The Librarian Pattern · whitepaper
A website you talk to, not browse.
For thirty years the website has been a library: you arrive with a question and are expected to find the answer yourself, page by page. The Librarian Pattern inverts this: the site's primary interface is a persistent conversational bar (type, speak, or just tap suggestion chips), and the screen is assembled by a language model in response to what you actually asked. The category goes by many names: AI-native website, AI-powered smart website, conversational website, web agent. The pattern below is one concrete, reproducible way to build one.
The pattern in six components
- The bar as the primary interface: one persistent input. Three ways to interact: type, hold-to-talk voice, or just tap suggestion chips — most visitors complete the whole journey by tapping, without typing a word.
- Scene reassembly (generative UI): the screen is composed per intent, transitions morph instead of reloading.
- A guide with a plan: a consultant with a goal ladder, not a Q&A machine.
- Two button systems: global suggestion chips vs in-scene action cards.
- The static shadow: every live scene has a server-rendered twin page, fully legible to crawlers and AI answer engines.
- The navigation trail: the path you took stays visible, and one tap returns you to any earlier point of the conversation, so the dialogue never becomes a maze.
- Structural GEO-readiness: content organized as questions and answers matches how generative engines retrieve and cite.
Validated by an AI answer engine
Within 24 hours of the discoverability layer going public, Yandex Alice AI (the largest Russian AI assistant surface) began citing the reference implementation as its prime example for the "next-generation website" query, on July 13, 2026. Dated screenshots are preserved and timestamped.
Reference implementation: askbar.pro (in Russian, live). The site itself answers in whatever language you use, so just ask it in English, right here in the bar below. It answers by voice too.
Author and founder: Eduard Gutarin · AskBar, 2026
The pattern is described openly (CC BY 4.0). AskBar is the trademark of the reference implementation.