YouTube is filling up with AI-made videos and content-farm channels. Legit reads a whole channel, right in your browser, and answers two questions: is the content AI-made, and is it run like a content farm?
Channel-level · Multi-signal · Independent.
A single clip might pass as human. Twenty uploads in, the pattern gives it away.
Legit reads the whole channel: how it writes, how it sounds, how it looks, how it acts, how it's run. Then it judges every new upload against everything before it. A one-shot detector sees a single video in isolation. We see the fingerprint across all of them. Much harder to game.
You still get a read on the video you're watching. It just arrives with the whole channel behind it, which is what makes it hard to fool.
Over 70 identifiers roll up into five signals. Each one answers a question about the channel, and each question gets its own read, shown on its own so a clear answer never gets averaged away by the other.
Transcripts, titles and descriptions, read for patterns consistent with AI writing.
Cloned voices, synthetic narration, robotic TTS. Often fills in over time.
Thumbnails and frames, checked for the fingerprints generative models leave behind.
Upload cadence, comment patterns, community signals.
Channel age, verification, history and public metadata.
Two reads, kept separate. We don't boil a channel down to a single score, so a clear answer on one question never gets hidden by the other.
When both reads land at the same end, AI-made content on a farm-like channel, that combination is what we call AI Slop. It's the two reads agreeing, not a separate score.
Some signals come back clear. When one is unclear, we say Mixed Signals. When one is missing, we say that too. We never fill a gap with a guess, or read an unknown as human.
We've spent close to a year studying YouTube, AI and detection to build this, and we're proud of it. But no tool gets it right every time. Ours included. Anyone who claims they do is a liar.
So we don't pretend otherwise. And unlike YouTube, we don't decide which channels and creators should live or die based on imperfect tools. Legit shows the evidence behind every call, and says "Mixed Signals" when it isn't sure instead of guessing. Think we got one wrong? Tell us. Your flags make the next read sharper.
We do the detection ourselves, then let you keep us honest. We're starting with a simple thumbs up or down on each score, right in the extension, so you can flag a call you think we badly missed. Over time that grows into a fuller community reporting system that sharpens our accuracy, and eventually lets you filter out the kinds of AI content you don't want to see.
YouTube does auto-label some AI now, but mostly when it can see a watermark or provenance tag from its own tools, Veo, or its partners, or spot obvious photorealistic AI. Use a different generator, strip the watermark, let AI write the script, or simply run a low-effort farm, and there's nothing to label.
And there's a conflict of interest baked in: YouTube is owned by Google, the same company building the AI tools creators use, like Veo and Gemini. It earns ad revenue when AI content runs, won't let you filter AI out of your algo, and sets the labeling threshold inside a black box no one can audit. The safe move for its business is to under-report AI, which is exactly why an independent read matters.
Legit doesn't wait for a watermark. We read what's public across a whole channel, and we answer the second question too: is this run like a content farm? YouTube's label is a data point, not the answer.
Identifying AI is step one. The bigger goal is putting you back in control of what actually reaches your feed.
Viewers deserve to know.
The part we're building now. Legit reads any channel across five signals and two questions, right in your browser, and shows you the receipts behind every call.
Live at launchViewers deserve to choose.
Once it's identified, you decide what to do with it. Hide, blur, or down-rank AI Slop across your feed, on your terms. Set it once and stop scrolling past channels you never wanted in the first place.
Coming soonno spam · you'll hear from us when we launch