Speedwrite AI writing guides
Chatbot writing has habits: the same few words, the same openers, "it's not just X, it's Y", lists of three and em dashes everywhere. Readers have learned to spot them. Paste any text below and our free checker highlights each habit with a fix, then read on for the twelve signs, how to judge them fairly, and how to make AI writing sound like a person wrote it.
Sentence lengths
This isn't an AI detector, and nothing can prove who wrote a text. It counts habits that make writing read like a chatbot, which people use too.
Chatbots reach for words that sound polished and say little. One is fine. Five in a paragraph is a pattern.
"In today's fast-paced world", "When it comes to", "Let's dive in" and "I hope this email finds you well" open; "In conclusion", "Overall" and "I hope this helps!" close. People who write for a living cut both and start with the point.
The single most recognizable chatbot frame. It sets up a claim nobody made so it can knock it down.
BrewBox isn't just a coffee subscription. It's a journey.
"Fast, reliable and affordable." "Plan, execute and measure." One list of three is good rhythm. When every sentence ends in one, the text sounds generated.
Chatbots use the long dash (—) far more than most people do, often several times a paragraph, to bolt on asides. See the em dash question below before you judge anyone for one.
Human writing speeds up and slows down: a long sentence, then a short one. Chatbot paragraphs tend to march along at one steady length, and so do the paragraphs themselves. Our checker draws your sentence lengths so you can see it.
"A rich and vibrant community." "Valuable insights." "A wide range of options." Real writers name the thing: which community, what insight, which three options.
"It's important to note that", "it's worth mentioning", "can help to", "plays a crucial role in". Delete them and the sentence usually gets better.
A short email doesn't need "In summary" and a restatement of everything above it. Chatbots add one by reflex.
Bold words mid-sentence, Heading: lines in an email, ### marks or ** stars pasted from a chat window, and emoji bullets (✅ 🚀 ✨) in places nobody uses emoji.
"Great question!", "Absolutely!", "Certainly!" and "Feel free to reach out" sound like a helpful assistant, because that's what wrote them.
No typos, no opinions, no names, dates, numbers or personal details. Smooth sentences that could describe any company or any trip. That's often the strongest sign of all, and no checker can measure it for you.
Look for several signs together, not one. A single "delve" or em dash proves nothing; a text with the stock opener, three "crucial"s, a "not just X, it's Y", same-length sentences and a summary at the end probably came from a chatbot with little editing.
Then be fair about it. People wrote like this before chatbots existed, and chatbots learned it from them. Writers in a second language often use formal, textbook phrasing that gets mistaken for AI. A Stanford study found seven popular AI detectors flagged more than half of essays by non-native English writers as AI-written, and OpenAI withdrew its own detector in 2023 for being too inaccurate. Treat signs as a reason to ask, not as proof.
Wikipedia's editors keep a long, careful list of these patterns for spotting AI-written articles; it's worth reading if you review a lot of writing.
Not on its own. Writers have used em dashes for centuries, and plenty of good writers love them. What changed is that ChatGPT and other chatbots use them constantly, often two or three per paragraph, so readers now notice them. One dash in a long piece means nothing. A dash in nearly every sentence of a short email, along with other signs on this page, is a tell. If you want to avoid the question, use a comma, a colon, parentheses or a new sentence instead.
A prompt that helps
Rewrite the text below so it sounds like a person wrote it. - Keep every fact, name and number. Don't add new claims. - Use plain, specific words. Mix short sentences with longer ones. - No em dashes, no lists of three, and no "it's not just X, it's Y". - Don't open by setting the scene, and don't end with a summary. - Don't use these words: delve, tapestry, testament, pivotal, crucial, seamless, robust, elevate, realm, landscape, navigate, foster, unlock, harness, moreover, furthermore, additionally. Text: [paste your text here]
Words like delve, tapestry, testament, pivotal, crucial, seamless, robust, elevate, realm, landscape, navigate, foster and unlock, plus transitions such as moreover, furthermore and additionally. Any of them is fine once; a cluster of them is what readers notice.
No. It highlights writing habits that make text read like a chatbot and measures how evenly your sentences run. It can't tell who wrote something, and neither can any detector with certainty. Use it to edit your own writing, not to accuse someone.
Sometimes, but not reliably. They miss lightly edited AI text and wrongly flag human writing, especially from non-native English writers. OpenAI shut down its own classifier because of its low accuracy. Turnitin says its document-level false positive rate is under 1%, but sentence-level flags are less certain.
Mostly. Each model has its favourite words, but they share the stock openers, lists of three, tidy summaries and even rhythm, because they learned from similar writing and are tuned to sound helpful and safe.
Millions of people ask the same few models for the same kinds of writing, and each model picks the most likely next words. The result is a safe, average voice. That's why Speedwrite gives you a different rewrite every time instead of one "correct" version.
From the makers of this page
Drafting with ChatGPT or Claude? Make it sound like you.
Speedwrite rewrites AI drafts one paragraph at a time into fresh, natural writing, with a new version every time you press Predict.