CRED-1: An Open Dataset Scoring Website Credibility from Multi-Signal Domain Features

> Paper: CRED-1: An Open Multi-Signal Domain Credibility Dataset for Automated Pre-bunking of Online Misinformation > Authors: Alexander Loth, Martin Kappes, Marc-Oliver Pahl > arXiv: 2604.20856 | 2026-04-28

---

1. A News Site That Is Hard to Distinguish as Fake

You see a link on social media with a sensational headline: "Scientists discover a cure for cancer!"

You click through. The site looks professional—there's a logo, navigation menus, author bylines.

But look closer:

• The domain was registered 3 days ago
• It is cited nowhere else
• Google Fact Check Tools show zero fact-check records
• Google Safe Browsing has flagged it as suspicious

This is a fake news site. But how would you know before reading the content?

---

2. Domain-Level Credibility Signals

CRED-1's core insight: judging truth at the content level is hard, but evaluating credibility at the domain level is comparatively easy.

A site's credibility can be inferred from multiple signals:

1. Domain age

• Trusted news organizations typically have years of history
• Fake news sites are often ephemeral—existing for days before vanishing

2. Web popularity

• Credible sites usually have substantial traffic and backlinks
• Fake news sites tend to exist in isolation

3. Fact-checking frequency

• Credible sites' content is frequently referenced or checked by fact-checking organizations
• Fake news sites usually never appear on the fact-checking radar

4. Threat intelligence

• Tools like Google Safe Browsing flag known malicious or fraudulent sites
• This is a strong signal

---

3. Pre-bunking

Traditional debunking is *reactive*: misinformation spreads first, then fact-checkers correct it.

But research shows that post-hoc debunking has limited effectiveness. Once people believe something, correcting it becomes difficult (the "belief perseverance" effect).

Pre-bunking takes the opposite approach: warn users that a source is unreliable before they encounter misinformation from it.

CRED-1's goal is to provide training data for automated pre-bunking systems.

---

4. Why an Open Dataset Matters

What does CRED-1 cover?

• 2,672 domains
• Multiple categories: news, health, politics, science, and more
• Four signals per domain: domain age, popularity, fact-checking frequency, threat intelligence

Why openness matters:

• Researchers can reproduce and verify results
• Developers can build better pre-bunking tools
• The public can audit the evaluation criteria

---

5. A Feynman-Style Judgment: Prevention Beats Cure

When teaching physics, Feynman always emphasized finding the fundamental question:

> "If you think you understand something but can't explain it to a beginner, you don't really understand it."

In the information ecosystem, the fundamental problem is not "how to correct misinformation" but "how to prevent misinformation from being believed."

Pre-bunking is built on exactly this insight: establish a defense before misinformation occupies the user's mind.

---

6. Takeaways

If you are building an information credibility system, ask yourself:

1. "Am I using multiple independent signals to assess credibility?" 2. "Is my assessment at the content level or the source level?" 3. "Do I provide preventive warnings, or only post-hoc corrections?" 4. "Are my data and algorithms transparent and auditable?"

CRED-1 reminds us: in information warfare, the best defense is not chasing every bullet, but identifying suspicious shooters before they fire.

When users see a credibility score for a source, they are already empowered to make informed choices.