Elicit + PRISMA 2020: What AI Can Automate in a Systematic Review Now
The question for AI systematic-review tools used to be whether they could find papers. It has moved on: Elicit now frames its systematic-review product explicitly around PRISMA 2020, describing the workflow as reproducible, traceable, and auditable at every step (elicit.com/blog/systematic-review-for-prisma-2020). That framing is worth taking literally rather than as marketing, because PRISMA compliance is a property of a documented process, not a checklist bolted onto a finished report.
Search: Safe to Accelerate
Elicit documents the databases, exact queries, and filters behind every search it runs, and offers a keyword mode the company describes as fully reproducible alongside an AI-assisted semantic mode for terminology a manual search might miss. This is the part of PRISMA reporting that AI genuinely strengthens: a search a human forgot to log is a bigger reporting gap than a search an AI ran correctly and then wrote down. Automating the logging, not the judgment, is the safe win here.
Screening: A Second Reviewer, Not a Replacement
Screening is where Elicit's own numbers require the most care. The same PRISMA 2020 post cites 97% sensitivity and 93% specificity on abstract screening, framed as approaching the accuracy of two human reviewers, and every exclusion decision comes with a captured reason and a supporting quote from the abstract. The documentation frames the tool as able to support two human reviewers or stand in as the second one — not as a replacement for reviewer judgment altogether. That distinction matters because a benchmark accuracy figure is an average across many reviews; it says nothing about how the tool performs on a specific, narrow, or unusually worded topic, which is exactly where a second human reviewer earns their place.
Extraction: Traceable, Still Needs a Read
Every extracted data point links back to the exact quote, table, or figure it came from, which turns "trust the number" into "click through and check the number" — a meaningfully smaller ask. That link is what makes extraction auditable. It does not confirm the extraction is correctly interpreted; a quote can be pulled from the right place and still be misread if the surrounding context (which arm, which subgroup, which timepoint) gets flattened in translation.
Reporting: Where the Audit Trail Actually Pays Off
At the reporting stage, Elicit exports the flow diagram, inline citations, and structured extraction data together, so a reader can trace a number in the results back to the paper, and the paper back to the search that found it. That chain is what a PRISMA-compliant report is supposed to demonstrate. It is also the piece most manual reviews under-document, since assembling it by hand after the fact is tedious enough that authors often reconstruct it loosely rather than precisely.
Where This Sits Next to a Workflow Builder
None of this replaces the exclusion-logic discipline covered in Elicit's Workflow Builder for systematic review screening: a documented audit trail is only as good as the criteria it is auditing. The PRISMA framing adds the reporting layer on top of that screening logic, and the same discipline connects back to the broader case for source-grounded AI in research workflows — traceability is the feature, not the summary text.
Writing up a PRISMA-compliant methods section still means describing that audit trail in prose an editor can follow. The Prompt Pack: Paper Structuring ($5, researchcraft.gumroad.com) includes prompts built for exactly that Methods-section handoff, from search strategy to flow diagram narrative.