They all write the same way, underneath
ChatGPT, Claude, and Gemini are built on the same basic idea. Each one writes by predicting the most likely next word, over and over. That is what makes their default writing smooth and even, and smoothness is exactly what a detector measures.
So a detector is not really looking for “ChatGPT writing” or “Claude writing”. It is looking for machine writing in general, which all three produce.
The differences are real but small
The models do have their own habits. One tends to be more formal, another more chatty, another fond of lists. A careful reader might notice. But for a detector, these style differences are minor next to the shared trait of being even and predictable.
Put simply: the gap between any two of these models is much smaller than the gap between all of them and a bumpy human draft.
Can a detector name the model?
Not reliably. Some tools claim to guess the source model, but there is no clean fingerprint that survives real use. As soon as text is edited or the model is asked for a different style, any small signal fades. Treat “this was written by model X” claims with heavy doubt.
What a detector can do is answer the useful question: does this look like AI at all? Which model is a much harder call, and rarely the one you actually need.
The small style differences you might notice
A detector does not care about these, but a careful human reader sometimes does. Each model has picked up its own habits. None of these are rules, and any of them can be changed with a prompt, but here is the rough feel people report.
ChatGPT
Often tidy and list-friendly. It likes clear structure, headings, and a helpful, slightly formal tone by default.
Claude
Tends toward longer, more flowing sentences and a warmer, more explanatory voice. It often "talks through" an answer.
Gemini
Frequently brisk and fact-forward, and quick to pull in current information when it can.
Notice that these are style, not fingerprints. The moment someone edits the text or asks for a different tone, the habit disappears. That is why even a sharp human guess about the source model is shaky, and a detector does not try to make it at all.
Why the model barely matters for detection
Here is the point that ties it together. A detector is not looking for “ChatGPT writing” or “Gemini writing.” It is looking for the one thing all these models share: smooth, predictable text. Their style differences are tiny next to that shared trait.
So the gap between any two of these models is much smaller than the gap between all of them and a rough, bumpy human draft. That is why one detector covers every model. It is measuring the family resemblance, not the individual face.
What this means in practice
You do not need a different detector for each model. One tool covers ChatGPT, Claude, Gemini, and whatever comes next, because it measures the shared traits of machine writing. And it carries the same limits for all of them: strong on raw text, weak on short or edited text, never proof on its own.