Under the Hood

How SatyaX Works

How SatyaX turns an article, headline, URL, or claim into an evidence-backed credibility assessment. Step by step. No black boxes.

🔬Evidence-first 🤖AI-assisted 🪟Transparent 🎯Honest about uncertainty

The 9-Step Analysis Pipeline

1
Input Collection
What can you submit?
Users can submit any form of informational content. The system accepts both direct text input and news article links for analysis.
📰 News article text 💬 Headline 📱 Social media post 🔗 Article URL 🎭 Satire article
Goal: Convert any piece of information into a standardized analysis pipeline.
2
Content Extraction
URL? We scrape the full article.
For URLs, SatyaX extracts the full article text using a 5-layer cascading scraper pipeline. For direct text, content moves immediately to the next stage.
Layer 1: Trafilatura Layer 2: newspaper3k Layer 3: BeautifulSoup Layer 4: Google Cache Layer 5: Archive.org
Only the article body is analyzed. Noise is removed before any AI processing begins.
3
Writing Style Analysis (BERT)
Linguistic pattern detection. Not truth detection.
A fine-tuned BERT model performs linguistic analysis — trained on 40,000+ real and fake news articles. Important: this module does not determine whether a claim is true or false. It identifies writing characteristics commonly associated with misinformation.
📊 Sensationalism Risk 🎣 Clickbait Probability ❤️‍🔥 Emotional Language 📰 Journalistic Similarity 🎭 Manipulative Patterns
Example: "Scientists HATE this one secret trick!"
→ High Clickbait Probability · High Sensationalism · Low Journalistic Similarity
4
Claim Extraction
Every verifiable claim, isolated.
The AI identifies individual factual claims made within the content and extracts them separately. Breaking content into individual claims allows more accurate, targeted verification instead of treating the entire article as one unit.
Input: "NASA confirms the Moon is made of cheese."

Extracted:
Claim #1 — NASA confirmed the Moon is made of cheese.
→ Each claim is verified independently in Step 7.
5
Real-Time Evidence Retrieval
Live web search. Not a static database.
The system searches the live web for supporting and contradicting evidence. This is not a pre-built fact database — results reflect current publicly available information.
📰 News reports 🏛️ Official announcements 🏗️ Government publications 🔬 Scientific sources 🌐 Public web
Goal: Find independent, real-time evidence related to each claim — both confirming and contradicting.
6
Source Reliability Assessment
Not all sources are equal.
Retrieved sources are evaluated for credibility based on domain reputation, editorial standards, historical reliability, and source authority. Each source is scored 1–10 and displayed alongside results.
✓ Trusted 9–10 ~ Moderate 5–8 ⚠ Low 1–4
reuters.com → 10/10 Trusted  |  bbc.com → 9/10 Trusted  |  unknown-blog.net → 4/10 Low Reliability
7
AI Fact Verification Engine
Evidence-based reasoning. Not style-based guessing.
The verification engine (powered by Google Gemini) cross-references the original content, extracted claims, retrieved evidence, and source quality to determine a verdict. Unlike traditional classifiers, the engine uses evidence-based reasoning.
✅ REAL
🚫 FAKE
⚠️ MISLEADING
🔶 PARTIALLY TRUE
💬 CONTEXT NEEDED
🔍 UNVERIFIED
8
Truth Score Generation
A credibility score from 0 to 100.
A credibility score is generated based on evidence strength, source quality, and claim verifiability. The score is a credibility indicator, not an absolute truth measurement.
80–100
High Credibility
Well-supported by evidence with strong source quality.
50–79
Mixed Credibility
Some valid information but lacks context or evidence.
0–49
Low Credibility
Major inaccuracies or unsupported/misleading claims.
9
Explainable AI Output
Not just a verdict. A transparent explanation.
The platform provides transparent reasoning. The goal is not only to classify content but to explain why a conclusion was reached. Users receive the full evidence trail.
📋 Verdict 📊 Truth Score 🎯 Confidence Level 🔎 Evidence Found 🎯 Key Claims 📄 Paragraph Analysis 🛡️ Source Reliability 🧒 ELI5 Explanation
What SatyaX Measures
Six dimensions of credibility, weighted by importance.
📝 Writing Style Analysis
Medium
Evaluates sensationalism, clickbait probability, and emotional language intensity via BERT.
🔬 Evidence Strength
High
Measures quantity and quality of supporting evidence found from real-time web search.
🛡️ Source Reliability
High
Evaluates credibility of retrieved sources based on domain reputation and editorial standards.
🎯 Claim Verifiability
High
Determines whether individual claims can be independently confirmed or refuted from public sources.
💬 Context Completeness
Medium
Checks if important contextual information is missing that would change the meaning of the content.
⚡ Contradictory Evidence
High
Identifies credible information that directly conflicts with the claims made in the content.
Architecture Overview
The complete data flow from input to verdict.
User Input
5-Layer Content Extraction (Trafilatura → BS4 → Cache → Archive)
📝 Writing Style Analysis (BERT)
🎯 Claim Extraction (Gemini)
🌐 Evidence Retrieval (DuckDuckGo Live Search)
🛡️ Source Reliability Analysis
🧠 AI Fact Verification Engine (Gemini Flash)
📊 Truth Score Generation (0–100)
✅ Explainable AI Results
What SatyaX Is Not
Honest limitations of AI-assisted fact analysis.
Not a fact-checking oracle
The platform provides AI-assisted evidence analysis and may occasionally make mistakes. Confidence levels indicate uncertainty whenever evidence is limited.
Not a replacement for experts
For medical, legal, financial, or safety-critical decisions, always consult qualified professionals. AI supplements — it does not replace — expert judgment.
Not censorship
The platform does not remove, report, flag, or suppress any content. It only provides an analysis when content is explicitly submitted by the user.
Not perfect
Every result includes confidence levels so users always know how certain the AI is. Low confidence results should be treated with extra skepticism.
🔒 Privacy First
What we process and what we don't.
✓ Only processes content you explicitly submit
✓ No social media accounts accessed
✓ No personal browsing history collected
✓ Only publicly available information used

Ready to verify something?

Paste any article, headline, URL or claim and get an evidence-backed verdict in seconds.

⚡ Start Verifying