What is the most common sign of AI writing?

Overuse of the em dash. AI models reach for the long dash as an all-purpose connector, often several times per paragraph, far more than a human writer would. Curly quotes and stock phrases like "it is important to note" are close behind.

Why does AI use so many em dashes?

The long dash is a flexible connector that fits almost anywhere, so a model trained to pick the safe, probable option reaches for it constantly. A human varies punctuation more, so a high density of em dashes is a strong tell.

Is it wrong to use words like "delve" or "leverage"?

Not in isolation. The problem is density. When several of these buzzwords cluster in one paragraph the writing feels generic, and swapping them for plain words almost always improves it.

Does cleaning up AI tells defeat AI detectors?

No, and that is not the goal. These are editing tools that make text read better and sound like you. They make no claim about any AI detector and are not designed to evade detection.

What is the fastest way to clean an AI draft?

Paste it into the AI Text Cleaner to remove em dashes, smart quotes, Markdown and assistant lines in one pass, then read it aloud and cut the stock phrases you hear.

AI Text Cleanup

How to Clean Up AI-Generated Text

AI writing tools are fast, but their output has an accent. Once you know the tells you cannot unsee them, and neither can your reader. This guide lists the punctuation, phrases and structures that mark a draft as machine-written, shows what to replace them with, and gives a quick workflow for turning a ChatGPT draft into text that sounds like you wrote it.

The clearest signs of AI writing are overused em dashes, curly quotes, stock phrases like "in today's fast-paced world" and "it is important to note", a rule of three in every sentence, and leftover assistant lines like "Certainly! Here is". Remove those, vary your sentence length, and add a specific detail, and the text reads as human.

Why AI text has a "tell"

A language model is trained to produce the most probable next word, and that pulls everything toward a safe, average, slightly formal register. The result is fluent but generic, and it leans on the same small set of connectors, openers and flourishes again and again. None of these moves is wrong on its own. The giveaway is the density: the same handful of habits repeated in every paragraph, with none of the friction and surprise of a human writer making choices.

The good news is that the tells are consistent, so they are quick to spot and quick to strip. Most of them fall into four groups: punctuation, phrasing, structure and leftover assistant lines.

Punctuation tells

The fastest signals are in the punctuation, because they survive copy and paste unchanged:

The em dash. AI reaches for the long dash as an all-purpose connector, often two or three times in a single paragraph, where a human would use a comma, a full stop or a colon. This one habit is the most common giveaway of all, which is why a dedicated em dash remover exists. You can also replace each one with a comma or with a colon.
Curly quotes. Pasted AI text usually carries smart quotes, the slanted curly version of quote marks and apostrophes, which clash with a site or document set in straight quotes. Straighten them with the smart quote remover.
Leftover Markdown. Text copied from a chat often arrives full of asterisks for bold and hashes for headings. Strip them with remove Markdown formatting.

Phrase and word tells

Certain phrases turn up in AI writing far more often than in human writing. These are the ones to search for and cut:

Stock openers: "In today's fast-paced world", "In the ever-evolving landscape of", "In an era where", "Picture this".
Empty hedges: "It is important to note that", "It is worth noting", "Needless to say", "At the end of the day", "When it comes to".
Buzz verbs and nouns: delve, leverage, navigate, foster, underscore, harness, streamline, plus the nouns tapestry, testament, realm and landscape, and the adjectives robust, seamless, vibrant and game-changing.
The closing flourish: "In conclusion", followed by a paragraph that restates the introduction without adding a single new idea.

None of these words is banned. The problem is the pile-up. If one paragraph contains "delve", "robust" and "leverage", a reader senses the machine even if they cannot name why. Swap each buzzword for a plain one: "use" beats "leverage", "handle" beats "navigate", "look at" beats "delve into". The AI text cleaner and the text tone cleaner flag and strip the most common offenders for you.

Structural tells

Beyond individual words, AI writing has a recognisable shape:

The rule of three, everywhere. "Clear, concise and compelling." "Plan, draft and refine." One triple is fine. A triple in every sentence is a fingerprint.
The "not only X but also Y" sandwich, used over and over.
Transition overload, where every paragraph opens with Moreover, Furthermore or Additionally.
Flat rhythm. Human writing mixes short, punchy sentences with longer ones. AI tends to produce sentences of similar, medium length, which reads as smooth but lifeless.

The fix is to read the draft aloud and break the pattern on purpose. Cut one item from every triple, shorten some sentences and merge others so the rhythm varies, and delete any transition word that is not doing real work.

Assistant leftovers

The most obvious tell of all is text the model wrote about the task rather than the topic. These lines should never reach a reader:

Openers: "Certainly!", "Sure, here is a draft", "Absolutely, here is".
Sign-offs: "I hope this helps!", "Let me know if you would like me to adjust anything", "Feel free to tweak".
Disclaimers: "As an AI language model, I".

Remove the opening prefaces and the closing lines in one click, or clear all assistant scaffolding at once.

A quick clean-up workflow

To turn a raw AI draft into finished text, work in this order:

Paste it into the AI Text Cleaner to strip Markdown, em dashes, smart quotes and assistant lines in one pass.
Read it aloud and cut every stock opener and empty hedge you hear.
Break the rhythm: shorten some sentences, merge others, and trim each rule-of-three down to the one item that matters.
Replace the buzzwords with plain words.
Add one specific detail, number or example that the model could not have known. A concrete specific is the surest sign of a human.

For targeted fixes there are focused tools: clean ChatGPT text, remove emojis, and the full AI text cleanup set. These are editing tools that improve how text reads. They are not detection-evasion tools and make no claim about any AI detector.