Objective: This lightning talk will provide a brief description of a scoping review designed to identify the extent to which generative AI is being used in the search methods of evidence synthesis reviews. We will compare tools and strategies used by review authors, and extract details on the level of reporting.

Methods: The JBI scoping review methodology will guide the conduct of this review. After calibration exercises on screening and data extraction were completed, an a priori protocol was published on OSF Registries. To be eligible for inclusion, a review must be a known type of evidence synthesis, and authors must have either used a generative-AI powered tool to develop database search strategies or used an AI search engine to locate references directly. We will search from Jan 2022 to present: Ovid Medline, Ovid Embase, Ovid PsycINFO, EBSCO CINAHL, EBSCO ERIC, ProQuest Sociological Abstracts, Elsevier Scopus, and Clarivate Web of Science Core Collection. We will conduct a supplementary full-text search in EBSCO MEDLINE, EBSCO CINAHL and Lens.org. We will independently screen in two stages in Covidence; disagreements will be resolved by consensus and discussion. We will extract study characteristics; characteristics related to the method, type of chatbot or AI search engine used; description of search method; and search reporting elements. The results will be presented in tables, accompanied by descriptive summaries.

Results and Discussion: This project will provide insights into the adoption and reporting of generative AI tools in KS searches.

Background: Searching trial registries is a mandatory item in the Cochrane MECIR guidelines, and two registries, the World Health Organization International Clinical Trials Registry Platform (ICTRP) and ClinicalTrials.Gov (CTG), are specifically recommended in the Cochrane Handbook of Systematic Reviews of Interventions (Chapter 4). As ICTRP contains records from CTG, a search of both resources will lead to duplicate records.

Automated tools and existing published deduplication methods are not suited to deduplicating registry records as they are based on comparison of fields such as title, author, journal name, year, and other metadata fields that are common to bibliographic records. Some of these fields are not commonly used in registry records, and even fields such as the title can differ across the same trial’s record from various sources.

Methods: This lightning talk will demonstrate the problem, including why software like Covidence fails at detecting duplicates between ICTRP and CTG. We will also present a method for deduplication of registry records in EndNote (desktop) using the unique study ID field as the sole deduplication parameter.

Conclusion: This method is particularly suited for librarians supporting reviews of interventions who have to search multiple resources that contain trials records (CTG, ICTRP, and even Cochrane CENTRAL).