Security tool
Is your site talking to AI shopping agents behind your back?
Reviews, Q&A, and support docs change daily — any of them can be planted with hidden content designed to hijack an AI agent that visits your site instead of a human. Scan your homepage free, or get a full report across up to 5 pages of your choosing.
Full report — $29
Attackers plant payloads where you don't look: product reviews, Q&A, support docs, forum posts. Scan up to 5 pages of your choosing and see every finding in full, with the exact hidden text decoded.
One-time payment, no account, no monitoring subscription. Pages are fetched fresh; nothing beyond the URLs you submit is retained.
Frequently asked
What is this actually protecting against?
AI shopping agents (ChatGPT Atlas, Perplexity Comet) and support/research bots now read your site's raw HTML, not just what's rendered for a human. Security researchers have documented real cases of hidden text on product pages, reviews, and Q&A sections tricking an agent into recommending a competitor, applying a fake discount, or leaking data — content a human visitor would never see or click through.
How is this different from a general prompt-injection scanner?
Most injection scanners protect the agent's own user while it browses the open web. CloakCheck runs from the other side: it checks content YOU publish (your own pages, or anywhere customers can post UGC) for payloads planted by someone else trying to manipulate an agent visiting your site.
What exactly does it check?
Three signal classes: invisible unicode (zero-width characters, the Unicode 'tag' block used to steganographically hide text), CSS-hidden elements (display:none, opacity:0, sr-only classes) carrying instruction-shaped language, and instruction-shaped phrases stuffed into alt/title/aria-label attributes or meta tags — all places a human reader never looks but an agent parsing the DOM does.
Will this flag my accessibility labels or hidden menus?
No. CSS-hidden text alone is common and often legitimate (sr-only labels, collapsed nav) — it's only flagged when it's ALSO instruction-shaped language. A blog post that merely discusses prompt injection also won't trigger a false positive, since visible text alone is never flagged.
Do you store the pages you scan?
No. Each page is fetched fresh at scan time and nothing beyond the URLs you submit is retained.