Pub-AI logoPub-AI: AI in Science
Literature Review & Evidence Synthesis⏱ Setup: 1–2 hours. Screening itself: depends on record volume and inclusion criteria.

Title and Abstract Screening for a Systematic Review Using Rayyan

A step-by-step guide to importing search results from Scopus or PubMed into Rayyan, configuring blind dual-review, using AI suggestions to prioritize screening, and resolving conflicts — producing a PRISMA-compliant screened record set.

AudienceResearchers conducting a systematic or scoping review who need to screen large numbers of records efficiently with one or more reviewers
Tools coveredRayyan, Scopus, Elicit, Zotero
Published July 2026

Where this fits in the systematic review workflow

A systematic review follows a documented pipeline. This tutorial covers the title/abstract screening stage:

1. Formulate question (PICO/PECO/SPIDER)
2. Register protocol (PROSPERO)
3. Database search → export records        ← prerequisite
4. Deduplication
5. Title/abstract screening                ← this tutorial (Rayyan)
6. Full-text eligibility assessment
7. Data extraction                         ← see Elicit tutorial
8. Risk of bias assessment
9. Synthesis and reporting (PRISMA)

Screening is where large initial record sets (often 2,000–20,000 records) are reduced to a manageable set of potentially eligible studies for full-text review. For PRISMA compliance, every decision needs to be documented with an exclusion reason.

Step 1: Export your search results

Run your searches in Scopus, PubMed, Web of Science, and any other databases specified in your protocol. For each database:

  1. Run your documented Boolean search string
  2. Export results in RIS format (recommended) or CSV
  3. Note the record count and date of each search — this goes in your PRISMA flow diagram

Expected record counts: A thorough systematic review search typically returns 500–5,000 records before deduplication, depending on topic specificity. If you’re getting fewer than 200 from a multi-database search, your search string may be too narrow.

Step 2: Import into Rayyan and deduplicate

  1. Create a free account at rayyan.ai
  2. Create a new Review — name it clearly and set the review type (systematic or scoping)
  3. Upload your RIS files: click Add Articles → Upload and import each database’s export separately. Rayyan keeps track of which database each record came from, which you’ll need for PRISMA reporting.
  4. After uploading, go to Duplicates in the sidebar. Rayyan auto-detects potential duplicates; review and confirm or reject each suggestion.

Manual deduplication check: Rayyan’s deduplication is good but not perfect. After auto-deduplication, do a spot check — search for known duplicates by author name to verify they were caught.

Record your pre- and post-deduplication counts. For PRISMA:

  • “Records identified through database search”: your total before dedup
  • “Records after duplicates removed”: your deduplicated count

Step 3: Define and document inclusion/exclusion criteria

Before screening, your inclusion and exclusion criteria must be documented — not developed during screening. For PRISMA compliance, criteria should be defined in your protocol (if you registered one in PROSPERO) and clearly stated in your methods section.

In Rayyan, you can add your criteria as Labels that reviewers can apply when excluding records. Create a label for each exclusion reason:

  • Wrong population (e.g., “Non-human subjects”)
  • Wrong intervention
  • Wrong outcome
  • Wrong study design (e.g., “No control group”)
  • Wrong language
  • Conference abstract / grey literature (if excluding)

Consistent labeling is what makes your exclusion decisions auditable.

Step 4: Set up blind dual review

For PRISMA-compliant systematic reviews, title/abstract screening by two independent reviewers is standard:

  1. Invite your co-reviewer to Rayyan (they need a free account)
  2. In the review settings, enable Blind Mode — reviewers see each other’s decisions only after both have decided on a record
  3. Assign the full record set to both reviewers

Each reviewer should independently apply include/exclude/maybe decisions with exclusion labels. Rayyan tracks decisions without revealing the other reviewer’s choice until both have screened a record.

For single-reviewer scoping reviews: Single-reviewer screening is acceptable for scoping reviews if documented. Skip blind mode setup and proceed directly to screening.

Step 5: Use Rayyan’s AI suggestions

Once you’ve screened at least 50–100 records, Rayyan’s AI model has enough signal to start generating relevance predictions. In the Insights panel:

  • Records with a high AI relevance score are more likely to be relevant based on your pattern of decisions
  • Records with a low score are likely excludable

How to use this productively:

  • Sort the remaining unscreened records by AI score (lowest first) to batch-screen clearly irrelevant records more quickly
  • Never use the AI score as the decision — use it to prioritize your order of review and to spot-check low-scoring records you might otherwise rush through

Document your approach: if you use AI suggestions to batch-screen low-priority records, describe this in your methods (“AI-assisted prioritization was used to order screening; all records received human review”).

Step 6: Resolve conflicts

After both reviewers complete screening, Rayyan surfaces Conflicts: records where one reviewer included and the other excluded.

Standard reconciliation approaches:

  1. Discussion: reviewers review conflicted records together and reach consensus
  2. Third-party adjudication: a senior author or third reviewer decides on unresolved conflicts

Document the number of conflicts and resolution method for your PRISMA methods section. A high conflict rate (>20% of records) suggests your inclusion criteria need clarification — revisit and tighten before proceeding if possible.

Step 7: Export included records

After reconciliation, you have your included record set — the papers that will proceed to full-text review.

  1. In Rayyan, export included records as RIS
  2. Import into Zotero as your “Full-text review” collection
  3. Record the count for your PRISMA flow diagram: “Records screened”, “Records excluded (with reasons)”, “Full-text articles assessed for eligibility”

Step 8: Document everything for PRISMA

PRISMA 2020 requires a flow diagram showing:

  • Records identified (by database)
  • Duplicates removed
  • Records screened
  • Records excluded (with numbers per reason)
  • Full texts retrieved
  • Full texts excluded (with reasons)
  • Studies included in synthesis

Rayyan’s built-in PRISMA Flow Diagram generator populates this automatically from your decisions — export it for your manuscript.

After screening: moving to full-text review

Title/abstract screening gives you your candidate set. Full-text review and data extraction are the next stages. For data extraction, see our tutorial on using Elicit for systematic literature review.

Time estimates by record volume

Records after deduplication Approximate screening time (single reviewer)
< 500 2–4 hours
500–2,000 1–3 days
2,000–5,000 1–2 weeks
> 5,000 Consider AI-assisted prioritization; allow 3–6 weeks

These are rough estimates for experienced reviewers. Add ~30–50% for new reviewers unfamiliar with the topic.

Related content

← Back to Learn