</> aicodedetect

Claude Code Detector

Paste a snippet to check whether it looks like it was written by Claude, Anthropic's AI assistant, or by another AI model. Free, instant, no sign-up.

0 / 8,000 characters

Two ways Claude writes code

"Claude code" covers two quite different situations, and they leave different evidence. In the chat app, someone asks for a function and copies the answer into their project. That produces a self-contained block of code with the usual assistant habits, much like output pasted from any chatbot. Claude Code, the command-line coding agent, works differently: it reads the repository, edits several files, runs commands and can commit the result. Its changes are spread across a codebase and follow the conventions already there, so any single file looks less like a pasted answer and more like the project's own code.

Check the commit history first

For agent-written changes, the repository is often more telling than the code. According to the Claude Code documentation, the agent adds attribution by default: commits get a git trailer such as Co-Authored-By, and pull request descriptions get a line of plain text. Users can change or remove this in the attribution setting, so its absence proves nothing, but its presence is a direct statement. A CLAUDE.md file at the root of a repository is another clue: it holds instructions for the agent and shows the project uses Claude Code, though not which lines it wrote.

Signals in pasted Claude output

Code copied from a chat answer tends to share the habits of every major assistant, and this detector reads those rather than looking for a Claude-specific fingerprint:

  • Edge cases nobody mentioned. Empty inputs, missing keys and invalid types all handled in a snippet that was meant to cover one case.
  • Comments written for a stranger. Explanations of why a step is needed, aimed at a reader who has never seen the problem, in a file the author maintains alone.
  • Leftovers from the conversation. A comment that says the function was "updated" or "fixed", refers to an earlier version, or answers a question that is not in the code.
  • A finished shape. No dead code, no debug output and no half-done ideas, which is unusual for code someone worked through by hand.

Read more

Frequently asked questions

Can this tool prove that Claude wrote the code?

No. Claude does not embed a watermark in the code it writes, so any detector reads style and produces a probability with reasons. That is a reason to look closer, not proof of who wrote something.

Can it tell Claude apart from ChatGPT, Gemini or Copilot?

Not reliably, and we do not claim to. The major assistants share most of their coding habits. A high score means the code looks AI-written in general.

Does it detect code written by Claude Code, the command-line agent?

It reads any code you paste, including code an agent wrote. For agent work, the repository history is often more informative than the code itself; see the section on commit attribution above.

What if the author removed the Co-Authored-By trailer?

Then the commit history no longer says anything about Claude, and you are back to reading the code. The detector can help with that, with the same limits as for any other assistant.

Is my code stored?

No. The snippet goes to a third-party AI model provider (Groq, with Google Gemini as a fallback) for the analysis and the result comes straight back to you. We do not save the code or write it to our logs.