Quick Guide to Using SciSpace for Literature Review (Agentic & Non-Agentic Methods)

SciSpace’s literature review tool splits the workflow into three modes—Standard, High Quality, and Deep Review—then contrasts that “non-agentic” setup with an AI agent that can actively reshape searches and generate fully customized reports. The biggest practical difference is that Deep Review expands coverage by transforming the original query into multiple search directions and then following citations and references, producing far more papers and a more comprehensive, report-style output.

In Standard mode, a user query returns a short insight built from the top five most relevant papers (with an option to expand to top 10). Results appear as a relevance-sorted list, and users can inspect cited papers and open each paper’s page. A key workflow feature is the ability to add columns to the results table—either predefined extraction fields or custom columns with instructions—so the system extracts additional information per paper and updates the table accordingly. Export options are available for the table and insights, but the overall structure stays fixed around relevance-sorted paper comparison.

High Quality mode keeps the same core layout and extraction features, but generates a longer, better-written insight by using a higher-capability model. The output becomes more detailed and easier to interpret when comparing criteria across papers, while still relying on the same basic query-to-results approach.

Deep Review is where the coverage jump becomes visible. Instead of using the query as-is, it asks follow-up questions when the request is underspecified, then constructs multiple transformed queries from the base question and the added parameters. Those queries run in parallel, yielding roughly 3x–4x more papers, and the example shown reaches about 1,750 papers versus far fewer in earlier modes. Deep Review then traverses citations and references from the newly found papers to discover additional relevant work, filters and reranks everything by relevance, and generates a large report with fixed sections such as abstract, introduction, scope, methodology, screening papers, results, and critical analysis. The section set is not meant to change per query, but the report is substantially more comprehensive than Standard or High Quality.

The non-agentic limitations become clearer when comparing report control and workflow flexibility. Standard/High Quality/Deep Review outputs are templated: users can’t request arbitrary file formats through instructions (beyond built-in export), can’t add or remove report sections, can’t target specific databases or journal sites via the report prompt, and can’t easily run follow-up tasks on top of prior results as part of the same instruction chain.

SciSpace’s agentic approach addresses those constraints by letting the system interpret detailed criteria and execute multi-step tasks. In the example, the agent searches PubMed for peer-reviewed research published after 2020, transforms the query into PubMed-compatible forms, runs multiple PubMed queries, combines and reranks results, and then produces a Word-formatted report with a target length (1500 words) and specific content requirements—strengths and limitations, diagnostic analysis, current consensus, open challenges, future research directions, and more. The agent also supports iterative refinement via follow-up questions, and it can export in common formats (including Word and PDF) and save papers into a SciSpace library with folder organization for later retrieval.