AI phrases checker.
Paste a draft. See every stock AI phrase, "not X, but Y" pattern and em dash in it, with a plain fix for each. Runs in your browser, nothing is uploaded, no signup.
Check your text
| Found | Times | What to do |
|---|
Nothing flagged. Either you wrote it, or you edited well.
What it checks.
Three kinds of tell, each scored separately. Light tells are common words AI overuses. Heavy tells are phrases and punctuation almost nobody writes by hand.
Stock phrases
Over 100 words and phrases that AI models reach for and people rarely do: "delve", "tapestry", "in today’s fast-paced world", "it’s important to note", "game-changing", "leverage" and the rest.
Sentence patterns
The shapes, not just the words: "it’s not X, it’s Y", "not just X, but Y", "no X, no Y, no Z", "whether you’re a founder or a marketer", and the one-word question that answers itself ("The result?").
Punctuation
Em dashes, en dashes used as a sentence break, and double hyphens. The em dash is the single clearest AI tell in long text, so it is counted on its own.
How to read the result.
The score is tells per 100 words. Under one is clean. One to three is noticeable to a careful reader. Over three and most readers will feel it, even if they cannot say why.
Fix the heavy tells first.
Every em dash becomes a comma, a colon or a full stop. Every "it's not X, it's Y" becomes the plain claim.
Then cut the filler.
"In today's fast-paced world" and "it's important to note that" delete cleanly; the sentence reads fine without them.
Then swap the overused words.
"Leverage" is "use". "Robust" is usually nothing. "Crucial" is "important", or cut.
Do not replace a flagged phrase with another one from the list. If no plain wording comes, cut the sentence.
Frequently asked questions.
No. An AI detector guesses whether a machine wrote the text and is often wrong. This checker does something narrower and more useful: it finds the specific phrases, patterns and punctuation that make text read as AI-written, so you can fix them. Clean human writing scores zero. Human writing full of corporate filler scores high, and should.
It is the banlist Pilcrino uses on every blog post it writes, plus the words readers and editors have called out as ChatGPT tells since 2023: "delve", "tapestry", "testament", "embark", "elevate", "seamless", "robust", and so on. The list lives in the page and you can read it in the page source.
No. The check runs in your browser with a few hundred lines of JavaScript. Nothing leaves the page, nothing is stored, and there is no account. You can turn off your network connection after the page loads and it still works.
Under one tell per 100 words reads as human. Zero em dashes, always. A 1,200-word post with three or four flagged words is fine once you fix them. If the count is in the dozens, the problem is the draft, not the edit: rewrite the sections, do not patch the words.
On its own, once, it is fine, and the checker says so by scoring it as a light tell. AI text uses these words far more often than people do, and the pile-up is what readers notice. Treat each one as a question: is there a plainer word, or can the sentence go?
Yes, if the tool that writes for you strips them before you see the draft. Pilcrino runs this check, and a fuller rewrite pass, on every post it writes, so what reaches you for approval has already been through it.