Website Roast

Audit the page you are on and get the critique as a roast: real findings, real fixes, and a card worth sharing.

Every AI website audit produces the same beige listicle nobody reads. Every roast generator produces jokes with no substance. This does both at once, and the joke is load-bearing - humour is what makes someone read a criticism they'd otherwise dismiss. EVERY CRITICISM COMES WITH A FIX. Not "your CTA could be more prominent", but "your CTA is the same grey as your footer links - give it your brand colour and put it above the fold." HALF OF IT ISN'T AN OPINION. Accessibility, SEO, performance and conversion are scored by rules that run on your device and read the page itself. The same page gets the same findings every time. A language model writes the review and scores five genuinely subjective things - copywriting, branding, originality, modern feel, emotional appeal - and those are labelled as its opinion wherever they appear. It is not allowed to contradict the measurements, or to make a claim it can't point at something on the page to support. CLICK A FINDING, SEE THE ELEMENT. Every finding that has one carries a selector, and clicking it outlines that element on the real page. That's the difference between entertainment and a tool. FOUR VOICES. Savage reads the page like a critic who's seen ten thousand of them. Investor reads it in forty seconds and asks what you sell. Designer notices the 3px misalignment before the headline. Support reads it as the person who'll answer the emails it generates. Three intensities change the jokes and never the findings. YOU CHOOSE THE MODEL. By default it uses Chrome's built-in on-device AI, so nothing leaves your machine. You can supply your own Anthropic, OpenAI or Google key instead for better writing - in which case the page content goes to that vendor, on your account, and the extension tells you so at the moment you choose. There is no server in the middle, because there's no company behind this. IT KNOWS WHEN NOT TO BE FUNNY. Memorial pages, medical and crisis resources, charity appeals, government services and personal sites switch automatically to a mode that finds what works and says why - and it tells you why it switched. IT SHOWS YOU WHAT CHANGED. Roast a page you've roasted before and it tells you which findings you fixed, which are new, and how the measured scores moved - which only works because those scores are reproducible. Then take the shareable card and post the score, or copy the whole audit as Markdown into your issue tracker. PERMISSIONS: activeTab, scripting, storage. No host permissions. It cannot read a page you haven't asked it to review.