How Heretix Checks a Claim

Heretix helps you see what leading AI models already believe about any statement you submit.
When you press “Ask the model,” here’s what happens behind the scenes.

⸻

Step by step

1 · Choose your lens You decide how the AI should think.
- Model-only: the AI answers using what it has already learned from its own training.
- Internet-informed: the AI also searches the web for new evidence before answering.
This choice lets you compare what the model believes on its own versus what changes when it sees new information.

⸻

2 · Check the claim from many angles Heretix rewrites your statement in several simple ways and asks the AI each version. This helps keep the answer balanced, so it’s not swayed by one particular wording.

⸻

3 · Combine the results Heretix looks at all the answers and blends them into one clear score—a number that shows how likely the AI thinks your claim is true. You’ll also see how confident the model is, so you can tell whether it’s certain or unsure.

⸻

4 · Show the reasoning Finally, Heretix explains why the AI leans the way it does. You’ll see:
- A short verdict such as “likely true,” “likely false,” or “uncertain.”
- A few plain-language points summarizing what shaped that belief.

⸻

What the percentage means

The number you see isn’t “truth”—it’s the AI’s belief.
A 90% result means the model thinks your claim is very likely true.
A 10% result means it thinks it’s very likely false.
And values near 50% mean the model is genuinely unsure.

Why this matters

By showing both the model’s internal belief and its internet-informed view, Heretix helps you understand how AI sees the world—and where it might be biased, outdated, or missing key facts.

Tips

Be clear and specific in your claim so the AI knows exactly what to judge.
Try both modes—Model-only and Internet-informed—to see how the AI’s view shifts.
Read the short explanation to decide what’s worth exploring next.