At a time when conversational AIs are taking over our screens, it’s becoming increasingly difficult to tell a human text from one generated by a machine.
ChatGPT, Yiaho, Gemini, Grok, Claude, Deepseek, and other models are competing in sophistication, but each leaves behind clues—”fingerprints” that reveal their artificial origin.
In this article written by the Yiaho team, we’ll break down the obvious signs specific to these AIs, with concrete examples and specific markers. And don’t forget we also have an online AI detector, if you need it!
1. ChatGPT: mechanical transitions
ChatGPT, created by OpenAI, is probably the best-known and most-used model to date, especially with our free version of ChatGPT. But that popularity comes with some very visible habits. One of its most striking markers is its frequent use of connectors like “moreover,” “furthermore,” “in addition,” or “nevertheless.” These words, while correct, often appear in a mechanical way, almost like a repeated formula to structure the text. For example: “Moreover, this technology helps improve business productivity. Furthermore, it offers an intuitive interface.”
Another clue: ChatGPT tends to over-explain, even for simple questions. Its answers are often long, with detailed sentences and a linear logic that can feel a bit too academic.
It also likes to use generic phrases like “it should be noted” or “in this context,” which can make the tone feel impersonal at times.
If you’re reading a text packed with predictable transitions and exhaustive explanations, there’s a good chance ChatGPT is behind it.
We wrote a dedicated article on this topic: how to recognize a text written with ChatGPT.
2. Gemini: smooth, cautious perfection
Gemini, developed by Google, stands out for its fluent, concise writing that can fool you at first glance. But that perfection is precisely its weak point: its texts are often too well structured, without the small irregularities or hesitations typical of a human. Among its recurring markers are turns of phrase like “it’s interesting to note that,” “this makes it possible to,” or “to some extent.” A typical example: “It’s interesting to note that this innovation simplifies everyday tasks, making it possible to optimize time.”
Gemini also adopts a neutral, cautious tone, avoiding strong opinions or overly bold wording. Its sentences are balanced, almost calibrated to please everyone, which can give an impression of a lack of personality.
If a text feels too smooth, almost artificially polished, think Gemini.
3. Grok: dashes and straight-to-the-point humor
Grok, created by xAI, is an AI with a very recognizable style. Its most obvious hallmark is its frequent use of dashes—like these—to structure ideas, add clarifications, or give rhythm to sentences. This habit quickly stands out once you notice it. For example: “AIs are fascinating—really—but they still don’t replace a good old conversation over coffee.”
Beyond the dashes, Grok stands out for a direct tone, often with a hint of humor or light sarcasm, especially when faced with offbeat questions. Maybe the “Elon Musk touch”? Its answers feel more spontaneous than those of other models, with a personality that stands out in a landscape of neutral AIs.
If you come across a text peppered with dashes and a relaxed style, there’s a good chance Grok is behind it.
4. Claude: empathy and human warmth
Claude, designed by Anthropic, aims to be the most “human” and reassuring AI possible. Its style is marked by an unusual warmth for a machine, with expressions like “I’m happy to help,” “let’s look at this together,” or “that’s an excellent question.” Example: “I’m delighted to answer that for you! This technology is exciting, isn’t it?”
Claude also favors empathetic, reassuring phrasing, avoiding overly technical or abrupt terms. Its texts have a softness that can be appealing, but that constant kindness can also become a clue: humans vary their tone more.
What’s more, it has a slight tendency to sidestep controversial topics with phrases like “it depends on your point of view,” reflecting its caution-focused programming.
A text overflowing with kindness and encouragement? Claude might be behind it.
Also read: Is AI plagiarism? Here’s what the law says
5. Yiaho: the AI that’s too human?
Our free AI platform Yiaho takes a radically different approach.
Unlike its competitors, it’s programmed to avoid any recognizable markers, making its texts almost undetectable to the naked eye. No suspicious repetitions, no obvious stylistic tics: Yiaho mimics human naturalness with unsettling precision.
We’ve pushed mimicry so far that it can vary its tone, structure, and vocabulary to escape the usual AI patterns. For example, it might write: “Honestly, this question makes me think, and I’d say it depends on the day.”
But the irony doesn’t stop there: our platform Yiaho also offers an AI detector, a tool designed to analyze any text and determine whether it was generated by an artificial intelligence or written by a human. A clever way to stand out while playing the transparency card.
If a text seems too perfect to have the typical flaws of AI, Yiaho might well be the culprit—or at least, you can check with their detector!
6. Other lesser-known AIs: more obvious tics
Some less common or older AIs leave even more visible traces. For example, LLaMA (developed by Meta AI) can produce texts with repeated keywords or sentences that lack flow, as if the model struggles to vary its vocabulary. A typical answer might be: “Artificial intelligence is powerful. Artificial intelligence is changing the world.” You often see this kind of sentence at the start of a paragraph.
On their side, models like Bard (Google’s former competitor, before Gemini) had a habit of summarizing their answers with overly obvious concluding phrases like “in summary, this shows that.” These more experimental or less refined AIs often give themselves away through awkward transitions or logic that can feel disjointed at times.
Universal markers of AI-generated texts
Beyond the specifics of each model, some clues are common to almost all AIs:
- Impeccable grammar: No spelling mistakes or typos, but sometimes a lack of spontaneity or personal style.
- An overly rigid structure: Answers often follow an introduction–development–conclusion pattern, even for trivial questions.
- Standardized vocabulary: Words like “efficient,” “innovative,” “significant,” or “crucial” come up frequently, lacking a unique creative touch.
- A lack of raw emotion: Even in humorous or empathetic texts, you can sometimes sense a limit to emotional authenticity.
How can you train yourself to spot them?
To sharpen your eye, read online texts—articles, comments, forum replies—and look for these markers:
- Do the “moreovers” pile up (ChatGPT)?
- Do dashes punctuate every idea (Grok)?
- Is the tone too warm and polite (Claude)?
With a bit of practice, you’ll start to guess who’s behind each paragraph. And if you’re still unsure, ask the author a trick question: AIs often struggle to improvise when faced with the absurd or the unpredictable!
A digital detective game
Spotting an AI-generated text is like cracking a code. ChatGPT lines up its mechanical transitions, Gemini shines with smooth perfection, Grok sprinkles dashes with a touch of humor, Claude wraps everything in softness, and Yiaho covers its tracks and can detect AIs. Each model has its strengths, but also its stylistic weaknesses.
So next time you read a text, have fun looking for these clues. Who knows—maybe this one already bears the marks of its creator?
