Visual Improvements and clarity - Kian Widget
This release focuses on transparency and trust. Users now see exactly what Kian is doing as it reasons through their question — and the widget becomes a navigation tool, not just a chat box.
The WTM widget has been rebuilt from the inside out. The thinking animation makes the AI feel deliberate rather than instant. Source cards replace the old pills — clicking navigates in-page, keeping the conversation alive. And Kian now detects when users are frustrated or repeating themselves, escalating naturally to a Gratis Erstgespräch.
On the intelligence side, this update adds tighter prompt rules, multilingual detection, rich page context injection, and a referrer-aware greeting — adapting its opening line based on whether the visitor came from Google, LinkedIn, or direct.
Thinking animation
Skeleton bars appear while Kian processes — latency feels intentional, not broken.
The thinking state uses skeleton bars that pulse while the AI reasons. Progress stages cycle through: 'Analysiere Kontext…' → 'Suche passende Seminare…' → 'Formuliere Antwort…'. Previously the widget showed nothing during processing — users couldn't tell if their message had been received.
Source cards replace source pills
In-widget navigation — clicking a source scrolls the page, widget stays open.
Pills gave no context about what the source contained. Cards show the seminar title and content type, making the link trustworthy. Clicking navigates to the anchor on the current page — the widget overlay stays open so the conversation continues.
Smart escalation to Gratis Erstgespräch
Detects frustrated or repeated questions — opens an in-chat booking overlay instead of cycling.
The prompt scores frustration signals (repeated semantic questions, negative phrasing). Above a threshold, Kian stops trying to answer and offers the Erstgespräch via [ACTION: book] — triggering the in-chat booking calendar. No page navigation. The conversation stays alive.
Multilingual & referrer-aware greetings
English visitors get English. Google/LinkedIn referrers get a context-adapted opener.
Language is detected from the user's first message and persists for the session. Page context injected on load includes document.referrer — the prompt adapts the greeting when it sees google.com or linkedin.com in the referrer.