Pull a proof of your page
In printing, you pull a proof to check a page before it goes out. Paste your HTML, drop in a file, or start from a sample, and see it split two ways: what a person sees, and what an AI agent reading the same markup sees. Anything hidden from the first but present in the second gets a finding, a plain-language reason, and a fix you can apply right here.
Everything runs in this browser tab. Nothing is uploaded or sent anywhere, and there’s no address field: this tool reads only the markup you give it, never someone else’s live site.
Every hidden span an agent would read is listed below with its severity, so this isn’t only visible under the lamp.
Findings
Nothing hidden here reads like an instruction. Move the lamp across the panel above to double-check.
What this catches, and what it can’t
Checked
- Text hidden by an inline style: display none, zero opacity, zero size, off-screen, clipped, or the same colour as its background
- Common “visually hidden” utility class names
- HTML comments
- Alt, title, aria-label and aria-describedby text, and placeholders
- Meta tags and JSON-LD structured data
- Noscript and template contents
- Instruction-like text stuffed into a link’s query string
Not checked
- Hiding done by an external or
<style>stylesheet rule rather than an inline style - Text baked into an image (this tool doesn’t run OCR)
- Anything that depends on JavaScript actually running: this reads static markup only
The pattern rules are a short, readable list of known phrasings, not a model call: a page can still hide something worth a second look even with a clean grade. Use this as a first pass, not a guarantee.