Clean the metadata around your content
HTML pages, SVG drawings and Markdown documents can carry creation information outside their visible content. Free AI Wiper inspects supported metadata, separates recognized AI markers from ordinary fields, and lets you remove the selected categories in your browser.
For example, an HTML page can contain a generator field naming Gemini while its body contains a useful article. Cleaning that field leaves the article, styles and scripts intact. A mention of Gemini in a headline or description is not treated as creation evidence.
What each format supports
| Format | Supported cleanup | Retained content |
|---|---|---|
| HTML / HTM | Supported meta name fields in an explicit document head; recognized provenance comments |
Body text, styles, scripts, SEO descriptions, viewport and encoding declarations |
| SVG | SVG-namespace metadata blocks, the root generator attribute and recognized provenance comments |
Paths, shapes, titles, accessible descriptions, styles and embedded image data |
| Markdown / MD | Supported flat scalar frontmatter fields; recognized standalone provenance comments outside code | Prose, ordinary frontmatter, code examples, line endings and UTF-8 byte-order marks |
Supported ordinary fields include author, creator, generator, producer, software, copyright, rights and creation/modification dates. Recognized provenance fields include AIGC, C2PA, Content Credentials, digital source type, provenance and explicit AI-generation or generator-attribution fields. A finding records metadata presence; even a field named ai_generated is not independent proof of authorship.
Inspect, select and download
- Open the cleaner, choose Documents, and select an HTML, SVG or Markdown file up to 1 MiB.
- Review the findings. AI metadata is selected by default; ordinary metadata is optional.
- Clean and download after verification. The output check compares the complete file with the intended source edits and inspects it again.
The app treats markup as source and does not execute or preview its scripts. It preserves every byte outside selected metadata. A file with no selected findings is returned unchanged. Removing metadata may still affect software that intentionally reads that metadata; keep an original copy.
Supported boundaries
Inputs must be UTF-8. HTML needs an explicit html element and a closed head; simple HTML5 doctypes or no doctype are supported. Custom declarations, ambiguous attributes and legacy escaped script content are refused. Metadata with identifiers or additional semantic roles is refused when selected. JSON-LD, arbitrary data-ai attributes, embedded data-image metadata and metadata in templates or foreign content are retained.
SVG requires one balanced root in the standard SVG namespace. DTD/entity declarations and recognized XML digital signatures are refused. Selected metadata containing identifiers is refused because other content may reference it. Cleanup covers the supported metadata carriers; it does not remove a visible logo, text label or pixel watermark. Supported SVG pictures in standard DOCX/PPTX/XLSX media folders use this same cleaner; see Office cleanup.
Markdown frontmatter is limited to flat scalar fields. Nested YAML, lists, block scalars, aliases, anchors and tags are refused. Operational fields such as model are retained. Comment recognition is conservative and limited to explicit, standalone provenance comments; inline comments and examples inside code, raw HTML or containers are retained. Body Unicode cleanup is a separate choice in the text cleaner.
Inspection is capped at 20,000 markup tokens or Markdown lines, 256 attributes per tag, 128 nesting levels and 10,000 metadata carriers. These are processing limits, not a browser performance guarantee. EPUB and ODT are not supported yet. A clean report means selected supported metadata was absent on reinspection, not that every AI signal or watermark has been removed.