Adopting AI for systematic literature reviews — quickly, but cautiously
10 June 2026 | 5 min read
AI tools are already proving to be time savers for life science professionals, but maintaining accuracy and transparency is paramount.
As you might assume of any job that has the word “systematic” in its title, systematic literature reviews (SLRs) are time and labor intensive. And in a highly regulated industry where lives are often at stake, these reviews are critical to ensuring informed decision-making.
That’s why it’s encouraging to hear from Dr. med. Katharina Friedrich, a regulatory consultant and owner of Katylistic, Inc., that AI is already helping to speed up a typically arduous process.
“I can definitely say, based on several SLRs that I did in the last weeks and months, that I probably saved around 20 to 30% of the time, considering the whole SLR flow chart. Especially because you save a lot of time when you don't have to create tables manually, but also because it really speeds up the title and abstract screening,” she reports.
“All the information about the population, indication, comparatives and outcomes, that's information that you usually already get from the abstract. And I think the accuracy of AI features is already very high. I would say maybe 95%, although this is just a very subjective estimate.”
Dr. med. Katharina Friedrich, regulatory consultant and owner of Katylistic, Inc.
Integrating AI across the SLR workflow
Amy Ferguson, a medical research librarian and founder of Evidence Search Lab, sees ways that AI tools can be useful at each step of the question and protocol development in the SLR process. “Chatbots are helpful for brainstorming ideas, coming up with what you might want to research and developing PICO questions.”
She also uses AI to support exploratory searching in the early stages of a review. “That wouldn't be considered one of your official searches, but it helps you find some of that literature that you need to get started because you do need some information about the topic to do research on it.”
As noted, the screening phase is where AI shines the most, helping researchers spend less time on that “very tedious part,” as Amy puts it.
Amy Ferguson, medical research librarian and founder of Evidence Search Lab
“Typically, when I do a search, I'll end up uploading hundreds or sometimes more than a thousand results,” she says. “So at least two researchers have to go through each one of those titles and abstracts, which can be very time-consuming. But a lot of our systematic review tools have priority screening now, which means the reviewers only need to review a set amount of the results.”
The system learns from the reviewers, and as it better understands what they’re looking for, it can start supplying the most relevant results first, making the process quicker.
For the data extraction stage, AI’s contribution is imperfect but quickly improving. “I've seen some tools that are really getting better at the data extraction,” she says. “They're not just pulling the data from the papers, but they're also highlighting within the paper where they're finding that data.”
“There's still a human involved,” she adds, “confirming that information is correct. But the human doesn't have to comb through a 100-page PDF to find that data.”
Similarly, in the risk of bias phase, AI can flag what it sees as low or high risk, providing the sources that led to that choice – but ultimately, human judgment is still a critical part of the process.
Transparency and traceability are essential
SLR professionals take their responsibilities very seriously, and traceability and auditability are key to ensuring that their reviews are accurate and trustworthy. AI tools can be incorporated into their work, but only in ways that maintain necessary transparency.
“For me, it's very important that I can clearly find the information where the output comes from so that I can just very briefly cross-check,” explains Katharina. “As a regulatory writer or researcher, you really have to make sure that the information is correct. You can only do this in a good way when the information is highlighted in the full-text article or whatever document you're working on.”
Amy emphasizes that you must “document, document and document,” recording what tools you’re using, which version of those tools and how you are using them. “Somebody should be able to take a look at your methods, follow those methods, get similar results to you and reach the same conclusion. That's why transparency and reproducibility are hallmarks of the systematic review.”
Why human-in-the-loop is critical to AI in SLRs
While AI is well-equipped and getting better at handling the many tedious tasks involved in reviews, Amy sees human involvement as an absolute necessity: “AI is good at going through information and maybe finding things, but it doesn't actually understand what you're putting in there and what it's saying, so making the judgment calls is still going to be with the humans.”
She sees data extraction as the biggest risk area where SLR experts need to be vigilant, making sure that the AI is pulling all of the data requested and not skipping anything or misrepresenting it. “AI prioritizes fluency over factual accuracy, so it's possible it can pull out something that sounds plausible but is actually entirely false – just a hallucination.”
While there aren’t established standards on AI and SLR to help reviewers navigate these potential pitfalls, support is on the way. The International Collaboration for the Automation of Systematic Reviews is a coalition of organizations now working on recommendations that they call RAISEopens in new tab/window (Responsible AI in Evidence Synthesis). “We are getting guidance on these things, but that guidance is still in development because all of this is so new and it changes so fast,” says Amy.
Advice on bringing AI into your workflow
We asked Katharina and Amy to share best practices for SLR professionals looking to incorporate more AI tools into their research workflows. Katharina suggests only using AI features that provide clear sourcing, so you know where the output is coming from. “One of my favorite examples is always Embase AI, because you insert a question, then you get an answer, but all the information is referenced. You can see, ‘Okay, the sentence comes from this publication,’ and then you can get into the publication and see if it makes sense.”
Amy also values that traceability. “A benefit of Embase AI as opposed to a lot of the other search tools is that you actually see the search string that it uses … you can see what subject terms it's used,” she says. “And that can start as a basis for your final search string.”
Future of AI in SLR
While AI continues to evolve in speed and accuracy, much of the day-to-day efficiency in SLR workflows still comes from automation. “I think many of the things that make my life easier when it comes to SLRs are actually automation,” says Katharina. “For example, the automatic creation of a Prisma flow chart, or the automatic creation of tables with an overview of inclusion/exclusion criteria. These basic things, this is all automation and not AI.”
At the same time, AI could be opening brand new possibilities in reviewing. Amy believes it could enable what she sees as the future of SLRs: living systematic reviews. “Right now, you work on a systematic review, it takes months, and of course, by the time it's published, there's already been new research that's come out that could affect it,” she explains. But with a “living” review, AI could be used to monitor the literature in real time.
Humans, with the aid of AI, would set up the living systematic review at the start, developing protocols and methodology. Then AI could continue monitoring the literature (with human oversight, of course) in an ongoing review.
“So, when healthcare providers are making decisions, they are able to make decisions on current evidence and not have to rely on evidence that could be out of date just because systematic reviews take so long,” says Amy.
Evolving SLRs into living reviews may be a more long-term development. In the meantime, one of the great hopes for AI, across the industry, is that it can take over or speed up the most time-consuming tasks, allowing for greater focus on what matters most.
“I somehow feel like I lose the oversight because I'm just so busy extracting information from the publication and putting it into a document,” says Katharina. “I really hope that I will have less repetitive work in the future, and I'll have more time to really think about: what kind of information did I get? What does it mean for my project? What does it mean for the research question that I want to answer?”
Watch the recent webinar “Integrate AI into systematic reviews without compromising scientific rigoropens in new tab/window” to get more insights from Amy and Katharina.
Related Articles
The cost of waiting: why delaying AI adoption is a strategic risk for R&D
What 3,000 AI projects taught us about implementation