Does my writing sound like AI?

Writing sounds like AI when it leans on a small set of habits readers now spot at once: stock words such as “delve”, sentences that all run to the same length, emoji bullets, and claims with no number or name behind them. One habit is normal in human writing. Several in a short text is what makes a buyer decide a machine wrote it and stop reading. This checker looks for 129 of these patterns in 11 groups, highlights each one in your own text with what to write instead, and scores the text as weighted tells per hundred words. It runs in your browser, so nothing you paste is sent anywhere.

No signup. The result is yours before we ask for anything.

Run it on your own writing

What you are checking

Your text

Or try oneChecked in your browser. Nothing is stored, and no AI reads it.

Paste a draft to see its tells

129 patterns, weighted by how strongly each gives a text away. A short text is scored as if it had sixty words, so one tell in a tweet does not decide it.

Quill and Relay · AI agents

Drafts in your voice, checked against this list before you see them

Quill writes your posts, and Relay your cold emails, from what you sell and how you write. Every draft is held to this list and to the rules you add, and a draft dense with tells is written again before it reaches you. Nothing goes out until you approve it.

Hire GTM agentsInstinctGTM builds your GTM plan first, free, from your website.

How this works

1129 patterns in 11 groups: words, stock phrases, sentence shapes, rhythm and structure, imagery, substance, formatting, chatbot residue, posts, cold email and landing pages.
2Each pattern carries a weight from one to three. A word a person might well use counts one; a line only a chatbot leaves behind, such as a placeholder still in brackets, counts three.
3The score is the weighted total per hundred words. A text under 60 words is scored as if it had 60, so one tell in a short post is not a verdict.
4Four bands: under 1.5 reads like a person, under 4 has a few tells, under 8 reads as generated, and 8 or more reads as machine-made.
5The score from 0 to 100 is the same density on a scale where each band takes a quarter: under 25 reads like a person, and 75 and over, a density of 8 or more, reads as machine-made.
6The cleaner version comes from fixed rules, not a model. They cut stock openers and lines an assistant leaves behind, swap a word such as “leverage” for “use”, turn em dashes into commas and emoji bullets into dashes, and trim a wall of hashtags. Each change is listed, and anything that needs judgment is left for you.
7Some checks apply to one kind of writing only. A cold email is checked for length, for more than one ask and for more I and we than you. A post is checked for a wall of hashtags and a link in its first line.
8Everything runs in your browser. Nothing you paste is stored or sent, and no AI model reads it.
9It is the same list InstinctGTM holds its own writing agents to. Every draft is checked against it before the founder sees it, and a draft dense with tells is written again.

What it cannot tell you

It does not tell you whether a person or an AI wrote a text. People write every one of these patterns, and a model can be told to avoid them all. It measures how the text will read, not where it came from.
It reads patterns, not meaning. A text can pass every check and still say nothing, or carry a claim that is wrong.
It reads English only. A text in another language gets no useful score.
A pattern can fire on a word used plainly, such as a journey that is an actual trip. Read the list as a guide to what a reader will notice, not as a verdict.

Questions people ask

What is AI slop?

AI slop is text that reads as generated: stock phrases, inflated words, an even rhythm and claims with nothing behind them. The term describes how a text reads rather than who wrote it, which is why a person can write slop and a model given the right rules can avoid it.

Can this tell if a post was written by ChatGPT?

No. Tools that claim to name the author guess from statistics a short post does not have enough of, and they misjudge people who write formally. This checker does something narrower and checkable: it shows the exact phrases that make a reader think a machine wrote the text, so you can change them.

Which words make writing sound like AI?

The words that turn up far more often in machine text than in a person’s: “delve”, “tapestry”, “pivotal”, “leverage”, “seamless” and “elevate” are the best known. Shapes give a text away faster than words, though. Sentences of one even length, lists that always come in threes and a question answered in the next sentence are noticed before any single word.

Why does sounding like AI matter in a cold email or a post?

Because a buyer decides in the first line or two. Buyers now get machine-written messages every day, and once one reads as generated they stop looking for the reason it was sent. A message that sounds like one person who knows something specific gets read to the end, which is the only way it gets a reply.

How do I make AI writing sound human?

Cut the stock phrases, then add what only you know: a number from your own work, the kind of customer you mean, a date. Vary sentence length, write in paragraphs rather than a line break after every sentence, and end on one real question or none. Then check the text again and fix what is left at the top of the list.

Is one AI word enough to give a post away?

No. One tell is normal in human writing, and plenty of people write “crucial” without help from a model. What gives a text away is density, several tells in a short text, which is why the score here is weighted tells per hundred words rather than a list of forbidden words.

The short version, if you would rather not run it

Buyers have read thousands of machine-written posts and emails by now, and they recognise the habits before they reach the point. The reaction is fast and unfair: a post with a real insight in it still gets skimmed if it opens on a stock phrase, because readers have learned that stock openers rarely go anywhere.

The habits come in a few kinds. There are words a model reaches for, and shapes such as the not-this-but-that contrast, the dramatic colon and the list of three. Layout gives a draft away too, with emoji bullets and a line break after every sentence. Absences count as well: a long text with no number, name or date in it reads as written about nobody.

None of these is wrong on its own. People write them too, which is why a list of banned words fails: it flags good writing and misses the shapes. Counting tells per hundred words, with the strongest weighted higher, is closer to how a reader reacts.

For a founder doing their own go-to-market, this matters most where attention is shortest: the first line of a post and the first two lines of a cold email. InstinctGTM holds its own writing agents to this list, plus any rules the founder adds, before a draft reaches them.