html-to-ai
Static page (GitHub Pages) that slims down HTML copied from a web page before pasting it into an AI (Claude Code, etc.).
Try it: https://leobastiani.github.io/html-to-ai/
How to use
- Copy the page's HTML: in DevTools, right-click the element → Copy → Copy outerHTML; or just select a part of the rendered page and copy it (the page uses the clipboard's
text/html).
- Paste it into the left pane.
- Click Copy in the right pane and paste the result into the AI.
Everything runs in the browser, nothing is sent to any server.
What is removed
script, style, noscript, template, link, meta, iframe, canvas, ... and HTML comments.svg content (only <svg> with its useful attributes remains).- Attributes that don't help the AI:
style, on*, data-*, tabindex, srcset, jsaction, ... Only an allowlist is kept (href, src, alt, role, aria-*, type, name, data-testid, ...). See KEPT_ATTRIBUTES in src/cleanHtml.js.
- Attributes that state the obvious:
role equal to the element's own (<td role="cell">), aria-* with the default value (aria-required="false", aria-sort="none"), aria-label equal to the text, aria-describedby/aria-labelledby pointing to ids that no longer exist, and href="#".
- Classes and
ids with no meaning (vnskjfndj, css-1x2y3z, sc-bdVaJa, Button_root__a1b2c) and those from component libraries (ng-star-inserted, mat-mdc-cell, cdk-row, MuiButton-root), keeping human-readable names (product-card, mainMenu, btn--primary, col-6). The heuristic lives in src/isSemanticName.js.
data: URIs are truncated and very long URLs are cut.- Empty elements.
Options
- Format:
HTML (default) or Markdown. Markdown is generated from the already cleaned HTML (respecting the other options): headings, lists, links, emphasis, code/pre and tables become Markdown; div/span/components keep only their content; form fields and svg stay as inline HTML. See src/toMarkdown.js.
- Remove
div/span/components that only wrap (on): span, a and custom elements (mat-card, fire-search-bar) without attributes are replaced by their content; a div without attributes and with a single child is replaced by that child.
- Remove hidden (on): elements with
aria-hidden="true" or hidden (icons, ligatures like more_vert).
- Remove classes and ids: when only the content matters.
- Remove buttons and fields:
button, input, select, textarea, comboboxes and menus.
- Shorten lists: keeps the first 3 siblings of each tag (except table cells) and replaces the rest with
<!-- 13 more <tr> -->, enough for the AI to understand the structure.
- Indent HTML (on).
Development
pnpm install
pnpm dev # serves the page locally
pnpm test
pnpm lint
Deployment to GitHub Pages happens on every push to main (see .github/workflows/pages.yml). Static SEO files (robots.txt, sitemap.xml) live in public/ and are copied to the site root on deploy.
