AI-slop detector
Paste a caption. We'll tell you how much it sounds like an AI wrote it.
What this tool does
There's a particular texture to AI-written copy and your reader's brain learned to spot it in 2023. The em-dashes everywhere. The hedge phrases ("it's important to note"). The grandiose verbs ("unleash", "navigate", "harness"). The sentences that all clock in at the same length. Once you see the pattern, you can't unsee it — which is bad news if you're posting that texture to your own audience.
How to use it
- Paste a caption, post, or paragraph into the box.
- We score it on six dimensions: AI-vocabulary density, em-dash count, hedge phrases, engagement-bait, sentence-length variance, and adjective load.
- Read the flagged phrases — those are the words a careful editor would cut first.
- Rewrite, or run it through the caption rewriter to strip the obvious tells.
Why it matters
Your audience is scrolling past dozens of AI captions a day. They're being trained — by repetition, not by reading think-pieces — to mentally tune out the texture. Posts that read human are starting to massively outperform posts that read generated, on the same accounts, with the same following. The fix isn't writing without AI. The fix is editing the AI texture out before you publish.
Questions people actually ask.
Does this actually detect AI, or is it heuristic?
Heuristic. There's no perfect AI detector and there never will be — modern models can write in any register. What we *can* do reliably is flag the specific textures that scream 'AI default voice' (overused verbs, em-dash spam, hedge cadence). That's enough to catch the obvious tells and most of the borderline ones.
My human-written post got a high score. Is that bad?
Not necessarily. Some humans naturally write with the same textures AI defaults to (lots of em-dashes, lots of hedges). The score is a flag, not a verdict. Look at the highlighted phrases and decide whether each one is doing real work or just adding texture.
Will my draft get sent to a server?
No. Everything runs in your browser. Nothing leaves your device, nothing is logged, nothing is used to train anything.