Artificial intelligence (AI) is a quickly evolving field. Over the past years, AI-based approaches evolved from simpler machine-learning methods to more advanced neural network–based tools expanding AI to a relational dimension. Adding to that, the recent generalization of access to large language models (LLMs) has affected the use of AI-based tools in the health research field (1). Examples include the development of reporting checklist extensions for trials of interventions involving AI and AI-based tools for supporting several stages in the conduct of systematic reviews (2–4).

Nevertheless, AI remains a concept difficult to be defined. Although several definitions exist (5), the High-Level Expert Group on AI of the European Commission defines AI as “systems that display intelligent behaviour by analysing their environment and taking actions—with some degree of autonomy—to achieve specific goals” (6). Despite the vague formulations of “some degree of autonomy” or “specific goals,” this definition allows to distinguish the concept of AI from those of “automation” or of an “algorithm.” Automation and algorithms are based on predefined conditions with setting of deterministic automation rules, without that corresponding to AI.

In past years, several frameworks were developed for the use of AI, including specific frameworks for AI use in health care authored by medical societies, universities, and other institutions (7–9). Some of these frameworks discuss ethical aspects like respecting human values and being inclusive (9). In addition, some of these frameworks address the need for transparency in the use of AI, including by discussing the aspects that should be obligatory to disclose (7, 8). Responsibility for AI-driven suggestions, education on how to use AI, and cybersecurity-related concerns are other important topics commonly addressed (7–9). A joint initiative of several evidence-related entities—including, among others, Cochrane, the Campbell Collaboration, and the Joanna Briggs Institute—has drafted recommendations for responsible use of AI in evidence synthesis (https://osf.io/cn7x4). To our knowledge, the use of AI in health guidelines has not been addressed in these frameworks.

In the health guideline context, AI may support the processes of planning, development and adaptation, reporting, implementation, impact evaluation, certification, and appraisal of recommendations; we will collectively refer to all these stages as “guideline enterprise.” For example, AI tools have the potential of supporting data analysis and pattern identification (for example, supporting the analysis and synthesis of evidence), supporting classification and methodological tasks (for example, rating the certainty of evidence), generating output complementary to the work of guideline panel members (for example, on question generation), or improving the quality of dissemination materials and tailoring them to different target audiences. However, there are many issues that require clarification and ground rules when it comes to AI and guidelines. Despite the potential of AI tools to render the guideline enterprise process more efficient, it is important for guideline developers to know their limitations and how to use them in the most adequate ways. Generally speaking, the use of AI in the guideline enterprise does not need to be seen as an “all-or-nothing” decision or as a “one-size-fits-all” recipe.

Established in 2002, the Guidelines International Network (GIN) (www.g-i-n.net) connects more than 100 guideline developing member organizations. The GIN has published highly influential documents—for example, proposing principles for disclosure of interests and management of conflicts of interest (10). Given the impending increased use of AI in guideline development, and considering the related uncertainty and concerns, the GIN board sought to propose principles for the development and use of AI tools or processes to support the health guideline enterprise. Thus, the objective of this work is to describe principles for the use of AI in the development and implementation of health guidelines. We did not focus on the use of AI-based tools in clinical practice. The principles are primarily relevant for those involved in practical tasks of the guideline enterprise, as well as for those participating in oversight of the processes, such as guideline developers and oversight committees. If appropriately reported by guideline developers, the principles will also allow guideline users (health care professionals, patients, and decision makers, among others) to evaluate if AI use adhered to the principles.