Does this sound like AI?

Paste a post and see the habits readers link to AI writing: “not just X, but Y”, delve and the other AI words, em dashes, emoji bullets and more. Each one comes with why it stands out, what to write instead and where the list comes from. It's a second read, not a verdict.

Runs in your browser. Your text and images never leave it.

Updated Oct 20, 2026

Your post

What we noticed

Paste or type a post to check it.

AI phrases to avoid, and what to write instead

All but the last one are on the list Wikipedia's editors keep of signs of AI writing; hidden characters are here because chat text can carry them. Each has a plainer way to say the same thing.

Habits the checker looks for
Habit	Looks like	Try instead
“Not just X, but Y”Sentence shapes	It's not a tool, it's a movement.	Say the Y part on its own. If the contrast matters, say who thought X.
Lists of three, again and againSentence shapes	Fast, simple and friendly. Clear, calm and kind.	Keep the one item you can back up, or list as many as there really are.
A clause tacked on with “-ing”Sentence shapes	We rebuilt the docs, highlighting our commitment to users.	Stop at the comma, or give the tail its own sentence with a real detail in it.
Big claims about importanceWord choice	This launch stands as a testament to a pivotal moment.	Show what changed, with a number or a name, and let the reader decide it's big.
“Serves as” instead of “is”Word choice	The page serves as a guide for new teams.	Write is, are or has.
Words chatbots overuseWord choice	Let's delve into the rich tapestry of ideas this showcases.	Use the word you'd say out loud: look at, show, mix, important.
Brochure wordsWord choice	A groundbreaking café nestled in the heart of town.	Say what's actually there: the street, the price, the thing people order.
“Experts say” with no expertWord choice	Experts say consistency is key.	Name who said it and link to it, or say it's your own view.
“It's important to note”Word choice	It's important to note that results may vary. In conclusion, try it.	Delete the signpost and start with the point.
Lots of boldFormatting	𝗕𝗶𝗴 news: our 𝗻𝗲𝘄 𝗳𝗹𝗼𝘄 is 𝗹𝗶𝘃𝗲.	Let one line carry the weight. Put the key point first instead of bolding it.
Emoji as bullet pointsFormatting	✅ Faster setup ✅ Clearer steps ✅ Better support	Use plain lines or a short sentence. Keep an emoji where it says something.
Headings In Title CaseFormatting	Our New Onboarding Flow Is Live	Drop the heading, or write it like a sentence: Our new onboarding flow is live.
Em dashesFormatting	We shipped it — and it changed everything.	Try a comma, a full stop or brackets. Keep it if it's how you write.
Chat leftoversLeft over from a chat	Certainly! Here's a polished version of your post. I hope this helps!	Cut them, and fill in any [placeholder] before it goes out.
Hidden charactersLeft over from a chat	A zero-width space between two words	Take them out with the invisible character remover.

Source: Wikipedia: Signs of AI writing (WikiProject AI Cleanup advice page)

AI words to avoid, and when they were in fashion

The words change with each model, so a list from 2023 is half out of date.

Wikipedia's editors group them by era. Delve, tapestry, testament, pivotal, underscore, intricate, meticulously and vibrant were the 2023 to mid-2024 words. Fostering, showcasing, highlighting, enhance and align with took over from mid-2024 to mid-2025. Delve, the famous one, dropped off sharply in 2025.

The research behind those lists counts words, not opinions. In more than 15 million biomedical abstracts, Kobak and colleagues found delves used 28 times as often in 2024 as earlier years predicted, underscores 13.8 times and showcasing 10.7 times. Of the 379 style words in excess that year, 66% were verbs: the change was in how people wrote, not what about.

Juzek and Ward found 21 words like it, delve, intricate and underscore among them, and their tests are consistent with the human feedback step of training playing a part. The same words turn up outside science too: Liang and colleagues estimate 6.5% to 16.9% of peer reviews at four AI conferences in 2023 and 2024 could have been substantially changed by a chatbot.

We flag the rarer words. Crucial, key, robust and valuable are on the lists too, but people use them so often that a check for them would light up every post.

Sources: Wikipedia, Signs of AI writing: High density of "AI vocabulary" words; Juzek and Ward (2025), Why Does ChatGPT “Delve” So Much? COLING 2025; Kobak et al. (2025), Delving into LLM-assisted writing in biomedical publications through excess vocabulary, Science Advances; Liang et al. (2024), Monitoring AI-Modified Content at Scale, ICML 2024

Em dashes and AI

The em dash got famous as an AI tell. It's a weak one on its own.

Wikipedia's editors note that chatbots use far more em dashes than non-professional writers do, in places a person would put a comma, brackets or a colon, often to punch up a line like sales copy, and usually with spaces around them.

They also file it under historical signs, from 2022 to September 2026. Because it became notorious, AI companies started making newer chatbots hold back on it, OpenAI's GPT-5.1 most of all, and a 2026 study cited there found ChatGPT now uses fewer em dashes than professional writers.

So one em dash proves nothing either way. If you like them, keep them. If there's one in every sentence, and other habits from the list around them, that's when it's worth a second look.

Source: Wikipedia, Signs of AI writing: Overuse of em dashes (2022–September 2026)

What this checker can't tell you

Who wrote your post. Nothing can tell you that for sure, and pretending otherwise would make this tool worse.

People write like this too. Every habit here is in plenty of human writing. The editors who keep Wikipedia's list say so themselves: not all text with these signs is AI-generated, and a sign is a reason to look closer, not proof.
Detectors can't tell you either. Wikipedia advises against relying on them: they do better than chance, but they have real error rates, and paraphrasing or a model they haven't seen throws them off. One Stanford study found popular detectors kept flagging essays by non-native English writers as AI.
Neither can most readers. In experiments with 4,600 people, Jakesch, Hancock and Naaman found people couldn't tell AI-written profiles from human ones, and leaned on cues that don't work, like contractions and first-person pronouns.
So there's no score. A percentage would claim a certainty nobody has. What the list can do is point at the phrases that make a post sound like everyone else's, so you can decide whether they sound like you.

Sources: Wikipedia, Signs of AI writing: AI detection tools; Liang et al. (2023), GPT detectors are biased against non-native English writers, Patterns; Jakesch, Hancock and Naaman (2023), Human heuristics for AI-generated language are flawed, PNAS

Does Shaflex's AI write posts that sound like this?

Shaflex's composer has an AI menu: Fix grammar, Improve writing, Rephrase and Fit to platform limit. It works on your draft, it never posts anything for you, and it doesn't check for these habits, so read what comes back before you schedule it. When it sounds like you, one draft goes to Threads, X, Bluesky, Mastodon and LinkedIn, each copy fitted to its limit.

See how Shaflex schedules for Threads, X, Bluesky, Mastodon and LinkedIn, how one draft goes to every network, and what the free plan includes.

Where these rules come from

Wikipedia's editors keep the list of signs, and peer-reviewed studies back the word lists and what we say about detectors. We last checked them on Oct 20, 2026.

More free tools

Each one runs in your browser too.

Questions

Does this sound like AI? How do I check?

Paste it into the box at the top of this page. The checker marks the habits readers and Wikipedia's editors link to AI writing, like “not just X, but Y”, delve and em dashes, and says what to write instead. It's a list of hints to look at, not a verdict on who wrote it.

What are the AI phrases to avoid?

The ones that come up most: “not just X, but Y” and “it's not X, it's Y”, “stands as a testament”, “plays a pivotal role”, “serves as” where “is” would do, a clause tacked on with “highlighting” or “underscoring”, “experts say” with no expert named, “it's important to note”, and chat leftovers like “I hope this helps” and “Certainly!”.

Is there a list of AI words to avoid?

Yes, and it changes with each model. Wikipedia's editors list delve, tapestry, testament, pivotal, underscore, intricate and meticulously for 2023 to mid-2024, and fostering, showcasing, highlighting and align with for mid-2024 to mid-2025. One or two mean nothing. Lots of them together is what stands out.

Do em dashes mean AI wrote it?

No. Chatbots overused them, so they got a reputation, but writers have always used them and newer models use far fewer. Wikipedia files the em dash under historical signs, most useful alongside other ones. Keep yours if it's how you write.

Can this tool tell if AI wrote my post?

No, and we don't think any tool can be sure. Detectors have real error rates, and one study found they often flag writing by non-native English speakers as AI. That's why this checker shows habits and their sources, not a percentage.

Why bother, if people can't spot AI anyway?

Because these habits sound the same in every post that has them. Readers may not know who wrote it, but a post full of “pivotal” and “not just X, but Y” sounds like a lot of other posts. Swapping them for what you'd actually say makes it sound like you.

Is my text saved or sent anywhere?

No. It's checked in your browser, and your text isn't stored or sent.

Write the next post once.

Shaflex fits it to Threads, X, Bluesky, Mastodon and LinkedIn, schedules it on your rhythm and brings the replies back to you.

Start free Compare plans

Free plan, no card needed.