From a research question to an appraised evidence base
Five steps, and you stay in control of every one. The mechanical parts are done for you, the methodological judgments stay yours, and the result holds up when a peer reviewer or an auditor checks it.
Protocol
Start from a question, not a blank template.
You describe the review in plain language, and Aibstracts drafts a structured protocol for you: population, intervention, comparator, outcomes, eligibility criteria, and a Boolean search strategy. Every version is saved, so the protocol stays auditable as it changes. You edit anything you want and lock it when you are ready. The result is a defensible starting point you can react to, not a blank PICO template you have to populate.
- A complete PICO and eligibility draft to react to
- A Boolean search strategy written for you
- Every version saved and traceable
Search
Find the relevant work without the noise.
Your strategy runs against a corpus of more than 220 million papers. Results come back deduplicated, and every record keeps its source, whether it came from the database search, citation snowballing, similarity, or a manual upload. Because provenance is recorded as you go, your PRISMA flow diagram is accurate instead of reconstructed from memory at the end.
- Search across 220M+ papers
- Duplicates removed automatically
- PRISMA source counts that are audit-ready
Screening
Spend reviewer time on judgment, not triage.
Every record is scored across relevance, evidence value, credibility, and more, then classified by study design. Aibstracts routes each study and gives an include, uncertain, or exclude recommendation with a reason, for both the title and abstract pass and the full-text pass. You confirm or override, and the trail records who decided what and why.
- Priority scoring across six dimensions
- Study design classified and routed
- A reason captured at every decision
Appraisal
The hardest, slowest part of the review, handled.
For randomized trials, Aibstracts runs RoB 2 in its parallel, cluster, or crossover form. For non-randomized studies of interventions, it runs ROBINS-I. The model answers each signalling question with a quote pulled straight from the source PDF, and deterministic decision rules, the published algorithms, turn those answers into domain and overall judgments. Studies that fit neither framework get a structured Evidence Profile instead of being dropped. You get a defensible appraisal for every study, with the methodological scaffolding handled for you.
- RoB 2 and ROBINS-I, run for you
- Every answer anchored to a source quote
- Judgments from published decision rules, not guesswork
Verify
Sign off with confidence, and prove it.
Override any signalling answer and the judgments recompute on the spot. Reconciliation flags where a full-text reading disagrees with the abstract-stage call, so nothing slips through unexamined. The most direct way to gauge trust is to re-run a review you have already done by hand and compare the appraisals study by study. What you are left with is a verified, structured, exportable evidence base that is ready to synthesize and ready to defend.
- Override anything and see judgments update
- Abstract and full-text conflicts reconciled
- A verified evidence base ready to export
What makes the result reliable
Reliability comes from specific choices about how the platform is built, layer by layer, from how a PDF is read to how the final judgment is made.
None of it shows up in the output, and that is the point. It is the plumbing that lets the appraisal hold up when a peer reviewer or an auditor checks it.
- Output A defensible, exportable evidence base auditable end to end
- Human Reviewer verification and sign-off you own the call
- Judgment Published RoB 2 / ROBINS-I decision rules deterministic code, not a model
- Grounding Source-anchored retrieval and embeddings the exact passage, not a paraphrase
- Corpus 220M+ papers · Semantic Scholar, OpenAlex, trials, grey lit coverage beyond one index
- Ingest GROBID + Docling document parsing structure, tables, references preserved
See it on your own review
Bring a real question. The fastest way to judge the workflow is to run one.
A free plan is available, no card required