Guest post from Jenn Huck, Associate Director of Research Data Services & Social, Natural, and Engineering Sciences. Huck assists researchers and teachers in identifying and accessing an array of numeric and geospatial data and is the liaison to the School of Data Science.
You may have heard that generative artificial intelligence (AI) can sometimes “hallucinate” citations to scientific articles. In 2025, for example, a report from the presidential Make America Healthy Again Commission referred to made-up research papers, leading the White House to remove the report and later release a corrected version. Later that year, The Independent, a Canadian newspaper, revealed that a government-commissioned Deloitte health care report contained citations pulled from academic papers that did not exist and also credited real scientists with articles they hadn’t worked on.
If you plan to use AI to assist with your research, I want to share a new tool with you that will help you avoid citing hallucinated articles.
Introducing Consensus
Consensus is an AI-powered search tool for academic research. It identifies academic papers related to your research question, then it uses AI to synthesize an evidence-based summary of findings. It always includes citations to real papers.
You may be wondering: What checks does Consensus have in place to make sure that the output is trustworthy? First, it is a closed system — its large language model (LLM) is trained on academic papers only. Second, it is verifiable — every claim links to a real paper. Third, Consensus uses a “search first” approach — Consensus searches the academic literature first, then uses AI to synthesize the results. It is not possible for Consensus to make up fake sources or cite wrong facts. It is technically possible for Consensus to mistakenly summarize real sources, but there are models in place that minimize that risk. Overall, hallucinations in Consensus will be very rare.
How does Consensus compare to Google Scholar or other databases?
Consensus summarizes the top results in a way that Google Scholar and disciplinary databases do not. Each Pro search in Consensus will provide a summary of the most relevant 20 results (or 50 in a Deep Search). Google Scholar and other databases will provide as many hits as possible that match your keywords — that could be a dozen or in the millions depending how specific, common, or rare your search terms are. Consensus will also use its LLM to generate a search strategy with words that share the same meaning; if you ask a question about “self-driving cars” it will know to also search for “autonomous vehicles.” Google Scholar and other databases won’t do that for you — the researcher would have to try (and know to try) a few keyword terms to get the best results.
Anyone interested in transparent, repeatable searches will be better off relying on a disciplinary database like PsycINFO, which offers rich subject-term searches, and other filters like date. It is far from guaranteed that Consensus will retrieve the exact same citations with the same summary, even with the exact same prompt.
Who should use Consensus?
Consensus is strongest for scientific and social science fields and less so for arts and humanities. The tool is very effective at providing results for disciplines where there is a strong culture of article publication; it is even stronger for disciplines that have a strong culture of Open Access article publication. That means that Consensus is very good for research questions in the sciences and most social sciences. It is less helpful in fields where book publication or non-textual materials are the main product of scholarship — most of the arts and humanities fall in this category.
Graduate students, faculty, and anyone else focused on research should check out the advanced features that Consensus offers:
- Deep Search Mode — Deep Search conducts more searches with more varied search strategies (up to 20 targeted searches) and returns 50 citations.
- Connect your Zotero library to Consensus — this allows you to find gaps in your existing collection of research papers and find papers that help fill those gaps.
- Explore how citations are connected to each other — You provide the Graph feature with a few relevant papers, and Consensus will find other relevant works based on citation patterns.
- Integrate Consensus into your chat workflow — You can take advantage of everything that Consensus offers in Claude or ChatGPT. You can also combine the best of Consensus and your AI tool for new abilities, such as searching for grants based on gaps in the research literature.
Get started by signing up at www.consensus.app with your UVA email address. Our UVA license gets you unlimited Pro searches and 50 Deep searches a month. You can also connect Consensus to UVA Library subscriptions, making it easier for you to access the full text versions of cited articles. Consensus uses LibKey (a favorite tool of ours for getting to articles quickly). Go to Account > Settings > Preferences > University of Virginia to get started.
Learn even more about Consensus at our Consensus research guide. For more about ethical use, citations, and considerations for use of generative AI, check out our Generative AI at UVA research guide.