Sharp. Fair. Verifiable.
Follow the claim.
Inspect the evidence.
The process is automated in places. The limits should always be visible.
Quick answer
How does FactCheckJoe fact-check Joe Rogan?
We pull the transcript of each Joe Rogan Experience episode, extract the checkable claims, research each one against published sources, and label it false, distorted, one-sided or unverified. Every check links the moment in the video and the evidence behind the finding, and it says when the speaker hasn't been confirmed or no source was captured. Unverified means we couldn't settle it, not that it's false. Accurate claims aren't listed, and the Jamie Meter is a plain count of misses, Joe's own once we've matched the voices, not a percentage or an honesty score. We're not against Joe. We're for checking.
1. Identify a checkable assertion
The worker discovers episodes, obtains transcripts and extracts factual assertions. Context screening aims to preserve who said what, qualifications and hypotheticals. It can still make mistakes. A transcript summary is not a verified quotation.
2. Research the specific claim
The worker attempts to research claims using web grounding. Availability and source quality vary; some records come from fallback passes. A source link alone does not prove that the source supports the conclusion. Each claim page shows the evidence captured and whether editorial context review is recorded.
3. Explain the finding and its limits
False, distorted and one-sided are different findings. Unverified means we cannot establish the answer from the recorded evidence; it is not a false verdict. The public list intentionally focuses on flagged claims and omits accurate claims. It is not a complete list of what was said or a measured falsehood rate.
Attribution stays unconfirmed until independently documented. We label exact quotations separately from summaries, give credit for verified on-air corrections, and distinguish one curated video check from a full episode audit.
The Jamie Meter
Our familiar shorthand counts false, distorted and one-sided findings. It is a count, not a percentage or a judgment of someone’s honesty. Unverified claims do not count as misses. Fewer than 5 rated claims means no meter. Extraction and selection affect the number.
We also match each claim to the voice that said it. NVIDIA’s open Nemotron 3 Diarization model separates the speakers in the episode audio, and we start every run with known clips of Joe and Jamie so it can tell them from the guest. On an episode where the voices separate cleanly, the meter counts only Joe’s own misses, and the guest’s are listed separately. A claim goes to a speaker only when one voice clearly owns the moment it was said. Otherwise it is marked as not matched and counts for no one. Episodes we haven’t matched yet still count every speaker’s misses, and their page says so. A speaker we confirmed by hand always overrides the voice match.
- Legit · up to 5 misses
A handful at most. Jamie barely had to move.
- Entirely possible · 6 to 12 misses
A few needed a second look. About what three hours of talking gets you.
- Look into it · 13 to 20 misses
Enough got stretched that you'll want the sources open.
- That's crazy · 21 to 30 misses
A miss every ten minutes or so. Jamie was busy.
- Have you ever tried DMT? · 31 or more misses
Off the charts. More than thirty things that didn't hold up. Jamie needs a nap.
4. Keep the record open
Material revisions should preserve the previous finding, the new finding, the date, the reason and supporting evidence. Our initial revision log does not reconstruct edits to older checks. It says so rather than pretending a complete history exists.
The automated checks can misread context, omit sources or reach the wrong conclusion. A check with no recorded human context review should not be mistaken for one that has received it.
Read revisions or challenge a finding →