How the Docognito disguise works
A visual overlay sits in front of the real page while the assistant keeps running underneath it, unchanged.
A layer drawn on top of the real page, not a rewrite of it
the visible page, redrawn as a document editor. The real claude.ai or ChatGPT interface is still there, still running, sitting underneath that layer the entire time. The extension does not delete it, replace it, or strip it out, it simply draws a fully opaque document view in front of it, at the very top of everything else on the page.
That distinction is what keeps the assistant working normally while it looks like something else. Nothing about how ChatGPT or Claude loads, sends a message, streams a reply, or handles a file changes underneath the overlay, because none of that code was touched. Only what sits in front of it, the part your eyes actually meet, is different.
Your message still goes to the real assistant
The document-styled box you type into is not a separate note kept off to the side. Each character you type is relayed into the actual ChatGPT or Claude message box running beneath the overlay, the same box that would be there with the extension switched off entirely. Pressing send submits through that real box, and the reply streams back into view the same way it always would.
What this does not change is worth stating plainly: the conversation itself, which OpenAI and Anthropic still receive and store. The extension redraws what a message looks like on your screen. It has no effect on what OpenAI or Anthropic receive or retain once you press send, because the message itself travels unchanged.
The tab title and favicon change at page load
Two things sit in a browser tab before any content does: the words on the tab and the small icon beside them. Docognito replaces the browser tab title and the tab icon the moment a tab loads, ahead of the point the assistant itself would set its own title, so neither one is ever visible even for an instant on a fresh load.
Both assistants rewrite a tab title of their own accord as a conversation continues, most often to reflect the first line of a reply. The extension keeps watching for that and reasserts the document title each time it happens, rather than setting it once and letting a later rename slip through unnoticed.
The address bar is left untouched, deliberately
the address bar, which still reads claude.ai or chatgpt.com. This is not an oversight, and it is not a feature waiting to be added later. A version that rewrote the path to look like a document address was built and shipped, and it was removed once the problem it caused became clear.
Both assistants read their own web address to work out which conversation is open and where to send your next message. With a false path in place, that lookup failed, and sending stopped working entirely. The disguise looked complete and the product underneath it did not function, which is the wrong trade for a tool meant to be used, not just looked at.
A domain cannot be changed by any extension, working or not
What it hides, stated plainly
Three surfaces change, and only three. Everything above this list is covered. Everything below it was never in scope.
- the visible page, redrawn as a document editor
- the browser tab title
- the tab icon
What it does not do
A visual disguise changes what somebody looking at your screen sees, and nothing beyond that. Read this list before installing, not after, so nothing here is a surprise later.
- the address bar, which still reads claude.ai or chatgpt.com
- your browser history
- anything a workplace network, proxy, DNS log or monitoring agent can see
- anything a managed or employer-administered device can see
- the conversation itself, which OpenAI and Anthropic still receive and store
- anything captured by screen recording or proctoring software
- the page source, which anyone opening developer tools can read
Common questions
Does the overlay change what ChatGPT or Claude actually receives
No. Every character you type is relayed into the real message box running underneath the overlay, the same box that would be there without the extension, so the assistant gets your message exactly as it always would.
Can the real interface still be reached while the disguise is on
Yes, at any moment. Shift and Escape switches straight to the real page and back, and the popup offers the same switch through a click. Nothing about the conversation depends on which view happens to be showing.
Does it work the same way on Claude and on ChatGPT
The overlay is tailored separately to each site, since the underlying page structure of claude.ai differs from that of chatgpt.com, but the result is the same document view on both, plus chat.openai.com.
Why not disguise the web address as well, the way the page itself is disguised
That was tried. Both assistants read the address to work out which conversation is open and where to send the next message, so rewriting it broke sending. It also would not have hidden the domain, which no extension can change.
Does the tab title stay changed for the whole conversation
Yes. Both assistants rewrite a tab title repeatedly as a reply streams in, and the extension keeps checking for that and reasserts the document title each time, rather than setting it once and leaving it exposed later.