03 · What You Need to Know
When AI Literature Discovery Becomes a Reportable Research Method
What Counts as AI-Assisted Literature Searching?
AI-assisted literature searching includes the use of artificial intelligence to identify, retrieve, recommend, screen, organize, or evaluate scholarly publications.
Different tools perform different activities. Some use semantic retrieval to identify papers related to a research question. Others generate summaries, recommend connected studies, classify abstracts, or help researchers construct search queries.
For example, a researcher might use an AI-powered discovery system to identify publications concerning self-regulated learning and generative AI. Another researcher might use a machine-learning screening tool to prioritize records during a systematic review.
These activities should not be treated as methodologically identical simply because both involve AI.
The appropriate reporting depends on whether the tool provided exploratory assistance or influenced a defined research procedure.
Why Is Preliminary Literature Exploration Different From a Systematic Search?
Preliminary exploration helps researchers become familiar with a topic, identify terminology, and locate potentially relevant publications.
It may involve searching databases, following citations, reading review articles, and considering recommendations from colleagues or software tools.
A systematic search, by contrast, is a documented method intended to identify evidence according to specified eligibility criteria and a defined review question.
For systematic reviews, the search strategy, information sources, dates, and selection procedures are central to evaluating the completeness and reproducibility of the review.
Exploratory AI-assisted searching
Using AI to discover possible sources, terminology, or research directions without treating the tool's results as a complete or formally defined evidence set.
Methodological AI-assisted searching
Using AI within a documented procedure that influences which studies are identified, screened, included, or analyzed.
The distinction is important because methodological AI assistance may affect the evidence base itself, not merely the researcher's familiarity with the literature.
Does Using ChatGPT to Find Research Papers Require Disclosure?
Not automatically under every publisher's policy.
A researcher may ask ChatGPT to suggest search terms or identify potentially relevant publications before independently searching scholarly databases.
Such assistance may be exploratory, although some journals require broad disclosure of generative AI use during research or manuscript preparation.
However, ChatGPT-generated bibliographic information should not be accepted without verification. A language model may generate nonexistent references, incorrect publication details, or inaccurate summaries of real articles.
Researchers should locate the actual publications through appropriate bibliographic sources and examine their relevance.
If ChatGPT materially determines which studies are included in a formal review, the assistance becomes more consequential and may require methodological reporting.
The absence of copied AI-generated text does not automatically eliminate disclosure obligations.
Do AI-Powered Academic Search Engines Require Disclosure?
The answer depends on the tool's function and the applicable policy.
Academic discovery systems may incorporate machine learning into ranking, semantic retrieval, citation recommendations, or search assistance.
Using an ordinary search interface that contains AI-based ranking is not necessarily equivalent to using a generative AI system to make substantive research decisions.
Researchers should avoid assuming that every algorithmic feature embedded in a database requires a separate AI declaration.
However, when a particular AI feature materially shapes a formal search or screening procedure, reporting its role may be necessary for methodological transparency.
For example, a review that relies on an AI-powered relevance-ranking system to decide which records receive screening should explain how that system affected the process.
The relevant issue is the tool's contribution to the evidence-selection method rather than the presence of AI somewhere in its technical infrastructure.
What Do Major Publishing Authorities Say?
The International Committee of Medical Journal Editors (ICMJE) recommends disclosure of AI-assisted technologies used in submitted work and distinguishes writing assistance from AI used in research procedures.
The Committee on Publication Ethics (COPE) similarly emphasizes transparency concerning AI tools used in manuscript preparation and research activities.
These positions provide general principles for reporting consequential AI assistance, although neither should be interpreted as prescribing an identical disclosure statement for every academic search feature.
Researchers should also consider reporting standards applicable to their research design.
For systematic reviews, PRISMA 2020 requires transparent reporting of information sources, search strategies, and study-selection processes. PRISMA-S provides additional guidance for reporting literature searches.
These reporting standards are not themselves universal AI-disclosure policies, but their requirements are directly relevant when AI contributes to the search methodology.
What Does PRISMA 2020 Require Researchers to Report?
PRISMA 2020 provides reporting guidance for systematic reviews, including requirements concerning information sources, search strategies, and selection processes.
Its checklist calls for authors to specify the databases, registers, websites, organizations, and other sources searched, along with the dates of the searches.
It also requires presentation of the full search strategies for the databases, registers, and websites searched, including relevant limits and filters.
For study selection, PRISMA 2020 asks authors to describe the methods used to determine whether studies met inclusion criteria, including the number of reviewers and whether automation tools were used.
This last requirement is particularly relevant to AI-assisted screening.
If an automation tool helped determine which records were eligible for inclusion, researchers should describe its role rather than presenting the process as entirely manual.
PRISMA 2020 should not be treated as a substitute for the journal's AI policy. Both sets of requirements may apply.
What Does PRISMA-S Add to Literature Search Reporting?
PRISMA-S extends the transparency of literature-search reporting by providing a structured approach to documenting search strategies and related procedures.
It addresses information sources, search strategies, limits, search updates, and other elements relevant to reproducibility.
For AI-assisted literature searching, the underlying principle is that readers should be able to understand how the search was conducted and evaluate whether important evidence might have been missed.
For example, reporting that an AI tool was used to find relevant studies would be insufficient if the manuscript does not explain the actual search procedure or how records were selected.
PRISMA-S does not establish a blanket rule that every use of an AI discovery tool must be disclosed. Its relevance lies in the need for transparent reporting of consequential search methods.
Does AI-Assisted Abstract Screening Require Disclosure?
AI-assisted abstract screening generally warrants methodological reporting when it contributes to determining study eligibility.
Screening decisions affect which studies enter a systematic review and therefore influence the resulting evidence synthesis.
An AI tool may prioritize likely relevant records, suggest inclusion decisions, or automatically exclude records according to a classification model.
These functions have different implications for the risk of missing eligible studies.
Researchers should describe the tool's role, how screening decisions were made, whether humans reviewed the outputs, and how potentially incorrect exclusions were addressed.
For example, an AI system used only to prioritize the order of human screening differs from a system authorized to exclude records without human review.
That distinction should be apparent in the methods section.
What Is the Difference Between AI Prioritization and AI Exclusion?
AI prioritization changes the order in which records are examined. AI exclusion removes records from further consideration, either automatically or through a decision influenced by the model.
The two activities create different methodological risks.
Prioritization may improve screening efficiency without necessarily reducing the number of records eventually reviewed.
Automatic exclusion can create a greater risk of missing eligible studies if the model incorrectly classifies relevant records.
Researchers should therefore report whether all records were ultimately screened, whether an early-stopping rule was used, and how the sensitivity of the screening procedure was assessed.
The adequacy of safeguards depends on the review design and the consequences of missed evidence.
A general statement that AI accelerated screening would not explain these important distinctions.
Can AI-Assisted Searching Introduce Selection Bias?
Potentially. AI discovery systems may favor publications that are well represented in their indexed sources, citation networks, or training data.
They may also rank studies according to relevance measures that do not correspond precisely to a review's eligibility criteria.
For example, a semantic search system may prioritize widely cited English-language studies while retrieving fewer publications from regional journals or less represented languages.
Such differences can affect the apparent scope of the literature.
However, the existence and magnitude of bias depend on the particular system and search context. Researchers should not assume that every AI search tool has identical limitations.
Where AI contributes to formal evidence retrieval, its coverage and potential selection effects should be considered alongside the broader search strategy.
Can AI Literature Discovery Replace Database Searching?
Not automatically. AI-powered discovery can complement established search methods, but its suitability as a primary retrieval method depends on the review question and the tool's documented performance.
For a systematic review, researchers must establish that the search strategy is appropriate for identifying relevant evidence.
A generative AI response listing several apparently relevant papers does not provide the coverage, reproducibility, or transparent search logic ordinarily expected of a systematic evidence search.
Even specialized AI search tools may have limitations concerning indexing, retrieval, ranking, and documentation.
Researchers should therefore justify the selected information sources and procedures rather than assuming that an AI-generated bibliography constitutes a comprehensive search.
AI assistance should be evaluated as part of the overall search methodology.
What About AI-Assisted Citation Chaining?
Citation chaining involves identifying additional publications through references cited by known studies or through later publications that cite them.
AI-powered tools may help visualize citation networks, recommend related papers, or prioritize potentially relevant connections.
When such assistance forms part of a formal review search, researchers should document the procedure sufficiently to explain how additional records were identified.
For example, a review might report the seed publications used for citation searching, the tool employed, and the date of the search.
Where AI-generated recommendations were screened for inclusion, the selection procedure should also be described.
In an informal literature review, the reporting requirements may be less extensive, depending on the journal's policy.
Does AI-Assisted Study Data Extraction Require Disclosure?
AI-assisted data extraction goes beyond locating publications because it influences the information recorded from included studies.
A model may extract sample sizes, intervention characteristics, outcome measures, or reported effect estimates.
Errors in extraction can affect evidence tables, meta-analyses, and the conclusions of a review.
Researchers should therefore report consequential AI-assisted extraction procedures and describe how extracted information was verified.
For example, a reviewer may independently compare AI-extracted values with the original articles.
Such assistance is methodologically significant even when the AI-generated extraction text does not appear verbatim in the manuscript.
It should not be described merely as literature searching if the tool also contributed to data processing or analysis.
What Information Should Be Reported About AI-Assisted Searching?
The appropriate detail depends on the role of AI and the reporting standard governing the review.
| Reporting Element |
Why It Matters |
| Tool and relevant version |
Identifies the system used, where this information is available and required. |
| Search or screening task |
Clarifies whether AI retrieved, ranked, screened, or extracted evidence. |
| Information sources |
Establishes which databases or collections were searched. |
| Search dates and strategies |
Supports evaluation and possible reproduction of the search. |
| Eligibility criteria |
Explains how studies were judged relevant to the review. |
| Human verification |
Describes how automated suggestions or classifications were evaluated. |
| Automation limits |
Identifies whether records were automatically excluded or screening stopped early. |
Not every item is required for every exploratory search. For formal reviews, reporting should satisfy the applicable methodological standard and journal instructions.
Where Should AI-Assisted Literature Searching Be Disclosed?
When AI contributes to the search or selection methodology, the methods section is generally the appropriate place to describe the procedure.
A separate AI-use declaration may also be required by the journal.
For example, a systematic review may describe AI-assisted abstract screening under study selection while separately identifying the tool in a designated disclosure statement.
Where AI was used only for preliminary topic exploration, reporting may be governed by broader AI-use policies rather than formal search-method requirements.
Researchers should distinguish methodological disclosure from acknowledgment of writing assistance.
What Should an AI-Assisted Literature Search Disclosure Say?
A useful methodological statement should explain the actual procedure rather than merely naming the tool.
Illustrative Methods Disclosure
AI-Assisted Abstract Screening
"An AI-assisted screening tool was used to prioritize records retrieved from the database searches. All retrieved titles and abstracts were subsequently evaluated by human reviewers against the prespecified eligibility criteria. The AI system did not independently exclude records from the review."
This statement is appropriate only if all records were actually screened by humans and no automatic exclusions occurred.
The specific tool, version, dates, and other details should be added where required by the journal or review reporting standard.
Researchers should not describe an automated exclusion procedure as human-verified screening if the records were not independently evaluated.
Does Using AI to Summarize Retrieved Articles Require Disclosure?
AI-assisted article summarization is distinct from literature retrieval, although the activities may occur within the same workflow.
A researcher may use AI to generate preliminary summaries of publications or extract information for a review.
If these summaries materially influence the evidence synthesis, their role may require disclosure under the applicable policy.
Researchers must also verify that the summaries accurately represent the original studies.
A model may omit limitations, exaggerate findings, or attribute conclusions to a paper that the authors did not make.
For systematic reviews, AI-assisted extraction or synthesis should be reported as part of the relevant methodology rather than concealed under a general statement about searching.
What If the AI Tool Recommends a Paper That Is Never Used?
Not every unused recommendation necessarily requires publication disclosure.
Researchers routinely encounter irrelevant records during searching, and formal reviews report selection procedures rather than narrating every individual search result.
However, if AI recommendations formed part of a systematic retrieval process, the procedure may still need to be documented even when particular records were excluded.
The distinction is between recording the method and reporting each unsuccessful suggestion.
Researchers should follow the applicable review reporting standards and preserve appropriate search records.
Can AI-Assisted Literature Searching Be Used Without Human Verification?
That depends on the task, but consequential AI outputs should not be accepted without appropriate validation.
AI-generated citations should be checked against genuine bibliographic records. Screening decisions should be evaluated according to the review's eligibility criteria. Extracted data should be compared with original sources.
The extent of verification may vary depending on the system's validated performance and the research design.
However, a general claim that an AI tool is accurate does not establish that its outputs are reliable for every topic, database, or study population.
Human researchers remain responsible for the evidence included in the final review.
What If the Journal Does Not Mention AI Literature Searching?
Researchers should review the journal's broader AI policy and the reporting standards applicable to the study.
For systematic reviews, methodological reporting obligations remain relevant even when the journal has no separate AI-search declaration.
If uncertainty remains about a consequential AI-assisted procedure, the editorial office may be consulted.
Researchers should retain accurate records of the search strategy, tool involvement, and verification performed.
Policy silence should not be interpreted as permission to omit information necessary for readers to understand how the evidence base was assembled.
