ISPOR 2026 exposed a growing tension between commercial momentum and best practice guidelines in AI-assisted literature reviews.
On one side, an expanding commercial ecosystem is increasingly marketing AI-assisted literature reviews, with platforms and vendors touting staggering efficiency and quality claims.1, 2 While accuracy claims at ISPOR Europe six months earlier rested on relatively small and simple data sets, data presented at ISPOR 2026 suggested a higher extraction accuracy for subgroup efficacy and adverse event data, although these findings may not yet be peer-reviewed.
On the other side, ISPOR’s Good Practices Task Force on GenAI for systematic literature reviews (SLR) urges caution, arguing that full end-to-end automation is not yet appropriate for literature reviews.3 In their rapid review assessment, accuracy outcomes for the use of AI in literature reviews were heterogenous. Without a universally positive outcome, they stopped short of issuing a blanket recommendation for AI in literature reviews. Instead, the focus was on AI-assisted triage at screening and first-pass data extraction. This aligns with Costello Medical’s own peer-reviewed research on using AI for data extraction in literature reviews, which demonstrated that human oversight remains essential, particularly for complex and nuanced data extractions.4
A recent opinion piece published in Nature by the Editor-in-Chief of the Cochrane Collaboration calls into question the efficiency gains touted at ISPOR International; it argued that using AI took longer than manual methods due to the training and oversight required.5
There is also a regulatory and governance challenge for AI in literature reviews. Copyright licensing and data-access issues were at the forefront of discussions at ISPOR 2026, more so than at ISPOR Europe. These areas remain uncertain and unresolved. And while position statements have been issued by bodies including NICE,6 there is no precedent for a fully AI-assisted literature review being accepted by a health technology assessment (HTA) body.
So where does this leave those of us seeking to implement AI in literature reviews responsibly, rigorously, and efficiently?
We can pursue exploratory studies and pragmatic literature reviews that are not intended for regulatory submission, within a human-in-the-loop framework and reporting AI use transparently in line with best-practice guidelines, such as RAISE.7 But until HTA and regulatory bodies and journals are seen to accept AI-assisted literature reviews, and copyright licences allowing content to be using in generative AI become commonplace, AI-assisted literature reviews are unlikely to become widespread for these purposes. Additionally, it remains unclear what the real efficiency savings will be once human review for quality is factored in.
Despite these challenges and questions, Costello Medical welcomes this shift toward governance-aware AI use and will continue to work with clients, regulators and publishers to align innovation with scientific rigour. We urge platform developers and sponsors to embrace external validation, publish limitations, and engage with HTA bodies and publishers to support broader adoption in a compliant and rigorous way. Compliance, rigour and transparent methods, underpinned by robust external validation and clear alignment with HTA, regulatory and publishing requirements, are essential for AI in literature reviews to become a credible, widely accepted commercial tool.
References
If you would like any further information on the summary presented above, please get in touch, or visit our Literature Reviews page to learn how our expertise can benefit you. Liz Lunn (External Engagement Manager) and Saoirse Leonard (Evidence Development Director) created this article on behalf of Costello Medical. The views/opinions expressed are their own and do not necessarily reflect those of Costello Medical’s clients or affiliated partners.
