AI Systematic Review SoftwareSystematic reviews in days, with every decision auditable.

Explainable AI for the life sciences, making medical evidence available in real time.

  • Researchers at over 150 institutions in over 40 countries
  • Research data stored in the EU
  • No training on your content
  • Every decision in the audit log
Projects/Title & Abstract ScreeningDashboardMy QueueAI ScreeningResults

Audit Trail12 Nov 2026, 14:32AI Screening SystemScreened as Include

55 included · 2,271 excluded · 8 unclear

Screening Details

Physiotherapy-led group programme for back pain in primary care

Demo Primary Care Journal · 2022 · DOI 10.0000/demo.2022.0932

AI AnalysisAI DecisionUnclearAI Confidence58%

Overall Summary

A physiotherapist-led group exercise programme compared with usual care in primary care. The abstract states neither how long participants had pain nor how they were allocated. Eligibility depends on the full text.

Key Evidence

  • “weekly physiotherapist-led exercise classes”
  • “patients were allocated to the programme or usual care”

AI Reasoning 5 Satisfied0 Violated3 Unclear

  • unclear: Adults with chronic non-specific low back pain Population

    The abstract gives no duration of pain.

  • unclear: Randomised controlled trial Study design

    Patients were allocated to groups, the method is not stated.

  • unclear: Acute or post-surgical back pain Population

    Acute cases are not ruled out.

  • satisfied: Supervised exercise therapy Intervention

    Group exercise sessions are led by physiotherapists.

  • satisfied: Usual care, no treatment or another active treatment Comparator

    The comparison is usual care.

  • satisfied: Pain intensity or disability reported Outcome

    Pain and disability are reported at six months.

  • satisfied: Participants under 18 years Population

    All participants are adults.

  • satisfied: Protocol or conference abstract without results Publication type

    Results are reported in full.

Audit TrailAI Screening SystemScreened as Unclear12 Nov 2026, 14:33

Every AI decision shows its reasoning and confidence · demo data

<1 month

To a draft review with Puraite

16 months

Industry average

The AI does its part in days. With your reviewers checking every decision, you have a full draft in under a month. A typical review takes 16 months.

Supported by Europe's leading AI and life science organisations

eXISTRWTH InnovationHigh Tech NRW AcceleratorFounders FoundationGerman Acceleratorgarage33QuhrPaderborn Universitylife science factoryUniversity Hospital CologneGHA - German Health Alliance

Why Puraite

How Puraite differs from generic AI tools and legacy review software.

Every decision shows its reasons.

For every screening decision, the AI lists which criteria are met, unmet or unclear. It quotes the supporting passage when there is one and reports its confidence from 0 to 100%.

Projects/Full-Text ScreeningManual ScreeningBatchExport

PDF Viewer

Methods

We conducted a systematic search of MEDLINE, Embase and Cochrane Central. Eligible trials enrolled adults (≥18 years) with moderate-to-severe rheumatoid arthritis (DAS28 ≥3.2) who received any approved IL-6 receptor inhibitor. Primary endpoint was ACR20 response at week 12.

AI citation: Population criterion

Eligibility Criteria

  • Adult RA patients (DAS28 ≥3.2)
  • IL-6 receptor inhibitor arm
  • RCT or controlled trial design
  • Min. 12-week follow-up
  • ACR20 primary endpoint

AI Recommendation: Include

Check each criterion against the highlighted passage · demo data

Every step documented for your dossier.

Your PRISMA 2020 flow diagram builds itself from your screening decisions. The AI pre-fills risk of bias for your assessors to accept or change. GRADE certainty is rated outcome by outcome. Every judgement goes into the audit log.

Projects/PRISMA Flow DiagramPRISMA 2020
PRISMA 2020 (default)AI transparency (PRISMA-trAIce)SVGPNGExcelPowerPointWord
IdentificationScreeningIncluded

Records from databases:

  • PubMed (n = 812)
  • Embase (n = 1,046)
  • Cochrane CENTRAL (n = 398)
  • Total (n = 2,256)

Records identified from:

  • Websites (n = 0)
  • Organisations (n = 0)
  • Citation searching (n = 0)

Records removed before screening:

  • Duplicate records removed (n = 684)
  • Records marked as ineligible by automation tools (n = 0)
  • Records removed for other reasons (n = 0)

Records screened

(n = 1,572)

AI-screened
1,572 (100%)
Human-screened
1,572 (100%)

Records excluded

(n = 1,498)

By AI
1,498
By human
1,492
AI agreement
97%

Reports sought for retrieval

(n = 74)

Reports not retrieved

(n = 3)

Reports assessed for eligibility

(n = 71)

AI-assessed
71
Human-assessed
71

Reports excluded

  • Wrong population (n = 18)
  • Wrong intervention (n = 14)
  • Wrong comparator (n = 9)
  • Wrong outcome (n = 11)
  • Wrong study design (n = 13)
  • Total (n = 65)
AI-excluded
62
Human-excluded
65

Studies included in review

(n = 6)

  • Quantitative synthesis (n = 6)
  • Qualitative synthesis (n = 0)
The PRISMA flow diagram builds itself and shows who decided · demo data

A human makes the final call.

People resolve every conflict and can override any AI decision. Each override is logged.

Title & Abstract Screening/Audit TrailRead-only

Supervised exercise versus usual care in chronic low back pain: a randomised trial

  1. AI Screening SystemIncluded, confidence 94%Decision with rationale and source quotes stored
  2. R. Okafor · Reviewer 1IncludedAdopted the AI decision after reading the abstract
  3. L. Brandt · Reviewer 2ExcludedReason given: comparator is another exercise programme
  4. SystemConflict openedTwo human decisions differ, record leaves the queue
  5. M. Sato · Project ManagerConflict resolved: IncludedComparator criterion I3 allows another active treatment
  6. SystemSent to full-text retrievalPRISMA count updated
Who decided what and when, for every record · demo data

The whole review or just the step you need.

Take a review from protocol to PRISMA report in one audited workspace. Already screened elsewhere? Start at data extraction.

Demo organisation/Projects

Start your evidence project

Recommended

Guided Review

Step by step through a complete, audit-ready review.

Best for most reviews

or start another way

Data extraction onlyStandalone statistical analysis

Studies already chosen? Start at data extraction and skip search and screening.

Run the whole review or start at the step you need · demo data

No training on your content

We do not use your content to train or fine-tune AI models. Our AI providers retain inputs only for the periods set out in our data processing agreement. You can export or delete your data. Section 14 of our Terms of Use explains how and when.

Security and data handling

How it works

  1. 01

    Import your records or search open databases.

  2. 02

    The AI screens and extracts, with a reason for every decision.

  3. 03

    Your reviewers check, resolve conflicts and export.

From protocol to evidence in six phases

The full systematic review workflow in one place.

Planning & Protocol

Answer a guided questionnaire or let the AI draft the protocol from your research question.

  1. Define Research Question (PICO)
  2. Conduct Scoping Search
  3. Draft the Protocol
  4. Register Protocol (optional)
Projects/ProtocolProtocolCriteria BoardSearchScreeningExtraction

Research Protocol

A systematic review to evaluate the efficacy of IL-6 inhibitors in patients with moderate-to-severe rheumatoid arthritis, comparing treatment outcomes versus placebo or standard of care.

Extract Criteria

Criteria Board

P
Population
Adults with moderate-to-severe RA
I
Intervention
IL-6 receptor inhibitors
C
Comparison
Placebo or standard of care
O
Outcome
ACR20 response at week 12
Turn your research question into clear eligibility criteria · demo data

Search & Management

Search the open databases from inside the project and import everything else by file.

  1. Develop Search Strategy
  2. Execute Database Searches
  3. Search Grey Literature
  4. Deduplicate Records
Projects/Search & ImportFile ImportDatabase SearchDeduplication

SourcePubMed

Drop your .ris or .nbib file hereor click to browse
TitleSourceYearStatus
Efficacy of IL-6 inhibitors in RA...PubMed2024Imported
A randomized controlled trial...PubMed2023Imported
Meta-analysis of biologic therapy...PubMed2022Imported
Search open databases and import the rest in one place · demo data

Study Selection

Screen manually, with the AI as second reviewer or AI-first. Resolve conflicts in place.

  1. Title/Abstract Screening
  2. Resolve TIAB Conflicts
  3. Full-Text Retrieval
  4. Full-Text Screening
  5. Resolve Full-Text Conflicts
Projects/Title & Abstract ScreeningManualBatch AISettings
18 / 40 reviewed86% AI confidence

Efficacy of IL-6 receptor inhibitors in rheumatoid arthritis: a systematic review and network meta-analysis

Demo Journal of Rheumatology · 2024 · 1,847 participants

AI Rationale

Population matches adult RA patients (P criterion met). Intervention aligns with IL-6 inhibitor protocol (I criterion met). Primary outcomes reported at week 12 (O criterion met).

ExcludeInclude
Reviewers decide record by record, with the AI's reasoning beside them · demo data

Extraction & Quality

Dual extraction and risk of bias with the tool that fits the study design.

  1. Design Extraction Form
  2. Extract Data (Dual)
  3. Risk of Bias Assessment
  4. Resolve Conflicts
  5. Contact Authors
Projects/Data ExtractionPanel ViewTable ViewExport

Source Document

...enrolled 342 patients across 14 sites. ACR20 response was achieved in 68.4% of patients in the treatment arm vs 28.1% in the placebo arm (p<0.001). Mean DAS28 reduction of 2.1 points

3 fields linked to this passage

Extraction Form

Sample size
342 patients
ACR20 (treatment)
68.4%
ACR20 (placebo)
28.1%
p-value
< 0.001
DAS28 reduction
2.1 points
Every extracted value links back to its source · demo data

Synthesis & Grading

Meta-analysis with forest plots, GRADE and Summary of Findings tables.

  1. Meta-Analysis
  2. Qualitative Synthesis
  3. GRADE Assessment
  4. Summary of Findings Tables
Projects/GRADE AssessmentEvidence ProfileSummary of Findings
  • High
  • Moderate
  • Low
  • Very low
HTMLExport

Summary of Findings4/7

OutcomeStudiesParticipantsEffect Estimate (CI)CertaintyRisk with comparatorAbsolute effectPlain Language Summary
ACR20 response at week 1252,068RR 2.32 (95% CI 1.94 to 2.77)Moderate281 per 1,000371 more per 1,000 (from 264 more to 497 more)The intervention probably increases ACR20 response at week 12.
DAS28 remission at week 2441,656RR 5.02 (95% CI 3.41 to 7.39)Moderate54 per 1,000217 more per 1,000 (from 130 more to 345 more)The intervention probably increases DAS28 remission at week 24.
Serious adverse events62,678RR 1.23 (95% CI 0.89 to 1.70)Low52 per 1,00012 more per 1,000 (from 6 fewer to 36 more)The intervention may result in little to no difference in serious adverse events.
Withdrawal due to adverse events62,678RR 1.58 (95% CI 1.04 to 2.40)Moderate31 per 1,00018 more per 1,000 (from 1 more to 43 more)The intervention probably increases withdrawal due to adverse events.
Certainty of evidence, rated outcome by outcome · demo data

Reporting & Maintenance

PRISMA manuscript draft, flow diagram and checklist.

  1. PRISMA Manuscript Draft
  2. PRISMA Flow Diagram
  3. PRISMA Checklist
  4. Living Review Dashboard
Projects/PRISMA Flow DiagramPRISMA 2020
PRISMA 2020 (default)AI transparency (PRISMA-trAIce)SVGPNGExcelPowerPointWord
IdentificationScreeningIncluded

Records from databases:

  • PubMed (n = 812)
  • Embase (n = 1,046)
  • Cochrane CENTRAL (n = 398)
  • Total (n = 2,256)

Records identified from:

  • Websites (n = 0)
  • Organisations (n = 0)
  • Citation searching (n = 0)

Records removed before screening:

  • Duplicate records removed (n = 684)
  • Records marked as ineligible by automation tools (n = 0)
  • Records removed for other reasons (n = 0)

Records screened

(n = 1,572)

AI-screened
1,572 (100%)
Human-screened
1,572 (100%)

Records excluded

(n = 1,498)

By AI
1,498
By human
1,492
AI agreement
97%

Reports sought for retrieval

(n = 74)

Reports not retrieved

(n = 3)

Reports assessed for eligibility

(n = 71)

AI-assessed
71
Human-assessed
71

Reports excluded

  • Wrong population (n = 18)
  • Wrong intervention (n = 14)
  • Wrong comparator (n = 9)
  • Wrong outcome (n = 11)
  • Wrong study design (n = 13)
  • Total (n = 65)
AI-excluded
62
Human-excluded
65

Studies included in review

(n = 6)

  • Quantitative synthesis (n = 6)
  • Qualitative synthesis (n = 0)
The PRISMA flow diagram builds itself and shows who decided · demo data

See the whole product

Academic Network

Research groups working with Puraite

Universities and public research institutes run their systematic reviews on Puraite. Academic access is free, with no time limit. Run as many reviews as you need.

  • Harvard UniversityUnited States
  • University of OxfordUnited Kingdom
  • National Research and Innovation Agency (BRIN)Indonesia
  • Universidad de AntioquiaColombia
  • Università degli Studi della Campania Luigi VanvitelliItaly
  • University of AberdeenUnited Kingdom
  • Wayne State UniversityUnited States
  • All India Institute of Speech and Hearing, MysuruIndia

See the Academic Network

Puraite's end-to-end systematic review platform has allowed us to maintain a high level of research momentum without compromising human control. It has streamlined our evidence synthesis workflow, allowing our team to process and manage over 10,000 records in an auditable, highly efficient workspace.

Researchers at over 150 institutions in over 40 countries

Puraite is built in Bielefeld and Paderborn, Germany. Research teams on every inhabited continent use it.

BielefeldHeadquartersPaderbornPaderborn University
Countries where researchers use Puraite · as of November 2026
  • Australia
  • Austria
  • Bahrain
  • Belgium
  • Brazil
  • Canada
  • Chile
  • China
  • Colombia
  • Cyprus
  • Denmark
  • Egypt
  • Germany
  • Ghana
  • India
  • Indonesia
  • Ireland
  • Italy
  • Jordan
  • Lebanon
  • Macao
  • Malaysia
  • Mexico
  • Morocco
  • Netherlands
  • New Zealand
  • Nigeria
  • Norway
  • Peru
  • Poland
  • Portugal
  • Romania
  • Russia
  • Saint Kitts and Nevis
  • Saudi Arabia
  • South Africa
  • Spain
  • Sweden
  • Switzerland
  • Thailand
  • Türkiye
  • United Arab Emirates
  • United Kingdom
  • United States

What your reviews cost in researcher time

Set your own numbers. The starting values come from a published cost study.

Your reviews
5
5
11
Assumptions
15 h
EUR 40
70%

Researcher time saved per year

EUR 500,480

12,512 researcher hours

Researcher hours per year

Today
17,875 h
With Puraite
5,363 h

The starting values follow Michelson and Reuter (Contemporary Clinical Trials Communications, 2019): five co-authors working 15 hours a week for 11 months, about USD 141,000 of researcher time per review. The share of time the AI takes over is your assumption. The Puraite licence is not included.

See Puraite on one of your own reviews.

Funded by the German Federal Ministry for Economic Affairs and Energy and the European Union through the European Social Fund Plus (ESF Plus). Implemented by Projektträger Jülich as part of the EXIST programme.