A Chatbot Audit for the Widget That Interrupts Everyone

What We Keep Finding

Chat widgets are guests that act like owners. They load late, plant a launcher over your content, sometimes grab focus on arrival, and open panels that hold the keyboard with no way out. New messages arrive announced loudly, or not at all. Both are wrong, and the middle setting has a name, because a live region can be told to interrupt or to wait, and told to read the whole message or only the part that changed.

The keyboard question is more forgiving than most people think, and the answer surprises teams in a useful direction. A panel that needs Escape to release focus is not automatically a failure. The rule allows an exit that needs an unusual key, as long as the user is told what that key is. So "press Escape to leave the chat" turns a failure into a pass. What has no defense is a panel with no exit and no instructions, and that is the version we keep finding.

Here is why it is worth caring even though the widget is not yours. Keyboard traps are one of a small handful of rules that apply to everything on the page, including content you never chose to rely on. A trapping chat widget does not just fail its own corner. It stops the whole page conforming, at any level, no matter how good the rest of your work is. Third-party does not mean out of scope. It means somebody else's code is holding your page's result.

The focus-ring question is worth getting right too, because it is usually overclaimed. The rule at the level you are being held to says no part of the focused control may be entirely hidden. A launcher that partly overlaps a focused link passes. A launcher that sits fully on top of it does not. We test to the rule rather than to the stricter version, so you do not spend a sprint on something nobody is asking for.

Because the widget is third-party, the audit's real product is proof. Precise, dated evidence of what it does inside your pages, ready to send the vendor, plus every fix your own configuration allows today.