New mind-blowing AI research tool makes systematic literature review easy

A new tool from Scispace, called “deep review,” aims to compress a Prisma-guideline systematic literature review from weeks or months into roughly minutes. The workflow centers on running a “systematic literature review” task that opens a new window with a ready-to-edit prompt, then generating Prisma-aligned outputs—research question framing, protocol criteria, screening steps, and structured reporting—at a speed that would be hard to match manually.

The process starts with tailoring the prompt. Users replace the default topic with a specific research focus (the example shifts from microplastic contamination to professional discrimination against non-native English speaker teachers). They also set the time window, which defaults to the last 10 years but can be extended to 20 years. The transcript adds practical guidance for choosing dates: check whether similar systematic reviews already exist and whether recent work has appeared; if the topic is fast-moving (like AI), narrower windows such as the last 3–5 years may be more appropriate.

Next comes database selection, which directly shapes the Prisma flow. The example keeps Web of Science, removes PubMed for non-medical relevance, retains Scopus, and adds ERIC to target education research. Users can also include regional databases. The tool is configured to produce key Prisma artifacts such as a PRISMA flowchart and a risk-of-bias table, while allowing further scoping via the prompt—such as limiting to T-OLE programs (teaching English as a second language programs).

A major selling point is transparency and replicability. The system shows step-by-step actions aligned with Prisma: interpreting the research question, finalizing inclusion criteria, searching multiple academic databases, screening titles and abstracts, and applying exclusion criteria. It also surfaces the search strategy—keywords and search strings—so researchers can rerun the same queries themselves to verify results, check for missed papers, and reproduce the search process across databases.

As the search runs, results appear incrementally, with downloadable CSV files for each database and additional files that reflect the tool’s own review database. Users can compare papers by adjusting columns, and they can download documents such as the criteria review already formatted. The tool also handles labor-intensive tasks that commonly consume time in traditional reviews: downloading full text, deduplicating overlapping records from multiple databases, and extracting relevant data.

Finally, Scispace generates a structured report in LaTeX form and then compiles it into a single PDF. The output includes standard systematic review sections: background and rationale, significance, prior research, research questions, methodology (protocol, inclusion/exclusion criteria, search strategy, study selection process, and study characteristics), results organized by themes (e.g., cultural/ideological factors and organizational policies), and a conclusion with implications and recommendations for future research. The transcript emphasizes a key limitation: AI-generated reviews still need a novel contribution to be publishable, and it points to a follow-up video on finding high-impact research topics for publication.