Artur Nowak represented Evidence Prime at IQWiG’s third Information Retrieval Meeting in Cologne on 24 April 2026. The meeting focused on how large language models can be introduced into the day-to-day work of finding and selecting evidence.
AI in the information specialist’s workflow
The programme brought together practical examples from health technology assessment agencies, academia and industry. Robin Featherstone presented Canada’s Drug Agency’s test-and-learn programme. Workshops considered transformer models as second screeners of titles and abstracts, automation in horizon scanning, the selection of suitable AI tools, and the industry perspective on evidence synthesis.
This was a useful setting for examining the difference between a tool that can produce a plausible answer and a process that an information specialist can use with confidence. Introducing AI into a review changes how work is checked as well as how it is performed. Teams need to define the task, understand the errors that matter, and decide how performance will be evaluated in their own setting.
Validation as part of implementation
IQWiG’s report from the meeting describes the emphasis on responsible tool selection and validation. The discussion also included tools for grey-literature searching and examples of AI use in horizon scanning, illustrating the range of tasks that sit within information retrieval.
For Laser AI, these discussions connect directly with a recurring question in evidence synthesis: what needs to remain visible when part of a workflow is automated? Search methods, selection criteria and the basis for a decision all affect whether reviewers can assess the result and investigate disagreements.
Our attendance offered an opportunity to compare perspectives with information specialists working through these questions in practice. The meeting’s focus on implementation made the methodological detail central to the conversation, alongside the capabilities of the technology itself.
