Decades ago, if you wanted to listen to music on the go, you packed a stack of CDs (or tapes), a Walkman, and headphones. You were your own DJ, but only within the limits of what you could carry. The music lived in plastic cases, and if you forgot an album at home, you went without.
Research worked much the same way.
If you were doing hard science or scholastic research in a library, discovery was a physical act. You consulted a card catalog or a software-based version of one. You might then navigate the library with the aid of a placard map if it was big enough, or ask where a specific wing was. Once you were in the right spot, you would search for a specific call number, hoping the spine was actually on the shelf. Then you would pile up the books or journals you were looking for and sit down to see if they had what you were after. If the book or journal was missing, you initiated an inter-library loan and waited. Sometimes days, sometimes weeks, for a truck to arrive from a partner university with the physical text.
That memory underscores the fact that for decades, even centuries, the unit of science was the Paper: a linear, frozen narrative that required a front-to-back commitment. But today, the paper is being disintermediated.
The end of the frozen narrative
The internet and the rise of search engines were the first wave; generative AI is the second. The latter is still disruptive to the former in ways that are only beginning to be understood. “Typical search behavior for scientific queries used to be two or three searches, read an article, leave,” says Cameron Ross, SVP of Generative AI at Elsevier, as the company rolls out LeapSpace, an AI-powered research synthesis tool. “Now the average prompts per user per visit is somewhere between eight and ten. People are having conversations with research literature.”
Ross describes a fundamental inversion of the publishing model. “The power of AI assistance is really about complementing researchers so they can look around corners,” he says. “How do I traverse beyond what I know or was trained on years ago? Can AI glean new scientific ideas from the periphery I haven’t considered?”
If the scientific paper was once like a vinyl album, a fixed sequence of information, tools like LeapSpace are turning publishers into something like streaming services. The AI rips the “tracks” (data, methods, conclusions) out of the “album” (the PDF) and remixes them into a custom playlist that answers a specific question.
“We don’t see AI as a replacement,” Ross clarifies. “It’s about better speed reading so you can decide which article to dig deeper into, or which methodology to try in the lab.”
Solving the volume problem
LeapSpace is Elsevier’s answer to a scientific community drowning in volume. “Of course, I will always suggest you should read that full article,” Ross says. But with millions of scientific articles published per year, “people are becoming more and more used to using an AI assistant like this,” Ross said.
“Good grief,” he adds, “it’s about time we helped busy, overloaded, intelligent people spend less time on the admin of science.”
They’re not alone in that diagnosis. Startups like Elicit, Consensus and Scite have been chipping away at the literature review problem for years. Wiley, meanwhile, has taken a different strategic path: its AI Gateway makes content interoperable across Claude, Perplexity, Mistral and AWS. Their first co-innovation partner, Potato AI, is focused specifically on protocol generation, helping researchers not just find papers, but execute reproducible experiments.
Redefining the article for an AI world
Both Elsevier and Wiley are betting that verified, peer-reviewed content still matters in an era when foundation models can synthesize answers from anywhere, including, potentially, pirated corpora. Here, the music industry again offers similar parallels. The entertainment industry tried the alternative: lawsuits, DRM and “You Wouldn’t Steal a Car” warnings that played before movie showings in cinemas and on DVDs you’d already paid for. Observers have compared Sci-Hub to Napster, but most of that discourse stopped at the piracy question. What happens after? If the music industry is any guide, the answer has to do with infrastructure and evolution.
Elsevier’s platform synthesizes answers from a growing roster of partner publishers and Scopus, its abstract database covering over 100 million records. Ross describes working with a major European automaker, traditionally dominant in internal combustion engines, now using LeapSpace to upskill its mechanical engineers in electrical engineering and software design.
LeapSpace also offers what Ross calls “deep research” mode: a multi-agent process that generates structured reports with evidence ratings rather than quick answers. The feature mirrors a capability that has rapidly become a common feature across AI platforms. Google launched Deep Research in December 2024; OpenAI followed two months later; Anthropic, Perplexity, and Grok now offer variations. The difference, Elsevier argues, is what’s being searched. A general-purpose deep research tool scours the open web. LeapSpace queries a curated corpus of peer-reviewed literature, and increasingly, content from partner publishers, with provenance baked in.
