Systematic Literature Reviews (SLRs) are widely used in academic research to identify, evaluate, and synthesise existing evidence on a clearly defined research question. Unlike a traditional narrative review, an SLR follows a transparent and reproducible process that normally includes database searching, screening, eligibility assessment, quality appraisal, data extraction, and evidence synthesis.
In 2026, artificial intelligence is increasingly being used to support several stages of this process. AI tools can help researchers discover literature, refine search terms, screen large numbers of records, classify studies, extract information, summarise findings, and organise themes. These capabilities can save time and improve efficiency, particularly when researchers are working with hundreds or thousands of publications.
However, AI-assisted reviewing also introduces important risks. Automated tools may miss relevant studies, misclassify papers, generate incorrect summaries, invent references, overlook methodological limitations, or produce results that are difficult to reproduce.
The most effective approach is therefore not to replace systematic review methodology with AI, but to integrate AI carefully into a transparent, researcher-led workflow.
Why AI Is Becoming Important in Systematic Reviews
The volume of academic literature has increased dramatically across many disciplines. Researchers conducting an SLR may retrieve hundreds or even thousands of records from databases such as Scopus, Web of Science, PubMed, IEEE Xplore, or subject-specific repositories.
Manually reviewing every title, abstract, and full text can be time-consuming.
Artificial intelligence can help manage this workload.
AI-based tools may support:
keyword identification,
search-string development,
semantic literature discovery,
duplicate detection,
title and abstract screening,
study classification,
data extraction,
thematic coding,
evidence mapping,
summarisation,
and visualisation.
These functions can reduce repetitive work and allow researchers to devote more attention to interpretation and critical appraisal.
AI-Assisted Literature Discovery
One of the earliest stages of an SLR is identifying relevant literature.
Researchers traditionally begin with keywords, Boolean operators, controlled vocabulary, and database-specific search strategies. AI tools can assist by suggesting synonyms, related concepts, alternative spellings, and broader or narrower terms.
For example, a review on public transport accessibility might include terms such as:
"public transport," "public transit," "mass transit," "accessibility," "first-mile," "last-mile," "connectivity," and "mobility."
AI can help researchers identify related expressions that may otherwise be overlooked.
However, an AI-generated search strategy should not be accepted automatically. The researcher must test and refine it within each database.
Database searching remains essential because an SLR must be reproducible.
Semantic Search and Research Discovery
Traditional database searching often depends heavily on exact keywords.
AI-powered semantic search can go further by identifying papers that are conceptually related even when they use different terminology.
This can be particularly helpful in interdisciplinary research.
For instance, similar concepts may be described differently in urban planning, public health, transportation engineering, geography, or environmental studies.
AI-based discovery tools can therefore help researchers locate relevant literature outside their immediate disciplinary vocabulary.
However, semantic recommendation systems may not clearly explain why a particular paper was suggested.
Researchers should therefore distinguish between:
systematic database searching, which should form the documented core of the review, and
supplementary AI-assisted discovery, which can help identify additional sources.
AI for Title and Abstract Screening
Screening is often one of the most labour-intensive stages of an SLR.
After duplicates are removed, researchers must decide which records potentially meet the eligibility criteria.
AI tools can help prioritise records based on relevance.
For example, a machine-learning system may learn from the researcher's inclusion and exclusion decisions and then rank remaining records according to their likelihood of relevance.
This can make screening more efficient.
However, automated screening should be used cautiously.
A paper incorrectly classified as irrelevant may contain valuable evidence.
Researchers should therefore document:
which AI tool was used,
how it was trained,
what screening criteria were applied,
whether human reviewers checked excluded records,
and how disagreements were resolved.
Human oversight remains essential.
AI and Full-Text Screening
Full-text screening requires more detailed judgement than title and abstract screening.
A paper may appear relevant based on its abstract but fail eligibility criteria when examined fully.
AI may assist by locating important sections such as:
study population,
methodology,
geographic setting,
intervention,
outcome measures,
sample size,
or publication type.
This can help researchers screen more efficiently.
Nevertheless, final inclusion decisions should generally remain with qualified reviewers because eligibility may depend on subtle methodological details.
AI can support the decision, but the researcher should make the final judgement.
AI for Data Extraction
Once studies have been included, researchers typically extract structured information into a data-extraction table.
Common variables include:
author and year,
country,
study design,
sample size,
research objective,
methodology,
variables,
findings,
limitations,
and quality indicators.
AI can accelerate this process by identifying relevant information from full-text articles.
However, extraction errors are possible.
For example, an AI tool may confuse the total sample with a subgroup, report an adjusted result as an unadjusted estimate, or incorrectly interpret a statistical measure.
The safest practice is to use AI-assisted extraction followed by human verification.
Important numerical information should always be checked against the original source.
AI for Thematic Analysis and Evidence Synthesis
In qualitative or mixed-method systematic reviews, researchers often need to identify recurring themes across multiple studies.
AI can support preliminary thematic coding by grouping similar concepts and suggesting common patterns.
For example, a review of public transport user perceptions might identify recurring themes such as:
safety,
reliability,
affordability,
comfort,
frequency,
accessibility,
travel time,
information quality,
and cleanliness.
This can help researchers organise evidence.
However, AI-generated themes should not automatically become the final conceptual framework.
Researchers must consider:
study context,
theoretical meaning,
methodological quality,
contradictions,
and differences across populations.
Thematic synthesis requires interpretation, not merely text clustering.
Opportunities Offered by AI
The use of AI in systematic reviews offers several important benefits.
Faster Processing
AI can help researchers process large numbers of studies more quickly.
Better Organisation
Automated classification and tagging can make large evidence bases easier to manage.
Improved Discovery
Semantic search may identify relevant studies missed by narrow keyword searches.
Reduced Repetitive Work
Tasks such as duplicate detection, screening support, and preliminary extraction can be partially automated.
Greater Accessibility
AI tools can help researchers understand technical articles, translate text, or simplify complex terminology.
Support for Interdisciplinary Reviews
AI can identify connections across fields that use different terminology.
These advantages make AI particularly useful when combined with rigorous review methods.
Major Risks of Using AI in an SLR
Despite its benefits, AI introduces several risks.
Fabricated References
Generative AI may invent article titles, authors, journals, or DOIs.
Researchers should never cite a source unless it has been independently verified.
Incorrect Summaries
AI may simplify or misrepresent study findings.
Screening Bias
An automated system may prioritise certain types of studies and overlook others.
Lack of Transparency
Some AI systems provide recommendations without explaining the underlying decision process.
Data Extraction Errors
Numbers, sample sizes, effect measures, or study characteristics may be extracted incorrectly.
Overdependence
Researchers may become too dependent on AI-generated interpretations rather than reading the original studies.
Confidentiality Concerns
Uploading copyrighted or unpublished material into third-party AI systems may raise privacy, copyright, or licensing issues.
PRISMA and AI-Assisted Reviews
Researchers should continue to follow established systematic review reporting frameworks.
PRISMA remains an important framework for transparent reporting of systematic reviews.
An AI-assisted review should still document:
databases searched,
search dates,
full search strategies,
number of records retrieved,
duplicate removal,
screening process,
eligibility criteria,
excluded studies,
included studies,
and synthesis methods.
The use of AI should increase efficiency without reducing reproducibility.
If an AI tool played a significant role in screening, extraction, classification, or analysis, its use should be described clearly in the methodology.
Best Practices for Researchers
Researchers using AI in systematic literature reviews should follow several practical principles.
First, define the review protocol before using AI.
The research question, eligibility criteria, databases, search strategy, and planned analysis should be established in advance.
Second, verify all AI-generated information.
Never assume that a summary, reference, statistic, or classification is correct.
Third, maintain human oversight.
Important inclusion, exclusion, appraisal, and interpretation decisions should remain researcher-controlled.
Fourth, document AI use.
Record the tool name, purpose, stage of use, and verification process.
Fifth, preserve reproducibility.
Maintain search strings, screening records, data-extraction sheets, and decision logs.
Sixth, use validated quality-appraisal tools.
AI should not replace established instruments for assessing risk of bias or methodological quality.
Seventh, protect sensitive information.
Avoid uploading confidential manuscripts, restricted datasets, or participant information to unapproved systems.
Eighth, disclose meaningful AI assistance when required.
Researchers should follow the policies of their university, journal, and publisher.
AI Should Support, Not Replace, Critical Appraisal
The central purpose of an SLR is not simply to collect papers.
A high-quality review evaluates the reliability, relevance, methodological strength, and implications of available evidence.
AI can identify patterns, but it cannot automatically determine whether a study is methodologically sound.
For example, two studies may report similar conclusions but differ significantly in sample size, study design, bias, measurement quality, or generalisability.
The reviewer must interpret these differences.
Critical appraisal remains fundamentally a scholarly responsibility.
Conclusion
Artificial intelligence is transforming how systematic literature reviews are conducted.
AI tools can support literature discovery, screening, data extraction, thematic analysis, and evidence organisation. When used responsibly, they can reduce repetitive work and help researchers manage increasingly large bodies of academic literature.
However, AI also introduces risks such as fabricated references, incorrect summaries, classification errors, extraction mistakes, bias, and lack of transparency.
The most reliable approach is therefore a human-led, AI-assisted systematic review.
Researchers should continue to follow recognised review frameworks, document their methodology, verify AI-generated outputs, preserve reproducibility, and maintain responsibility for all inclusion decisions and conclusions.
For academic researchers and research-support organisations such as EduPub, AI offers substantial opportunities to improve the efficiency of systematic reviews. The key is to combine technological assistance with transparent methods, critical appraisal, and rigorous human verification.
In systematic reviewing, AI can make the process faster—but research quality still depends on the judgement, integrity, and methodological discipline of the researcher.
