Get AI summaries of any video or article — Sign up free
Elicit AI Can Now Do Your Entire Systematic Review (In Minutes) thumbnail

Elicit AI Can Now Do Your Entire Systematic Review (In Minutes)

Andy Stapleton·
5 min read

Based on Andy Stapleton's video on YouTube. If you like this content, support the original creators by watching, liking and subscribing to their content.

TL;DR

Elicit’s systematic review workflow screens hundreds of papers using inclusion/exclusion criteria and assigns each paper an inclusion score.

Briefing

Elicit AI’s paid “systematic review workflow” can screen hundreds of papers and extract structured data in minutes—turning what normally takes hours or days of manual reading into a guided, step-by-step process with adjustable inclusion thresholds. In a live walkthrough using a mindfulness-and-anxiety research question, the tool pulled in 499 candidate papers, evaluated them against inclusion/exclusion criteria, and then extracted data from the subset that met the user’s relevance bar.

The workflow starts with a research question and then asks for a speed/comprehensiveness setting (the walkthrough used a “fast” option). Elicit generates a research report by first finding papers, then screening them, and finally extracting data. A key feature is the structured “systematic review” pipeline: it offers suggested screening steps, lets users choose whether to use AI-generated suggestions or skip directly to screening, and supports importing papers already found elsewhere (including uploading up to 500 PDFs). The interface also exposes screening dimensions—such as age range, data separation, intervention duration/type, and related study characteristics—so inclusion decisions aren’t a black box.

During screening, Elicit evaluates each paper and assigns an inclusion score. The walkthrough showed the system processing all 499 papers autonomously, with progress updates and counts of papers likely to be included versus excluded. Once screening completes, the user can tune the score threshold using a slider to control how strict the review is. In the example, a high score (near 4.9 out of 5) signaled very close alignment with the research question, while lower scores corresponded to weaker matches. Adjusting the threshold changed the number of included studies dramatically—down to a single paper at very high thresholds, or up to 107 papers at a lower cutoff.

After inclusion is set, Elicit performs data extraction across the included studies. It can use suggested extraction columns (including study design and randomization details) or add custom fields. The walkthrough emphasized that extraction is where most of the time goes, because it pulls detailed information per paper—participant demographics, intervention details, and outcome/effect-size elements—then compiles everything into a report.

The results appear as both an on-screen report and downloadable PDF. The report includes typical systematic-review components: an abstract, methods and flow-style accounting of how many papers were screened, included, and ignored, plus tables of outcomes and secondary outcomes. The walkthrough also noted that the full report generation can take roughly 10–20 minutes depending on reference counts.

Beyond systematic reviews, Elicit retains its broader research workflow: semantic paper search, research-question-driven summaries, filtering, and iterative steps like creating tables, summarizing concepts across papers, and “chat with papers” using full text (not just abstracts). The overall pitch is practical: a guided pipeline that forces the same decisions researchers make—while automating the bulk of screening and extraction.

Cornell Notes

Elicit AI’s systematic review workflow automates the core labor of a systematic review: screening hundreds of papers and extracting structured data from the included subset. In a walkthrough question about mindfulness-based interventions reducing anxiety in university students, Elicit screened 499 candidate papers using inclusion/exclusion criteria and assigned each paper an inclusion score. Users can adjust a score threshold to control how many studies qualify (e.g., 107 included at a moderate cutoff, far fewer at stricter cutoffs). After inclusion is set, Elicit extracts data using suggested or custom columns and then generates a detailed report and PDF, typically taking longer during extraction and then producing the final narrative and tables in roughly 10–20 minutes. This matters because it compresses days of reading into a guided workflow with transparent decision controls.

How does Elicit’s systematic review workflow decide which papers to include or exclude?

It runs a structured screening process over a set of candidate papers (499 in the walkthrough). Each paper is evaluated against inclusion/exclusion criteria tied to the user’s research question, and the system assigns an inclusion score (shown as a fraction like 4.9 out of 5 for highly relevant studies). Papers are then categorized as included or excluded based on that scoring and the user-controlled score threshold. The interface also surfaces screening dimensions such as age range, data separation, intervention duration/type, and other selectable columns, making the criteria visible rather than purely opaque.

What role does the “score threshold” slider play in shaping the final review?

The score threshold determines how strict the relevance requirement is after screening completes. Raising the threshold keeps only the most closely aligned studies; lowering it admits more borderline matches. In the walkthrough, a very high cutoff (greater than about 4.8) resulted in only one included paper, while a lower cutoff produced 107 included papers. This directly changes the size of the dataset used for extraction and the scope of the final report.

What happens during data extraction, and what kinds of fields can be captured?

Once included studies are selected, Elicit extracts detailed information per paper using suggested extraction columns or custom fields. The walkthrough highlighted study design and randomization details, participant demographics, intervention characteristics, and outcome/effect-size-related elements. Each included paper becomes a row in the extraction table, with Elicit populating the selected columns. Extraction is the time-intensive step because it pulls many data points from each paper before the report can be generated.

How does the workflow handle papers already found or uploaded by the researcher?

It supports reusing prior work. If a literature search has already been done, papers can be uploaded and then used in the systematic review workflow rather than starting from scratch. The walkthrough mentioned uploading up to 500 papers and also selecting from an existing library. This means the screening and extraction steps can operate on a researcher’s curated set, not only on newly retrieved results.

What do the generated reports include, and how do “fast” and “systematic review” differ?

The systematic review output includes typical review components: an abstract, methods, and a flow-style accounting of screened versus included versus ignored papers, plus tables of outcomes and secondary outcomes and references. The walkthrough also contrasted a “fast” report mode, which is less exhaustive but still detailed—useful for quick literature understanding. The full systematic review workflow is more comprehensive because it performs screening and deeper structured extraction across the included subset.

Beyond systematic reviews, what other research tasks does Elicit support in this workflow?

Elicit also supports semantic paper search and research-question-driven exploration, producing summaries of top papers and a row per retrieved paper with clickable sources. It enables filtering and adding columns (e.g., limitations) for up to eight papers at a time. It can create additional steps like summarizing abstracts, building tables, and “chat with papers,” including using full text to answer questions such as “What are the main findings from this paper,” with citations tied to the referenced passages.

Review Questions

  1. In the mindfulness/anxiety example, how did changing the inclusion score threshold affect the number of included papers and the scope of data extraction?
  2. Which parts of a systematic review does Elicit automate in the walkthrough (screening, extraction, report writing), and which user decisions remain adjustable?
  3. What kinds of extraction columns were shown as suggested or selectable, and how could a researcher add custom columns?

Key Points

  1. 1

    Elicit’s systematic review workflow screens hundreds of papers using inclusion/exclusion criteria and assigns each paper an inclusion score.

  2. 2

    A user-adjustable score threshold controls how many studies qualify for extraction, dramatically changing the review’s breadth.

  3. 3

    The workflow follows a systematic-review style pipeline: paper selection for screening, screening evaluation, extraction, and then report generation.

  4. 4

    Data extraction can use suggested fields (e.g., study design and randomization details) or custom columns, producing structured tables per paper.

  5. 5

    Report outputs include typical systematic-review sections such as methods and flow-style counts of screened, included, and ignored papers, plus outcome tables.

  6. 6

    Extraction is the most time-consuming step; report generation then compiles the extracted data into a detailed narrative and PDF.

  7. 7

    Elicit also supports broader literature workflows—semantic search, filtering, concept summaries, and “chat with papers” using full text.

Highlights

Screening 499 candidate papers ran autonomously with progress updates, replacing hours of manual reading with an adjustable, score-based inclusion process.
The score threshold slider acted like a control knob for review strictness—one cutoff produced 107 included studies, while a stricter cutoff left only a single paper.
The final outputs weren’t just summaries: the workflow generated systematic-review-style methods, flow accounting, and tables of outcomes and secondary outcomes, exportable as a PDF.
Elicit’s “chat with papers” can use full text (not just abstracts), producing cited answers tied to specific passages.

Topics

Mentioned