> For the complete documentation index, see [llms.txt](https://factualai.gitbook.io/factualai-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://factualai.gitbook.io/factualai-docs/ai-fact-checking-engine.md).

# AI Fact-Checking Engine

#### Introduction

At the heart of FactualAI lies a powerful AI engine designed to **extract truth from chaos**. The decentralized and noisy nature of Web3 requires a system that can process natural language, access real-time data, and return meaningful conclusions—**instantly and intelligently**.

The FactualAI Fact-Checking Engine does just that.\
It reads, reasons, and responds.

***

#### 1. Natural Language Understanding

Users interact with FactualAI by submitting rumors in plain language. The system must interpret a wide range of phrasing—from rumors phrased as questions (“Is Solana partnering with Amazon?”), to speculative statements (“There’s a chance ETH ETFs will be approved next week”), to memes or slang.

To handle this variety, we use:

* 🧠 **Transformer-based NLP models** (BERT, RoBERTa variants)
* 📊 **Topic modeling** to identify relevant entities (project names, events, people)
* 🧩 **Intent classification** to distinguish:
  * Partnership rumors
  * Exchange listing claims
  * Exploit/security events
  * Token unlocks or market manipulation

This allows the engine to **normalize and categorize the rumor**, setting the stage for analysis.

***

#### 2. Multi-Source Data Aggregation

The AI Engine draws insight from a combination of **real-time and historical data sources**, including:

| Data Source                   | Purpose                                                   |
| ----------------------------- | --------------------------------------------------------- |
| 🔗 **On-chain data**          | Wallet activity, token movements, smart contract triggers |
| 📰 **News APIs**              | Press releases, headlines, article clustering             |
| 📱 **Social Media**           | Twitter/X, Telegram, Reddit – volume, velocity, sentiment |
| 📚 **Historical database**    | Previously validated rumors and their outcomes            |
| 👥 **Community voting layer** | Real-time fact-checker input and reliability scoring      |

Each rumor is **mapped to these data streams** to assess:

* Is the claim consistent with verifiable data?
* Have similar claims been true or false in the past?
* Is there unusual social or blockchain activity related to this rumor?

***

#### 3. Credibility Scoring

Once the data is collected, the engine calculates a **Credibility Score** using a multi-layer decision model:

* **Stage 1:** Data validation (timestamp checks, source consistency)
* **Stage 2:** Feature extraction (e.g., sudden wallet spikes, keyword co-occurrence)
* **Stage 3:** Confidence classification (True / False / Unclear)
* **Stage 4:** Score assignment (e.g., 0–100%, updated as new info arrives)

Credibility scores are **not binary**. Instead, they reflect the probabilistic nature of truth in decentralized environments. A claim may score 86% based on data now—and shift to 42% an hour later as a contradiction emerges.

***

#### 4. AI-Driven Investment Insight

FactualAI doesn’t stop at truth evaluation—it also acts as a **signal engine**.

Using the same data, it generates a real-time **Investment Suggestion**, classified as:

* 🟢 **BUY** (strong momentum or credible catalyst)
* 🟡 **HOLD** (observe; signal uncertain or neutral)
* 🔴 **SELL / AVOID** (low credibility, possible exit signal)

Signals are further augmented with:

* **AI confidence level** (Low / Medium / High)
* **Rationale summary** (1–2 sentence explanation)
* **Volatility indicator** (based on social and price data dispersion)

This creates a **usable output** for traders and decision-makers—not just information, but insight.

***

#### 5. Feedback Loop & Continuous Learning

Every interaction on the platform helps improve the engine.

* ✅ **User votes** become labeled data
* 🧠 **Model retraining** occurs weekly with new inputs
* 📉 **False predictions** are penalized in the engine’s scoring algorithm
* 🧪 **A/B testing** of different AI models ensures accuracy improvement

In time, the AI learns not just how to analyze facts—but how to **predict which types of rumors tend to be true**, and how markets are likely to respond.

***

#### Summary

FactualAI’s AI Engine is not just a detector of truth.\
It is a living, learning system that ingests unstructured chaos and outputs structured, actionable guidance.

Whether you're a crypto investor, builder, or researcher, the AI engine gives you an **unbiased second brain**—one that learns faster than the rumor mill can spin.

In the next chapter, we explore how this intelligence is tied to the platform’s native token:\
**$FCTAI — Truth, tokenized.**