An AI chatbot for Webflow that respects the design

Webflow occupies a particular niche: sites built by designers and studios, often for clients, where the visual result is the point. That context changes what a chatbot has to get right. It cannot fight the design, it cannot require rebuilding anything in the Designer, and on client work it cannot advertise a tool the client did not choose. Those constraints rule out a lot of the chat widgets on the market before the conversation quality is even considered.

Installation should be the least interesting part. Paperchat is a script tag pasted into Project Settings, Custom Code, in the footer section, applied when you publish. Because it lives outside the canvas it cannot interfere with layout, breakpoints or interactions, and it survives page duplications, template swaps and full redesigns. Handover to a client does not involve explaining where a chat component is hiding.

Training is where Webflow sites have an advantage. A well built Webflow project usually has real substance in its CMS collections, case studies, service pages, help articles, that visitors rarely dig through. Paperchat reads published pages including collection pages, so that material becomes answerable in a sentence. You can add a sitemap to ingest the site in bulk, upload documents for anything unpublished, and paste in the specifics a client tells you on a call but never puts on the site.

Appearance is configurable from the dashboard rather than in code: colour, bot name, avatar, subtitle and welcome message, with branding removal on higher plans. For studio work that last point is the difference between a chatbot that reads as part of the build and one that reads as a bolt on. Each site also has a domain allow list, so a script tag copied from a client's page cannot be reused on someone else's site.

The recurring value is worth planning for if you build for clients. A chatbot degrades as content changes, so retraining is natural retainer work, and the conversation log is the most honest research a client will see: the questions their visitors actually asked, in their visitors' own words, and the pages that failed to answer them. That turns a one off deployment into a monthly conversation about content, which is usually where the next piece of work comes from.