Credibility data · AI content
How can you tell if a YouTube video is AI-generated?
Faceless channels and AI voiceovers are flooding your feed. You can hunt for the forensic tells — but the question that actually protects your time is whether the video says anything at all.
From our data
82 / 100
The average substance score across 73 of the most popular videos we tested. Production polish keeps climbing; substance is the part a credibility check actually measures — and it is what AI-generated filler is missing.
Source: the public Clarion Credibility Leaderboard (n = 73 videos), each scored 0–100 by the same automated Vibe Check anyone can run on a URL. Snapshot, methodology-first, aggregate-level.
AI has made it almost free to produce a confident, well-narrated ten-minute video about anything. That is why your feed keeps filling with faceless channels, synthetic voices, and slideshows of stock footage. The instinct is to ask “is this AI-generated?” — and there are real forensic tells you can look for:
- A synthetic narration. Text-to-speech voices are getting good, but many still have flat prosody, odd emphasis, or mispronounced proper nouns.
- No first-hand experience on screen. The video never shows the creator actually doing the thing — only generic B-roll over a script.
- Fluent but empty writing. Paragraphs that sound authoritative yet never cite a source, name a number, or make a claim you could check.
- Confident, wrong facts. Hallucinated dates, invented statistics, or a stitched-together summary that contradicts itself.
Here is the problem: detection is a losing game. Every month the voices get more natural, the scripts get tighter, and the tells get subtler. If your only defense is spotting the seams, you will lose that race. And it misdirects the question — a careful AI-assisted explainer can be genuinely useful, while a lazy human-made video can be a total waste. How a video was made matters far less than whether it says anything.
That is why we score substance instead of chasing provenance. Across the 73 popular videos on the Credibility Leaderboard, substance scores ran from 35 to 95 — a 60-point gap — and production value did not predict where a video landed. The average was 82/100, but the ones near the bottom shared a single trait: they were fluent and confident and said almost nothing you could act on. That is the real signature of AI slop, and it is measurable without a detector.
So the practical test is not “did a model write this?” It is “does this video make specific, checkable claims, or is it filler dressed up as expertise?” Answer that and the AI question mostly answers itself.
Check any video in one hover
You can run these checks by hand, but you won't do it every time. That is what Clarion is for. Hover any thumbnail and it reads the transcript and metadata, then shows a 0–100 credibility score, a Watch / Skim / Skip call and three takeaways — the same automated check that produced the number above, on every video you browse.
Get the next data drop
We refresh the Credibility Leaderboard and publish new findings regularly. One email when the next drop lands — plus a free month of Pro for founding users.
Frequently asked questions
Can you tell if a YouTube video is AI-generated?
Sometimes, from tells like a synthetic TTS voice, generic stock footage, a script with no first-hand experience, and confidently wrong or hallucinated facts. But detection gets harder every month as the tools improve, so it is a losing game to rely on. The durable test is substance: does the video make specific, checkable claims, or is it fluent filler? Empty content scores low whether a human or a model wrote it.
Does it matter if a YouTube video is AI-generated?
Only insofar as AI makes it cheap to produce confident, well-narrated videos that contain nothing. A carefully made AI explainer can be useful; a lazily made human one can be worthless. What matters is whether the content is accurate and specific — which is exactly what a substance score measures, independent of how it was produced.
What are the signs of an AI-generated YouTube video?
A monotone or subtly off text-to-speech narration, visuals that never show the creator doing the thing, a script that stays vague and never cites anything, padded runtime, and factual errors stated with total confidence. Any one can be a false alarm; together they usually mean the video was assembled to fill a slot, not to teach.
How does Clarion handle AI-generated videos?
Clarion does not try to run forensic AI-detection. It reads the transcript and scores how much genuine, checkable substance the video delivers on a 0–100 scale. AI-generated filler tends to score low because it is vague — so you get the practical answer (is this worth my time?) without depending on a detector that AI keeps outrunning.