Free tools / AI Slop Checker
Free tool · em dashes, stock phrases, model habits
AI Slop Checker
Does this sound like AI? Paste your copy or a page URL. We mark every phrase people now read as machine-written, in place, and name what kind of problem it is: stock phrases, em dashes, the 'X, not Y' flip, teases, filler, and the words Claude, ChatGPT and Gemini each reach for. It's the gate our own writing passes before it ships.
Sample result
What a result looks like
A paragraph we wrote badly on purpose, checked live. The marks, the score and the categories are the tool's own output.
Scores 23 of 50, under the floor of 35 we hold our own copy to, with 24 flagged phrases and 1 structural pattern.
How it works
Our own content gate, turned outward
Every phrase rule we run on ourselves
61 plain-English phrases in 11 categories, read live from our list, plus the stock phrases and shapes built into the scorer. The whole list is published on the AI Slop Phrases page.
Five shapes, not only words
The 'X, not Y' flip, tail clauses, sentences opening on This is, possessive tails and em dashes. These are how copy sounds templated even with no banned word in it.
A score out of 50
Five readings of 0 to 10: directness, rhythm, trust, register and density. We hold our own published copy to 35.
Model habits, stated as habits
Three categories list vocabulary each model family is known for. We count them and tell you which way the text leans. People use those words too.
Reference
What the checker flags, with examples
| Category | Example | Why it reads as machine-written |
|---|---|---|
| Stock phrases | in today's, game-changer, moreover | Filler every style guide bans, and models reach for by default |
| Em dashes | the dash between two clauses | Not wrong, and readers now associate it with generated text |
| The binary contrast | it's not a tool, it's a system | Frames the point against a strawman nobody raised |
| Withheld tease | what they don't tell you | Promises a payoff instead of giving it |
| Pause for effect | let that sink in | A stage direction where a point should be |
| Claude vocabulary | tapestry, testament to, delve into | Words Claude models are known to overuse |
| ChatGPT vocabulary | furthermore, crucial role, in summary | Words ChatGPT models are known to overuse |
| Gemini vocabulary | here's the deal, let's break it down | Phrases Gemini models are known to overuse |
What to do with a result
What each result means, and what to do about it
| If the result says | It means | Do this |
|---|---|---|
| Score 35 or more, nothing marked | The copy clears the floor we hold our own writing to. | Ship it. Read it aloud once anyway; no checker hears tone. |
| Score 35 or more, phrases marked | The rhythm is fine and a few phrases give it away. | Fix the marked phrases. Each one names its category, so the rewrite is specific. |
| Score under 35 | Hedges, flat rhythm or stock phrasing run through the piece. | Fix the lowest of the five readings first. Directness and rhythm move most. |
| Em dashes marked | Readers now spot the dash as a machine habit. | Use a comma, a colon or two sentences. |
| Leans toward one model's habits | The vocabulary matches what we catalogue for that family. | Swap those words for plainer ones. It says nothing about who wrote the text. |
Is this an AI detector?
No. An AI detector claims to tell you whether a model wrote a text, and none can do that reliably. This tool claims less and delivers it: it marks the phrases and shapes readers now associate with machine writing, wherever they came from. A person can write every one of them, and a model can avoid them all.
It also doesn't predict whether an engine will cite your page. Cleaner copy is easier to read and easier to trust. Whether an engine names you depends on what the page says, whether it answers a buying query, and who else backs it up.
Clean copy is the floor. It can't tell you whether AI names you.
A free scan puts real buying queries to all five AI engines and shows who they name instead of you.
Check your AI visibility free →FAQ
Frequently asked questions
Does this tell me if AI wrote my text?
No. It reports phrases and shapes that read as machine-written, and the vocabulary habits we've catalogued for Claude, ChatGPT and Gemini. People use all of them. We describe the words on the page, never who or what wrote them.
Why are em dashes flagged?
The em dash is correct punctuation, and readers now treat it as a sign of generated text because models use it constantly. We flag it so you can decide. A comma, a colon or two sentences usually reads cleaner.
Where does the phrase list come from?
From the gate we run on our own blog, our customers' drafts and our app screens. It started with the published stop-slop ruleset and grows whenever we hear a phrase that gives copy away. The whole list is on the AI Slop Phrases page and updates there as it changes.
What does the score out of 50 measure?
Five readings of 0 to 10. Directness counts hedges and why-or-what openers. Rhythm rewards varied sentence length. Trust counts stock phrases and em dashes. Register reads adverbs, passive voice, formality and This-is openers. Density checks sentence length sits near twenty words.
Why is 35 the target?
It's the floor we hold our own published writing to. It's our standard, set on our own posts. No search engine publishes a number like it.
Does a better score get me cited by AI?
We won't claim it does, and our scorer's own documentation forbids the claim. Clean copy removes a reason to distrust a page. Being cited depends on answering a real buying query and on independent sources backing you.
Do you store what I paste?
No. The text is scored in memory and kept for fifteen minutes so a refresh doesn't re-run it, then dropped. Nothing is saved to an account.
Can I check a whole page instead of pasting?
Yes. Switch to Page URL and we read the page's main text, skipping navigation, headers, footers and forms.
Sources
Sources and further reading
- stop-slop: the published ruleset our checker started from
- AI Slop Phrases: the full list this checker runs, live