Signals that Python code was AI-generated
Python is unusually readable, which cuts both ways: it is easy for an assistant to produce clean, idiomatic code, and easy for a reviewer to notice when it is a little too clean. These are the Python-specific patterns that show up most often in generated code.
- A docstring on everything. Every function carries a complete docstring in the same format, with
Args:andReturns:sections, even for a two-line helper. Human projects are far patchier. - Type hints applied without exception. Full annotations on every parameter and return value, including
-> Noneon functions that obviously return nothing, in code where nothing else suggests a typed codebase. - A demo block. An
if __name__ == "__main__":section with sample data and printed output, attached to a module that was only supposed to expose a function. - Defensive scaffolding.
if not isinstance(x, list): raise TypeError(...), empty-input guards andtry/except Exceptionwrappers around code that nobody asked to harden. - Textbook idiom. Comprehensions,
enumerate,zip, f-strings andpathlibused correctly and uniformly, with none of the older or clumsier habits most people still mix in. - Imports that do not match the code.
from typing import List, Dictalongside modern builtin generics, or an import that nothing in the file actually uses.
Why Python is harder to judge than it looks
PEP 8 exists to make Python look the same everywhere, and most projects run Black or Ruff on save. That removes exactly the small inconsistencies — spacing, quote style, line breaks — that would otherwise mark a human author. Introductory Python is also the most heavily taught code on the internet, so a student following a course closely, and a model trained on the same material, will often produce nearly identical solutions to a standard exercise. Scientific and data code adds another wrinkle: pandas and NumPy pipelines are conventional by nature, and a tidy one is not evidence of anything.
How to read the result honestly
The output is a probability and a short list of reasons, not a verdict. Read the reasons first and check whether they describe something you can see in the code. A score driven by "consistent formatting" alone is weak; a score driven by an unused import, a demo block nobody asked for and a docstring style that appears nowhere else in the project is worth following up.
Context beats style every time. Compare the snippet with the author's other Python, look at how it was committed, and ask them to extend it by one small feature. That is the check that actually settles the question.
Read more
- ChatGPT code detector — the same tool framed around ChatGPT output.
- How to tell if code was written by ChatGPT — the full checklist of signals.
- AI code checker for homework — for Python assignments and graded work.
Frequently asked questions
Does the detector work better on Python than on other languages?
Python gives a detector a bit more to read than most languages, because docstrings, type hints and comments are all conventional places where an assistant adds text. The trade-off is that PEP 8 and tools like Black push everyone toward the same formatting, which removes some human signal.
Do I need to select Python from the dropdown?
No. The language is detected automatically from the snippet. Choosing Python explicitly can help with very short code, or with files that mix Python and something else such as SQL strings or shell commands.
Does code formatted with Black or Ruff look AI-generated?
It can push the score up, because consistent formatting is one of the signals. It should not dominate the result on its own, since the analysis also weighs comments, naming and structure, but it is a good reason to read the stated reasons rather than only the number.
How much Python should I paste?
A complete function or a module section of a few dozen lines works best. A five-line helper or a list comprehension carries almost no stylistic information, and any detector will return an uncertain answer for it.
Does pasting Python code store it anywhere?
No. The snippet goes to an AI model (Google Gemini) for the analysis and the result comes straight back to you. We do not save the code or write it to our logs.