ASReview LAB makes literature screening easier to audit

What this is about

ASReview LAB is an open-source tool for systematic reviews, meta-analyses, and other screening tasks with large text datasets. It helps researchers avoid reviewing titles and abstracts in an arbitrary order by showing likely relevant records earlier. The official website emphasizes open software, privacy, no tracking cookies, and use by universities, governments, and institutions.

The topic is not new, but it is highly practical in 2026: teams face larger literature sets while AI tools in scientific workflows are being examined more critically. ASReview is interesting because it does not merely promise that "AI does the review". It keeps humans in the decision loop and makes the screening process easier to document.

What ASReview LAB actually does

ASReview LAB imports references from common literature databases and shows reviewers individual records. For each record, a human decides whether it is relevant or irrelevant. The system learns from those decisions and prioritizes the next records. This is active learning: the software does not independently produce the final answer, but asks the human for the next useful examples as efficiently as possible.

Beyond screening, ASReview offers simulations and analysis. Teams can use fully labeled datasets to test which model, strategy, or stopping rule fits their review. The documentation describes ASReview LAB as a web-based application built on the ASReview Python package; the GitHub project points to Apache-2.0 licensing and active development.

Why it matters

Systematic reviews matter in medicine, social science, public policy, and technology, but they are expensive. A search can produce 1,000, 5,000, or 20,000 records. If a team screens every record manually in arbitrary order, days or weeks can pass before it is clear which studies are relevant.

ASReview targets that bottleneck. An independent evaluation in a medical context reported that ASReview could reduce screening time by 83 percent on average while identifying 95 percent of relevant studies. Those numbers are not a universal performance guarantee for every project. They do show why a transparent screening tool can be useful for real research teams: it saves more than clicks, because it forces documentation of decisions, models, and limits.

In plain language

Imagine a huge stack of job applications. Instead of reading every application alphabetically, you first show an assistant a few clear examples: suitable, unsuitable, maybe suitable. After that, she puts the most likely next folders on top of the stack. You still make the decision, but you no longer work blindly from top to bottom.

ASReview does that with literature records. It does not take away the scientific judgment, but it learns how to sort the stack.

A practical example

A research team studies digital interventions against burnout and finds 6,200 records across several databases. Three reviewers first screen 200 titles and abstracts together to calibrate relevance criteria. The team then uses ASReview LAB to prioritize the next records.

After 1,400 screened records, the curve shows that almost all remaining records are clearly irrelevant. The team does not stop blindly. It documents the model, version, training decisions, stopping rule, and a manual sample of deferred records. If reviewers or supervisors ask later, the team can explain what the software did and what remained a human decision.

Scope and limits

• ASReview does not replace a sound search strategy. If the database search never imports important studies, the tool cannot find them.
• Active learning depends on early human labels. Poor or inconsistent decisions at the start can skew prioritization.
• Not every discipline accepts AI-assisted screening without additional documentation. Teams must clarify reporting duties, reviewer expectations, and reproducibility before using it.

The best next test is small: take a completed review with known inclusions, run a simulation in ASReview, and compare how early relevant records would have appeared. Only then should teams move the tool into a live review.

SEO & GEO keywords

ASReview LAB, systematic review, active learning, literature screening, open source research tools, meta-analysis, human in the loop, reproducible research, Utrecht University, Nature Machine Intelligence, evidence synthesis, research workflow