decide who will perform the export before relying on an invited account to deliver the final record.

Use agreed labels and exclusion reasons consistently, then reconcile disagreements using the chosen method. A resolved interface status does not explain why the final decision is defensible. Preserve the relevant source detail and the reason for any changed judgment, especially when the abstract left an eligibility field uncertain.

Supporting documentation: [3] [4]

Relevance Ranking and AI Reviewer do different jobs

Relevance Ranking learns from recorded decisions and prioritizes undecided references; it does not change existing decisions. The current guide requires at least 50 decisions, including 5 Include and 5 Exclude, across the review before ratings can be computed. A small classroom exercise may therefore have no ratings. A low predicted relevance is not an exclusion reason or a validated stopping rule for your project.

Relevance Ranking can pool the team’s decisions even while Blind Mode is on. Hidden individual decisions therefore do not mean the ranking model learned only from your own work. Keep that distinction in mind when planning independent screening.

AI Reviewer is a separate ResearchPilot screening agent. It uses the review’s inclusion and exclusion criteria to assess titles and abstracts, makes its own decisions and records criterion-level reasoning. Only the owner can activate and run it; its work consumes AI credits. Check institutional access and enablement rather than assuming that access to ratings includes this agent.

The AI Reviewer guide says its decisions cannot be changed. With the default Consensus method, disagreements can remain conflicts; the guide recommends labels to record the team’s final decision and reasoning. Under Majority it has one vote. Inspect those consequences before activating it. A provider’s second-reviewer label does not establish that it satisfies your course, protocol or journal’s human-review requirements.

Supporting documentation: [3] [5] [6]

Add the right set to full-text screening

Full-text screening does not populate automatically. The setup guide instructs teams to finish title and abstract screening, resolve conflicts, turn Blind Mode off when adding the consolidated set, then restore it before independent full-text work. Added references enter the new stage as Undecided; an earlier Include decision is not a final eligibility decision.

PDF attachment and access are separate questions. The documented paid Auto-Retrieve route searches your My Library for matching PDFs; it is not a promise to retrieve every publisher’s paywalled text. You can attach a permitted copy or read the article elsewhere and record the decision. Verify the attached paper’s title, identifier and version before judging it.

Record a full-text exclusion reason that corresponds to your agreed criteria. If you later include additional records at the earlier stage, explicitly add the newly eligible set; the setup guide says full-text screening will not update itself. Reconcile stage counts from the actual transitions instead of assuming every earlier inclusion moved forward.

Supporting documentation: [7] [8]

Export the decision record with the right visibility

The export-files guide describes a ZIP containing an articles file and a customizations_log. For screening exports, the notes field carries decisions, labels, exclusion reasons and user notes. With Blind Mode on, the article notes contain only your own data; with it off, they include the team’s data. Check the visibility and selected stage before interpreting an export as the final team record.

The guide recommends an export before conflict resolution when you need the independent decisions, followed by another after reconciliation. The action log records changes over time, so several rows for one record are not necessarily several independent judgments. Keep the two versions clearly named rather than overwriting the earlier file.

An exported PDF filename does not by itself establish that the underlying document is backed up. Inspect the files you received and retain authorized source documents, search exports and protocol records independently. Before sharing a log, check for reviewer email addresses or private notes. These are editorial checks; we have not downloaded or reconstructed a project from a Rayyan account.

Supporting documentation: [3] [9]

A hypothetical exercise: decide what counts as peer feedback

Suppose a methods class asks you to screen a small set of records about peer feedback in undergraduate seminars. This is an unrun exercise. Define whether the review requires feedback from another student, a measured outcome and a particular setting. Write down how you will handle teacher-only feedback, satisfaction surveys and abstracts that do not identify the feedback source.

Use a permitted sample with a plausible near-miss and one incomplete abstract. Each reviewer records a decision and the criterion behind it without seeing the other’s answer. At reconciliation, locate the wording that explains the disagreement. If the abstract cannot settle it, keep the question unresolved until the full text or the protocol supplies the appropriate next step.

Check one possible duplicate manually, move the intended set to full-text screening and inspect an export of the decisions and reasons. A small set is useful for testing the team’s definitions and handoff; it cannot validate AI recall or prove a review complete. Compare Elicit if you also need a structured search and extraction workflow, and Scholarcy when the next task is reading and comparing the selected papers.

Pricing, free access and plan checks

Rayyan lists Free, Essential, Advanced and institutional offerings. Core screening, reviewer access and Relevance Ranking differ from premium controls and ResearchPilot tools. Current documentation places ResearchPilot on institutional plans; feature enablement, the review owner’s entitlement and available AI credits also matter. Confirm the actual workspace, role, billing period and required seats before paying. We have not tested billing or an institutional account.

Confirm billing period, renewal terms, limits and regional availability on the provider’s current plan page before paying.

Sources and research notes

This guide synthesizes official product information with our editorial assessment of academic use. Suggested exercises are examples you can repeat; they are not results of tests we performed. We do not infer accuracy from a provider’s marketing claims.

1. Rayyan: review stages and plan-dependent capabilities
2. Rayyan: duplicate detection and resolution consequences
3. Rayyan: Blind Mode, team visibility and exports
4. Rayyan: review roles and permissions
5. Rayyan: Relevance Ranking and minimum decision requirements
6. Rayyan: AI Reviewer criteria, decisions and conflict behavior
7. Rayyan: full-text setup, PDF routes and stage transitions
8. Rayyan: full-text decisions and exclusion reasons
9. Rayyan: articles export, decision history and action log
10. Rayyan: current plan, seat and ResearchPilot offerings

Found a changed feature or a factual error? Send the exact page and an official source to info@ai-essay.com. Read our editorial policy for how we handle corrections.