Advancing insight with AI: Best practices for rigorous research
Insights on choosing an AI tool that helps you work better, not just faster.
Choosing AI for research
AI is increasingly part of researchers’ daily work. In the Researcher of the Future report, a recent survey Elsevier conducted of 3,200 researchers worldwide, 58% of participants have used AI for work-related purposes, up from 37% in 2024.
These researchers — who included those in the physical sciences, health research, life sciences, and social sciences and humanities — showed the strongest interest in using AI to find and summarize the latest research, but were also interested in using it for literature reviews, drafting proposal grants, analyzing research data and drafting papers.
They see AI tools as a way to save time, but also to increase the quality of their work and research overall as well as accelerate the discovery of new knowledge.
Researchers are looking beyond convenience for support that helps them work better, not just faster. They’re concerned about productivity, but also about quality and trust in their research results.
With this in mind, use the best practices below to choose an AI tool fit for research.
Become familiar with research-grade AI
The first step in becoming a responsible user of AI is understanding the differences between a general-purpose AI and a research-grade AI tool. Any AI tool used for research should work the way that researchers do: iteratively, critically and with evidence that you can verify. Or in other words, "research-grade AI", which has three main features: trusted content as the foundation, responsible functionality and human oversight.
If you’re uncertain whether your tool is truly "research-grade", consider AI literacy training. Check your institution’s library for any resources. Trainings should include details about how AI tools operate, best use cases, ethical considerations and more. Understand how the tools you use generate responses and what content it contains. Our webinar series, "From insight to impact"opens in new tab/window, is an example of training that provides actionable guidance that you can apply immediately.
When you’re using LeapSpace, you can be confident that it is a research-grade tool. Not only does it draw on peer-reviewed, source-neutral citations indexed in Scopus, but it is evaluated by independent advisory boards (such as the LeapSpace Advisory Board) and developed with input from thousands of researchers, librarians and academic leaders worldwide.
Ensure that your AI has guardrails
Guardrails support the AI operating within certain parameters, and they are critical to enabling research-grade AI.
Some examples of guardrails include:
Developing and using responsible AI principles to inform how a new tool is developed
Using a set of curated content in an AI tool to avoid misinformation from the open internet
Including citations to verify responses
Keeping humans involved at every step of the process
Having an AI tool show its work when generating a response to provide transparency into its processes.
Why are guardrails so important? They help protect research. Your work to make breakthroughs relies on gathering credible evidence, documented through trusted sources and methods. Seen within the context of the broader research ecosystem, guardrails are vital to:
Mitigating bias and hallucination by grounding responses in trusted content and backing that up with citations
Protecting the scholarly record, and your reputation, from using false or out-of-date information
Maintaining trust in your research and in the scientific community
Retaining control over how and where you present your research outcomes
At Elsevier, we’ve developed and follow responsible AI principles. For the researchers using our products, this means responsible AI is not only a technical issue but a question of governance, explainability, fairness and accountability.
Verify your AI’s responses
AI guardrails are an essential protective measure, but they don’t think for you or replace your judgment. Critical thinking is the most essential aspect of “keeping humans in the loop” when using research-grade AI.
Always verify your responses with the original source, making sure you understand context and nuance. You can do this by:
Verifying claims and assessing citations: Ask yourself if the articles cited fit the claims made in the summary. Also, consider if the summary generally aligns with what you would expect.
Assessing the response for gaps and looking for possible misinterpretation: Based on your own knowledge of the topic, is there something missing or not considered? Is there a contradiction in the evidence that is returned? Use your responses to guide replies to the AI tool.
Being critical of any simple responses to a research question: As researchers know, science is built on consensus built through multiple studies rather than a single answer. An AI tool should be able to point out studies that conflict with another finding or tell you if there isn’t a clear consensus.
Some AI tools give an answer but not necessarily a path to verification. But research-grade AI tools use features like citations to support their users’ critical thinking.
LeapSpace uses Trust Cards, which gives a confidence rating for responses. This makes the evidence behind its answers transparent and helps show that a claim is verifiable. Researchers have asked for other features to help them evaluate responses as well. In response, we developed Claim Radar, which helps you understand how a claim is grounded.
Claim Radar finds related documents and shows whether they support a response, or if the knowledge base shows mixed results or contradictory evidence. You can also view the resulting Confidence Level or see follow-up suggestions.
Protect your privacy
Researchers often work with unpublished ideas, sensitive data, early drafts, collaborator information or institutionally governed materials. Before you upload your documents or write prompts that include information from your work, it’s important to understand what happens to that data, whether it is retained by the AI tool and whether it may be used for training or other downstream purposes.
Many available tools allow any information you provide to be used to train their models. If you’ve been using AI to think through your ideas or new concepts, you don’t want to use a tool that takes that information for its own purposes. Protect your privacy by using tools that don’t use your information to train their models.
Additionally, make sure you’re aware of your institution’s data privacy policies if you’re using tools that are not already approved. Your institution’s library can usually help with this.
When you work within LeapSpace, your privacy is respected. Elsevier is committed to protecting both individual researchers’ work and institutional data. Review our privacy principles for more information.
Next steps
Interested in learning more about LeapSpace? Sign up for email updates to get the latest news and articles with best practices, user insights, and product updates.
