What this ChatGPT code detector looks at
ChatGPT writes code the way it writes prose: completely, evenly and for a reader who has never seen the problem before. That habit leaves a texture you can read. The detector sends your snippet to a language model with instructions to review it the way an experienced engineer would, then reports a verdict, an AI probability and the specific reasons behind the number.
- Explanatory comments. A comment above nearly every block, often restating what the line already says, plus docstrings of uniform length on every function.
- Chat-shaped framing. An
Example usage:section, a demo block with sample inputs, or a comment that refers to "your requirements" or "the code above". - Unrequested completeness. Input validation, type checks and error handling for cases the task never mentioned, usually with generic messages like
"Invalid input". - Textbook naming.
result,items,is_valid,current_node— descriptive, consistent and applied without a single lapse. - No mess. No debug prints, no TODO notes, no commented-out first attempt, nothing abandoned halfway. Code that someone actually worked through usually leaves traces.
What it cannot tell you
Two limits matter. First, the detector cannot single out ChatGPT. Claude, Gemini, Copilot and Cursor share most of the same habits, so a high score means "this looks AI-written", not "this came from ChatGPT". Second, it reads only the text you paste. It knows nothing about the assignment, the rest of the repository or how the author normally writes, and those are often the strongest clues available to you.
False positives to expect
Some human code genuinely looks generated. Careful developers who document everything, code that has been through a formatter and a strict linter, and beginners copying the style of a popular tutorial all tend to score higher than they should. Short snippets are the worst case of all: in ten lines of a standard algorithm there is no room for personal style, so any detector is mostly guessing. Paste a complete function or a meaningful section of a file, keep the original comments, and check a few independent pieces rather than trusting one score.
Use the result as a prompt for a conversation. Asking someone to walk through their code, explain one design choice or make a small change on the spot tells you far more than a percentage does.
Read more
- How to tell if code was written by ChatGPT — the signals above in detail, plus a step-by-step way to check fairly.
- Python AI code detector — the same tool with Python-specific signals.
- AI code checker for homework — for students and teachers reviewing assignments.
Frequently asked questions
Can a ChatGPT code detector prove that ChatGPT wrote the code?
No. ChatGPT does not watermark the code it writes, so a detector reads style instead: comments, naming, structure and how finished the code looks. That produces a probability and a set of reasons, not proof. Treat it as a reason to look closer and ask the author about their work.
Can it tell ChatGPT apart from Claude, Gemini or Copilot?
Not reliably. Today's assistants were trained on overlapping data and have picked up similar habits, so the style they produce is hard to separate. This tool estimates whether code looks AI-written in general rather than naming the product behind it.
Does it still work if the ChatGPT output was edited?
It gets harder. Renaming variables, deleting the explanatory comments and merging the code into an existing file all remove the signals a detector relies on. Lightly pasted output is much easier to spot than output someone has reworked.
Is the ChatGPT code detector free?
Yes. You get up to 10 checks per day without signing up. The daily limit keeps the running costs manageable, and it applies per IP address.
Do you keep the code I paste?
No. The snippet is sent to an AI model (Google Gemini) for the analysis and the verdict comes back to you. We do not save your code or write it to our logs.