Your friend has had a rough week. You send a clumsy message to check in. Twenty minutes later the reply arrives, and it is beautiful. It thanks you for reaching out. It lists three things they are grateful for. It closes by hoping you are looking after yourself too.
It is a lovely message. And for a split second, before you feel any warmth at all, you wonder whether they actually wrote it.
The autocomplete era
Machines have been finishing our sentences for longer than most people realise. Old keypad phones guessed your word from a few button presses, often wrongly. Autocorrect made confident mistakes of its own. Then, about a decade ago, email apps began offering tiny ready-made replies. “Sounds good!” “Thanks for letting me know.”
Those buttons were a small convenience. They were also a quiet nudge. Tap one and you have said what countless other people said that day, in the same words. When the easy option is always there, plenty of people take it.
Keyboards now suggest the next word as you type. Some offer to rewrite a whole message in a friendlier tone. Each suggestion is reasonable on its own. Together, they pull everyone towards the same middle: polite, upbeat and slightly bland. The rough edges that made your texts sound like you get sanded down a little at a time.
That is the flattening of tone. You would never spot it in one message. You notice it when everybody’s “congrats!!” starts to look the same.
“Did ChatGPT write this?”
Once chatbots that write whole paragraphs became free and easy to use, a new question entered the group chat. Did a person write this, or a machine?
Tools like ChatGPT are built on an LLM, a large language model trained on vast amounts of human writing to predict the next word. That is why the output sounds so fluent. It is also why it sounds so familiar.
People have built up a folk list of tells. Over-polished structure, with a greeting, three tidy points and a warm sign-off. “I hope this message finds you well.” Lists that always come in threes. And, most famously, the em dash, the long line some writers use to break off a thought. Somewhere along the way it became a supposed giveaway.
Here is the problem. Humans wrote like this first. The chatbots learned it from us. “I hope this message finds you well” has been clogging office inboxes for decades. The rule of three is as old as speechwriting. Novelists and journalists have loved the em dash for centuries. Some now say they cut it from their own writing to avoid suspicion, which is a strange thing to have to do.
| The tell people point to | What it usually means |
|---|---|
| “I hope this message finds you well” | Someone who learned to write emails in an office |
| An em dash in the middle of a sentence | A person who likes punctuation, as writers always have |
| A neat list of exactly three things | One of the oldest habits in public speaking |
| Words like “crucial” and “furthermore” | School essays or years of formal work writing |
| A long, well organised reply to a short text | They care, or they had time on the train |
No single tell proves anything. Sometimes the polished message really was drafted by a chatbot. Often it came from a careful person now mistaken for one. The detective work says more about our suspicion than about the message.
Writing messy on purpose
So people have adapted. If polish looks artificial, mess looks human.
All-lowercase messages. Missing full stops. Typos left in because fixing them feels try-hard. None of this is new. What is new is that some people now do it on purpose, as a kind of proof of life.
Shorthand plays a big part. A word like lwk, short for low-key, or icl, short for “I can’t lie”, does something a chatbot’s default voice does not. It signals that you are talking casually, to someone you know, without trying to impress. “icl that film was mid” reads like a person because it is imperfect. It skips words. It assumes shared context.
Chatbots can imitate this if asked. But their default is tidy and complete, and most people never change it. So for now, a dropped capital letter says, roughly, “a human typed this with their thumbs, probably while doing something else.”
The irony is neat. Adults spent years complaining that texting abbreviations were ruining proper writing. Now they are one of the clearest signs of a real person on the other end.
Ghostwritten feelings
Emotional writing is trickier. Here, AI help raises questions that spelling suggestions never did.
Dating apps are the obvious example. Some offer to help write your bio or an opening line, and standalone apps promise rizz on demand. The appeal is clear, since messaging a stranger is intimidating. The risk is just as clear. If the witty person in the chat was partly a chatbot, the first date means meeting someone slightly different. For plenty of people, finding out a match’s charming messages were machine-made would be an instant ick.
Then there are the harder messages. Apologies. Condolences. Breakups. People ask chatbots for help with these precisely because they matter and hurt to write. Someone who freezes when a friend’s parent dies might want a starting point. Someone ending a relationship might want help being kind rather than blunt.
Does it count if a machine drafted it? It probably depends on what happened next. A message you drafted with help, then rewrote and meant every line of, is not so different from asking a friend how to phrase something. People have always borrowed words from sympathy cards and wiser friends. A message pasted in without reading closely is a different thing. The words may be fine. The effort, often the thing the other person wants most, is missing.
One useful test is simple. Would you be comfortable if they knew? If yes, the tool was probably a helper. If not, it may have done a job that was meant to be yours.
The slang people invented about AI
Whenever something big arrives, people give it nicknames. Naming a thing helps us get a grip on it. Mocking it a little makes it less overwhelming. AI has produced a whole vocabulary very quickly, following the usual pattern of how new slang spreads: small communities first, then everyone.
Clanker is the rudest of the bunch. It started as a Star Wars insult for battle droids and became a dismissive word for robots and AI in general. Calling a customer service bot a clanker is a tiny act of rebellion, and usually a joke.
AI slop describes the flood of low-effort machine-made content filling feeds. Odd images. Recycled articles. People in pictures with too many fingers. The word suggests something cheap, mass-produced and not quite fit to consume.
Vibe coding is the gentler one. It means building software by describing what you want to an AI and accepting whatever comes back, without really understanding the code. The AI researcher Andrej Karpathy coined it in early 2025. It caught on because so many people recognised themselves in it.
“Just ChatGPT it” has joined “just Google it” as the modern way of telling someone to stop asking you things. And people treat chatbots like that one overconfident friend. It answers everything with total certainty. It is sometimes completely wrong. In technical terms, a model can hallucinate, inventing facts, quotes or sources that do not exist. People have also noticed that chatbots love to yap. Ask a yes-or-no question and you may get four paragraphs and a summary.
None of this vocabulary is neutral. It is sceptical, playful and a bit defensive. That is how people sound when adjusting to a big change in real time.
What stays human
For all the hand-wringing, some parts of texting remain very hard to fake.
In-jokes are the clearest example. One word about a holiday three years ago means nothing outside the group. A chatbot was not on the holiday. Shared context works the same way. Your best friend knows that “fine.” with a full stop means you are not fine.
Timing matters too. A reply at 2am says something a reply at noon does not. So does a message the morning after a big day, just asking how it went.
And then there is the single emoji. Knowing when a crying-laughing face is enough. Knowing when a heart beats a paragraph. Knowing when the right response is silence for a while. That judgement comes from knowing a person, not from predicting the next likely word.
AI will keep getting better at writing, and keep changing how we text. But the messages people treasure are rarely the most polished ones. They are the ones only one person could have sent. If you want to keep up with the words we invent along the way, the full dictionary is a good place to start.
Questions people ask
Can you tell if a text was written by AI?
Not reliably. Tells like polished structure, the em dash or “I hope this message finds you well” were common in human writing long before chatbots, so usually the only honest way to find out is to ask.
Why do people type in lowercase on purpose?
Lowercase, typos and abbreviations make a message feel casual and personal. Now that polished writing can look machine-made, a little mess signals that a real person typed it.
Is it wrong to use AI to write a personal message?
Not necessarily. Using a chatbot for a starting point and rewriting it in your own words is like asking a friend for help, but sending something you do not mean tends to feel hollow.
What does clanker mean?
Clanker is a dismissive slang word for robots, chatbots and AI. It comes from Star Wars, where it was an insult for battle droids, and is mostly used jokingly.
What is AI slop?
AI slop is low-quality content churned out in bulk by AI, such as strange images, generic articles and uncanny videos. The word suggests something cheap and not worth your attention.