AI Information - VerifyScience | EU AI Act Disclosure

What is VerifyScience?

VerifyScience is an AI-assisted research credibility screening tool that uses leading AI models to analyse research papers and provide credibility assessments. By default, analysis is performed by Claude AI (developed by Anthropic). Professional and Enterprise accounts using their own API keys may instead select OpenAI, Google Gemini, Mistral, or DeepSeek models. Every result states, in plain language, exactly which AI model produced it. Our service helps researchers, journalists, investors, and policy makers quickly evaluate the quality and reliability of academic research.

Purpose of Our AI System

VerifyScience uses AI to:

✓ Provide rapid preliminary assessment of research credibility
✓ Help users prioritize which research to review in depth
✓ Identify potential methodological issues or concerns
✓ Assist in literature review and research evaluation processes

VerifyScience does NOT:

✗ Replace peer review processes or expert human judgment
✗ Provide legal, medical, or professional advice
✗ Make definitive claims about research validity or truth
✗ Guarantee accuracy or completeness of assessments

How Our AI System Works

1

Input Processing

You submit a research paper via URL, PDF upload, direct text paste, or the Chrome browser extension (which extracts metadata and abstract directly from PubMed, bioRxiv, medRxiv, and arXiv). Our system extracts the text content and prepares it for analysis.

2

AI Analysis

The content is sent to the AI model that will run the verification (Claude AI by default, or the model the account has selected) with specific instructions to evaluate the research across standardized criteria. The system automatically detects the paper's research domain and methodology type, and calibrates the assessment criteria accordingly. The AI analyzes the text and generates scores and explanations.

3

Results Delivery

You receive a confidence score (0-100%), detailed breakdown across criteria, and a professional PDF report. All results are stored securely and can be accessed anytime.

What Our AI Can Do

✓ Methodology Analysis

Evaluates study design, sample size, control groups, and experimental rigor

✓ Citation Quality

Assesses relevance, credibility, and recency of referenced sources

✓ Data Quality

Reviews statistical validity, transparency, and reproducibility indicators

✓ Peer Review Status

Checks publication venue, review process, and editorial standards

✓ Detailed Explanations

Provides reasoning for each score with specific observations

✓ PDF Reports

Generates professional, shareable documentation of findings

Important Limitations

⚠️ Critical Understanding

Our AI system has important limitations. Please read these carefully before relying on verification results.

⚠️ Interpretive, Not Definitive

AI analysis is interpretive and probabilistic. Confidence scores are guidance, not absolute truth or final verdicts.

⚠️ Context Limitations

AI may miss field-specific nuances, domain expertise, or context that human experts would recognize.

⚠️ Variable Results

AI can produce different scores for the same paper due to the probabilistic nature of language models.

⚠️ Cannot Verify Facts

AI cannot independently verify the factual accuracy of data claims, experimental results, or measurements.

⚠️ Text Analysis Only

AI analyzes text content but cannot verify actual experiments, lab work, or physical evidence.

⚠️ Training Data Limitations

AI knowledge is based on training data up to January 2025. Newer research methodologies may not be fully recognized.

⚠️ Not a Replacement for Experts

AI assists with preliminary evaluation but cannot replace domain experts, peer reviewers, or professional judgment.

Your Responsibility as a User

As a user of VerifyScience, you should:

→ Use AI results as one input among many, not the sole basis for decisions
→ Verify critical claims independently using primary sources and expert consultation
→ Apply your own domain expertise and professional judgment to all assessments
→ Not rely solely on AI confidence scores for important research or policy decisions
→ Understand that AI provides screening assistance, not definitive validation
→ Consult qualified experts when dealing with critical or high-stakes research evaluation

Technical Information

AI Providers	Anthropic (default, all plans) OpenAI and Google available on Professional and Enterprise plans with the account's own API key; Mistral and DeepSeek additionally on Enterprise. Every result discloses which model produced it.
AI Models Used	Claude Sonnet (paid tiers, default) · Claude Haiku (free tier) Optional user-selected models: Claude Opus, OpenAI GPT, Google Gemini, Mistral, DeepSeek — availability depends on plan and the account's own provider keys. The exact model is stated on each result and in each PDF.
Training Data Cutoff	January 2025
Temperature Setting	0.3 (optimized for consistency)
Processing Time	Typically 30-90 seconds per verification
Data Retention	Results stored securely; deletable anytime via GDPR controls

Human Oversight & Control

You maintain full control over the verification process:

✓ You decide which research papers to verify
✓ You interpret the results using your expertise
✓ You make final decisions about research validity
✓ You control your data (download, delete anytime)
✓ AI assists with preliminary screening, you decide what matters

Data Privacy & Security

🔒 GDPR Compliant

Full compliance with EU data protection regulations. Download or delete your data anytime.

🛡️ Secure Processing

All data transmitted securely via HTTPS. API keys encrypted at rest (AES-256-GCM).

🔐 Access Control

Only you can access your verification history and results. No sharing without consent.

📋 Transparency

Clear information about what data is collected, why, and how it's used.

Read our full Privacy Policy →

Questions About Our AI?

We're committed to transparency about our AI system. If you have questions about how our AI works, its limitations, or how to best use VerifyScience, please contact us.

Contact Us Try VerifyScience

AI System Information