Chat with papers you find or upload

Elicit’s new Notebook features add a flexible “chat with papers” workflow that lets researchers ask questions across papers they either search for in Elicit’s database, upload as PDFs, or combine from both sources. The core value is control: users can select a specific set of papers, then switch into an interactive chat mode to compare findings, drill into details, and refine which papers matter—without losing the grounding that comes from working directly with real academic documents.

The workflow starts from a new Notebook screen where users can search for papers on a topic (for example, “long-term effects of cash transfers”) and simultaneously upload their own documents. As results load, users can select any subset of papers—either from a single search query, from the uploaded set, or from a mix of both. Once the desired papers are highlighted, adding a new step opens chat mode tied to exactly that selection.

In chat mode, users can pose tasks such as “compare and contrast the papers.” Elicit highlights the four selected papers involved in the question, and users can choose whether to rely on full text when available or fall back to abstracts. That choice affects speed and cost: using abstracts is faster and cheaper, while full text supports deeper answers.

A key differentiator is that this chat experience sits alongside Elicit’s structured capabilities. Users can incorporate other Elicit actions—such as extracting main findings into columns—then use that structured output to decide which papers to chat with next. The interface supports an iterative loop: structured analysis helps narrow the set, and open-ended chat helps interpret and interrogate it.

The system also provides transparency and traceability. Clicking a paper reveals its full text and the tables referenced in the paper, and users can extract data from those tables when high-accuracy mode is enabled (Elicit Plus). Chat supports follow-up questions on the same paper set, including requests for specific study details like sample sizes. If the direction changes, users can select a new set of papers and start another chat step.

Under the hood, Elicit’s grounding is a major selling point. Because the tool operates over a database of public papers and can also work with uploaded PDFs, it avoids fabricating nonexistent citations—unlike some chat-first AI tools that may hallucinate references. Notebooks extend beyond chat as well: users can summarize selected papers, run additional queries to fetch more relevant literature, and then keep building the notebook in a tabular format when that better matches research needs.

Overall, the Notebook chat workflow is positioned as a practical bridge between exploratory questioning and structured literature analysis—especially useful when researchers don’t yet know what they’re looking for and need to scope criteria, evidence types, and study characteristics as they go.