Copy-paste

Claude artifacts HTML classes: the copy-paste trace

If you copy from a Claude artifact, preview, or rich card, the clipboard often carries HTML, not just visible words. Class attributes containing “claude” are one of the inspectable leftovers this checker is built to find. They are easy to miss in a Google Doc that still looks like plain prose.

Paste and checkHow the checker works →

What shows up in the paste

A typical artifact copy is a fragment like class="…claude…" wrapped around headings, code blocks, or layout divs. You may not see that in the rendered preview. Paste into a CMS, email composer, or “paste with formatting,” and the class names travel with the HTML.

The in-browser checker counts those attributes, plus zero-width characters and exotic/nbsp spaces. Em dashes and curly quotes are listed as stylistic only — not as proof you used Claude.

How to clean them

  1. Paste the full copy (including HTML if that’s what you grabbed) into the homepage tool.
  2. Read the findings. Deterministic hits are the ones worth cleaning.
  3. Click Clean traces. Class names containing “claude” are emptied, zero-width characters are removed, and exotic spaces become ordinary spaces. That step is free and stays in the browser.
  4. Paste the cleaned text as plain text when you can, so a later editor does not re-wrap it in leftover markup.

HTML classes vs the statistical watermark

Stripping claude from a class name does not neutralize Anthropic’s text watermark. That mark is a bias in token choice. Cleaning markup is hygiene; rewrite is the separate, paid job. The split: clean traces vs rewrite.

Also not C2PA

Artifact HTML is still text/markup. C2PA provenance lives on supported files (PNG, JPG, SVG). Screenshotting an artifact or exporting an image is a file-metadata problem. Pasting the artifact’s prose is a text problem.

FAQ

Why does pasted Claude HTML mention “claude” in class names?

Claude artifacts and some rich copies include HTML class attributes that contain the word “claude”. Those are inspectable traces in the markup, not the statistical watermark in token choice.

Does stripping class names remove Anthropic’s text watermark?

No. Cleaning class names, zero-width characters, and exotic spaces is a local cleanup. The statistical mark lives in which words were chosen. That takes a rewrite.

Is this the same as C2PA?

No. C2PA is signed provenance on files like PNG, JPG, and SVG. Artifact HTML classes are text/markup. Different system.

Clean artifact HTML for freeWhen you also need rewrite →

Related guides

Not affiliated with Anthropic. Updated August 30, 2026.