Guidelines for Guidelines

Overview

The AI in Guidelines: An Australian Lens workshop, held as part of Cochrane Australia’s Living Evidence 2026 Symposium, brought together approximately 70 Australian guideline developers, researchers and partner organisations to explore how artificial intelligence (AI) can support guideline development. Guideline developers recognised that AI is already being used across parts of the guideline development process particularly for high‑volume, repetitive and administrative tasks.

Participants emphasised the importance of ensuring AI should not be used as a replacement for expert judgement, but rather to assist teams in undertaking structured, rules-based tasks (for example, meeting minutes, de-duplication, summarising, screening support, and evidence surveillance). The greatest advantage to guideline development would be reducing workload pressures and creating capacity for additional activities that require deliberation, judgement, clinical and methodological expertise. Resources could additionally be redirected to other value-add activities that improve engagement with interest holders, tailor dissemination, support implementation and evaluate impact.

Participants identified several risks with the use of AI that could undermine trust and credibility if unmanaged, particularly hallucination, inaccuracies, inconsistent outputs, and privacy or confidentiality concerns. Participants noted that some of these risks could be managed by using a 'human-in-the-loop' approach, always treating AI outputs as draft material and ensuring outputs are subject to full review, verification and sign-off processes. Further reflections from workshop participants emphasised:  

• using AI strategically by prioritising its use for tasks that are high burden (for example, automation of repetitive work) without compromising methodological quality.
• embedding safeguards for responsible use by protecting confidentiality (for example, closed systems where required, de-identification and clear disclaimers), actively managing bias, and maintaining credibility through oversight and checks for known AI limitations.
• strengthening governance and transparency by developing standards and disclosure requirements to clarify acceptable tools, data handling expectations, validation requirements, and how AI assistance is reported in outputs.
• building capability by improving AI literacy and workforce confidence through training, shared learning opportunities and practical guidance to enable safe, effective and equitable uptake.

Overall, workshop participants were cautiously optimistic about the use of AI in guideline development but wanted further support in understanding the risks. Strong governance frameworks, standards and transparent disclosure of AI use were viewed as essential to building confidence in users, with a focus on ensuring the source data is understood and its application used appropriately and consistently across activities.

For further insights and a detailed summary of workshop outcomes, please view the full AI in Guidelines Workshop Report.