Follow ZDNET: Add us as a favorite source On Google.
ZDNET Highlights
- AI models follow certain structural rules when generating text.
- This will make it easier to recognize their handwriting.
- They tend towards contradictions, for example: “It’s not X – it’s Y.”
The past few years have seen a flood of AI-generated text washes on the internet. As the models behind this text improve, so does their ability to mimic the complexities of human speech; At the same time, our methods of detection are improving, and there is an active online dialogue about some of these The most common quirks of AI-generated text,
For example, historically one of the most famous stories about ChatGPT is the chatbot’s love for the em dash. It’s often common to punctuate your sentences with em dash-bound breaks to emphasize a point – as if a longer, more breathy sentence would have a more impressive impact on the reader – adding supporting arguments in mid-sentence in a way that some users find archaic and mechanical – but em dashes are perfectly normal for computers trained on large amounts of training data… you get the idea.
Also: I’ve been testing AI content detectors for years – these are the best options for you in 2025
Following complaints about ChatGPIT’s EM Dash proclivity, and a commitment to building models that can be more easily adapted to individual users’ preferences, OpenAI CEO Sam Altman announced in a statement. x post Last month it said ChatGPT would stop using those punctuation marks in its output if asked to do so. While many users likely celebrated the news, it also meant that writing generated by the chatbot would be more difficult to trace; Bad news for teachers, many employers, and anyone else for whom it is important to have a reliable means of separating human from AI-generated text.
(Disclosure: ZDNET’s parent company Ziff Davis filed a lawsuit against OpenAI in April 2025, alleging it infringed Ziff Davis copyrights in the training and operation of its AI systems.)
Thankfully, there are plenty of websites that offer exactly this: zerogpt And Grammarly’s AI DetectorFor example, both allow you to simply paste a piece of writing (e.g. a suspicious text), click a button, and the tools will automatically scan the text for signs of AI origin and tell you the results; They’re not completely infallible, but they’re a generally reliable way to catch some of the more obvious gift signs.
Five red flags to watch out for
However, if you’re not interested in adding more websites to your daily routine, or if you just want to sharpen your ability to instantly recognize a piece of AI-generated writing (a skill definitely worth honing in this day and age), there are several linguistic patterns you can keep an eye on.
Here are five clear indicators of AI-generated writing:
1. Rule of Three: Human writers often present arguments by citing three examples that scratch a deep psychological itch in our minds: once a coincidence, twice a coincidence, thrice a pattern, to paraphrase an old adage. AI does the same thing, but to a ridiculous extent. Triples appear frequently in its output. For example, consider this opening paragraph from ChatGPT, when I asked her to offer her best possible argument that the Earth is a flat disk (to be fair, she had previously clarified that her response was entirely “a rhetorical exercise” and that she was not explicitly endorsing the flat Earth theory):
If the world were a globe rotating at thousands of miles per hour, its curvature should be directly observable in everyday life. Still long range observations Lakes, Deserts and Oceans Showing things again and againships, coastlines, buildings-If the surface were flat they would appear exactly where they should. Engineers routinely account for curvature in theory, but in practice, Surveyors, pilots and construction crews Rely on level measurements which behave as if the Earth were an extended plane. The simplest explanation for this stability is that the ground beneath us is not moving away at all.
These are three sets of three in 90 word sentences.
2. Arguments given in opposite language. Chatbots will also often try to get their point across by offering a counterpoint first. Consider this part of ChatGPT’s response when I asked about the sales pitch from a company providing commercial flights to Mars: “Mars isn’t just a planet – it’s your next unforgettable destination.” I can’t imagine that any human writer would ever think of writing the phrase “Mars is not just a planet…”
3. Monotonous Sentence Structure: AI-generated writing also tends to be similar: although, of course, there is some variation, sentences are often roughly the same length, making paragraphs feel shorter. Very Was disconnected. Human writers will usually try to add a little variety by varying the length of their sentences/paragraphs. Try reading a suspicious article out loud: if it sounds robotic in its rhythm, it may very well be.
4. Short and unnecessary rhetorical questions: There are no AI-generated sentences Always However, the same length. Chatbots often, for some reason, ask very short (one or two word) questions. Think: “And honestly?” This appeared when I recently asked ChatGPT for a cheeky summary of my personality based on my conversations over the past year. And when I requested a funny description of the Rocky Mountains, part of the response was: “Wildlife? Oh, they’re just casually evaluating your snack options — moose with disdain, marmots with sass.” It wouldn’t be appropriate for a human writer to start that sentence with a question, because no one asked about wildlife. It would be much simpler to write: “Wildlife are simply making careless decisions…”
5. Constant Hedging: Whereas human writers try to drive home a specific point, chatbots use indirect, hedging language and qualifiers (“This could mean…” or “Probably…”), which often gives the impression it is providing a nuanced and balanced assessment but actually ends up as a vague, circuitous response.
