Agentic Research Directory

Literature screening · Editorial guide

Screen literature without hiding the exclusions

Move from a search export to a defensible screening ledger, using AI to prioritize attention rather than decide what counts as evidence.

Editorial suggestion — not end-to-end tested. These workflows combine documented capabilities with our suggested research checks. Source review is not independent reproduction, a tool endorsement or a promise of time savings.

Goal & fit

Produce a traceable set of records ready for full-text assessment. This guide covers title-and-abstract screening, not a complete systematic review or an automated synthesis.

Suitable when

A bounded review question and an imported collection whose records can be screened against explicit criteria. Active-learning prioritization is useful to pilot when the team can audit what remains unscreened.

Not suitable when

An undefined search question, missing abstracts that prevent reliable decisions, or a review where the team cannot justify early stopping. A ranking cannot establish that the search itself was comprehensive.

Inputs to gather

  • A dated search export with stable record identifiers, titles and abstracts; retain the original database and search strategy.
  • Written eligibility criteria, known relevant examples and an agreed process for uncertainty, disagreements and stopping.
  • Reviewers with subject expertise, a place to store decisions and an approved environment for the material.

Suggested workflow

The sequence below is editorial guidance. Numbered references identify sources for tool capabilities, not evidence that this entire workflow was tested.

  1. Freeze the question before the ranking

    Write inclusion and exclusion rules with examples of borderline records. Decide who resolves disagreements and whether all records must be screened. Save the protocol version separately from the tool so later changes remain visible.

  2. Import and reconcile the collection

    Preserve the original export, then deduplicate a working copy. Check that titles, abstracts and record IDs survived import. Keep a mapping from merged duplicates to the retained record; do not silently discard reports that describe the same study.

  3. Calibrate on a shared sample

    Have reviewers apply the rules independently to a deliberately varied sample. Discuss disagreements before accepting model suggestions. For an ASReview route, supply human-checked relevant and irrelevant examples as prior knowledge, as described in the paper [1].

  4. Screen with prioritization, not automatic exclusion

    ASReview describes an active-learning loop in which reviewer labels update the ranking [1]. Rayyan documents relevance ranking and screening decisions [2]. Whichever route you choose, record the human decision and reason against the original record. Send uncertain cases forward rather than treating a low rank as proof of irrelevance.

  5. Audit the stopping decision

    Follow the agreed stopping protocol. Inspect records beyond the highly ranked set and review exclusions according to your risk tolerance and review standard. If a check finds missed eligible material, reopen screening and record the change. Without a defensible audit, finish screening the collection; do not claim complete recall from a quiet run of exclusions.

  6. Export the handover

    Export decisions, reasons, duplicate mappings and unresolved cases with IDs. Reconcile imported, duplicate, screened and unscreened records before handing potentially eligible records to full-text reviewers. Preserve software settings and the date of the export.

Alternative tools & routes

Choose by project needs and governance, not by an assumed quality ranking. Current costs, licences and service terms need checking before adoption.

ASReview

Active-learning screening route

The primary paper describes researcher-led labeling and reprioritization. Requires a record collection, reviewer labels and a defensible stopping procedure. The paper's simulations are not a recall guarantee for your review.

Directory evidence: ASReview · Primary source [1]

Rayyan

Collaborative review route

Vendor documentation describes imports, deduplication, screening, relevance ranking and blind mode. Check current account features and institutional terms before committing; this guide does not audit prices or privacy terms.

Directory evidence: Rayyan · Primary source [2]

Expected output

A screening ledger with record ID, source database, decision, exclusion reason, reviewer and protocol version; a duplicate crosswalk; a separate list of unscreened and uncertain records; and a full-text retrieval queue. A manual reference-manager or spreadsheet route remains a valid baseline.

Validation checklist

Use these as acceptance checks in your project record. This page does not save completion state.

  • Known relevant studies are present in the imported collection; gaps trigger a search review, not just reranking.
  • Every exclusion has a recoverable record and a reason consistent with the protocol.
  • Reviewer disagreements and any rule changes are documented and resolved or explicitly open.
  • The counts reconcile, and unscreened records are not reported as excluded.
  • The stopping rationale and audit findings are recorded without an unsupported recall or time-saving claim.

Privacy & research integrity

  • Check database licences before uploading abstracts or full texts. Unpublished manuscripts, reviewer comments and restricted corpora may need an approved local or institutional environment.
  • Ranking can amplify early labeling errors and underrepresent unfamiliar terminology or languages. Audit those cases deliberately.
  • Full-text eligibility, study-level deduplication, risk-of-bias assessment and synthesis remain separate research tasks. Neither listed tool makes this workflow independently validated.

Linked primary sources

Based on stored primary-source extracts accessed on 15 October 2026. They were re-read for these guides, not freshly fetched or independently reproduced. Features and terms may change.

  1. van de Schoot et al. — ASReview framework and screening workflow

    Primary research paper; supports the active-learning loop and human labeling. Results are author-reported.

    Source accessed: 2026-10-15

  2. Rayyan — Getting Started

    Vendor documentation of review stages and ranking; not an independent evaluation.

    Source accessed: 2026-10-15

← Choose another research playbook