Webflow AI assistant that guides visitors before they bounce

Webflow attracts a specific type of builder. Designers who care about type hierarchy and spacing. Agencies that ship six-figure rebrands with custom interactions and Lottie hero scenes. In-house marketing teams at B2B SaaS companies who chose Webflow precisely because they wanted control over every pixel without fighting a developer queue. Brand sites for studios, fashion labels, boutique hospitality groups, and founder-led consultancies. What these projects share is a bias for design excellence — and an allergy to anything bolted on that looks like it came from a different era of the web. That cultural constraint shapes what kind of conversational layer actually belongs on a Webflow site: it cannot be a generic square speech-bubble widget with default blue styling, and it especially cannot be the kind of intrusive drive-by popup that breaks the scroll reading experience someone spent weeks refining.

The word assistant is doing real work in the name Webflow AI assistant. A chatbot, in the public imagination and honestly in much of the product category, is a reactive surface — a bubble in the corner that sits there waiting for a visitor to click it. Most visitors never do. A well-built assistant behaves differently. It watches the visitor's path through the site, reads context clues about intent, and offers help at the specific moment when help is most likely to be welcomed rather than tolerated. Someone who's been idling on a pricing page for forty seconds after scrolling past the second tier is telling you something about their decision process. An assistant that gently opens with "want me to walk through how the CMS plan's limits compare to Business for a blog-heavy build?" converts that moment. A dumb bubble in the corner does not. That's the difference this product leans into.

Installing on Webflow is genuinely pleasant compared to most platforms, because Webflow's Custom Code surface is documented, discoverable, and behaves predictably. You open Site Settings from the gear icon next to your project name, navigate to the Custom Code tab, and find two text fields labeled Head Code and Footer Code. Paste the Asyntai snippet into Footer Code rather than Head Code, because assistants belong after paint — you want the designer's hero render to happen first, then the assistant to initialize quietly underneath. Save the panel, then publish the site from the Designer top bar or from your usual staging-to-production flow. That's the entire install path. The snippet itself is a standard async script tag, so it plays nicely with whatever else is already in your Custom Code — Memberstack, Outseta, Finsweet attributes, GA4, Segment, Hotjar, a pixel or two. Order of loading is handled by async semantics rather than manual coordination.

Plan gating is worth being candid about because Webflow's pricing structure confuses people who haven't shipped on it recently. Custom Code site-wide is available on CMS, Business, and Enterprise Site plans, not on the free Starter plan. If a client's project is still on Starter, the assistant will preview fine on the .webflow.io staging URL but won't carry over when the site goes live on a custom domain until the plan is upgraded. This is a Webflow constraint, not an Asyntai constraint, and it applies equally to every third-party script anyone would want to site-wide on Webflow — pixels, chat tools, analytics, A/B testing platforms. In practice, almost every Webflow project that's serious enough to want a conversational assistant is already on CMS or Business for reasons that have nothing to do with our product: they want blog collections, team-member dynamic lists, or site search. So the plan requirement rarely surfaces as a real obstacle.

What the assistant knows about your site starts with a crawl of your published URLs. Drop your Webflow domain into the Asyntai dashboard, and the indexer walks through every reachable page — Home, About, Services, Work, Blog, individual case-study pages generated from a CMS template, individual blog posts, the careers page, pricing, FAQ, contact, privacy, terms. The index includes both human-written static pages and dynamically-generated CMS collection pages, because Webflow renders those as real HTML to crawlers rather than hiding them behind client-side rendering. That matters: a typical agency site might have fifty static URLs but three hundred CMS-rendered pages across a blog and case-study library. An assistant that only read the static pages would miss most of the substance. One that reads the CMS output reflects the actual published site.

Private knowledge rounds out what the assistant can reference. There's always material you don't publish on a client-facing Webflow site — internal pricing ladders, playbooks for how discovery calls should flow, the SOW template language you use for retainer scopes, the brand voice guide that keeps junior writers on track. You upload those PDFs or paste them as text into the Asyntai dashboard, and the assistant draws on them privately when answering visitor questions. Behavioral rules layer on top in plain English. "If a visitor asks about project timelines, always ask what their launch date is first." "If someone mentions they already have a Webflow site and want a redesign, prioritize the rebrand-and-migrate case study in your answer." "When someone says they're evaluating Webflow versus another platform, stay neutral and link them to the platform comparison post." Rules like that get followed consistently across every conversation, in every language.

For agencies running Webflow as a service, the economics of this feature are usually obvious the first time you look at an assistant's conversation log after a week. The top question category is almost never what the team predicted. It's not "what's your pricing" or "do you work with startups" — those get answered by the site itself. It's oddly specific lead-qualification questions that were invisible before: "do you build with Memberstack," "can you migrate from Squarespace without breaking SEO," "do you offer ongoing retainers or just project work," "can you work with a technical founder who already has a brand guide." Every one of those questions represents a qualified prospect who would have clicked away rather than fill out a four-field contact form with a freeform message box. The assistant turns that friction into a short dialogue, captures the email address at the natural moment, and routes the full context to wherever you work from.

Webflow Ecommerce deserves its own paragraph because it changes what the assistant is doing during a session. Commerce on Webflow tends toward direct-to-consumer brand stores — small-batch skincare, independent fashion, design objects, curated homeware — rather than sprawling marketplaces. Catalogs in that range sit comfortably inside an assistant's working knowledge, so shoppers can ask about specific product details, shipping zones, return windows, or sizing language and get answers grounded in the actual product copy on the store. The assistant doesn't touch the Webflow Ecommerce checkout flow, doesn't rewrite product pages, doesn't alter inventory — it only sits alongside the store experience and fields the questions that would otherwise cause hesitation or abandonment. That separation of responsibilities is intentional. Checkout is sacred; the assistant's job is to deliver a visitor to checkout with fewer unanswered questions.

Membership layers are common on Webflow, and the integration story there matters. Memberstack and Outseta are the two names that show up most often when Webflow projects need gated content, paid memberships, or authenticated dashboards. On Standard and Pro plans, Asyntai reads a window.Asyntai.userContext object that you populate right after your membership SDK finishes initializing the session. That object can carry name, email, plan tier, signup date, renewal date, and any custom field you already have — the assistant then personalizes its replies against that identity. A logged-in paid member asking about an upgrade gets handled differently from an anonymous visitor asking the same question, without the assistant having to re-collect information the member already provided during signup. The integration is a few lines of JavaScript alongside your existing Memberstack or Outseta boot code, dropped into the same Footer Code panel where the Asyntai snippet lives.

Design integration with the widget itself is worth calling out because Webflow designers are particularly sensitive to visual mismatches. Every surface aspect of the Webflow AI assistant configures from the Asyntai dashboard with live preview — the primary accent color to match your brand palette, widget corner radius to align with the rest of your component system, font family selection to echo the site's typography, welcome-bubble copy, avatar imagery, corner position, open/closed default state. The widget is designed to look like it belongs on a design-led site rather than imposing a generic chat aesthetic. Designers tend to iterate on the palette and radius for ten minutes, screenshot it against the live Webflow page, and lock it in — the same muscle memory they use when fine-tuning a Webflow component.

Analytics on the Asyntai side reveal patterns that a Webflow dashboard on its own won't expose. Which pages trigger the most assistant conversations. What time of day visitor questions cluster. Which topics come up repeatedly (which is a direct signal about what your site copy is silently failing to communicate). How often a conversation converts into a captured lead versus trails off. How long the average conversation runs. Which languages your international traffic actually speaks. None of this data is available inside Webflow Analytics, which tops out at traffic and form fills — the assistant layer adds a qualitative analytics dimension that's arguably more useful for iterating on the site than raw pageview counts.

Pricing is deliberately shaped around how Webflow projects grow. Free covers 100 messages a month against a single Webflow site — enough to prove the concept on a staging domain or handle a quiet portfolio. Paid plans open at $39 per month for 2,500 messages and two sites, which suits a solo designer running their own site plus one active client project. Mid-tier steps cover three sites with the Standard plan and support User Context for membership-aware replies; the Pro plan covers up to ten separately trained assistants, which maps well to a mid-size agency's active client roster. Before you hit a monthly cap, warning emails arrive at sensible thresholds so a sudden traffic surge from a Product Hunt feature or a newsletter mention doesn't silently cut off the assistant mid-month.

A short setup pass for a Webflow AI assistant: register a free Asyntai workspace, copy your snippet, open your Webflow project in the Designer, click the gear icon next to the project name, open Custom Code, paste the snippet into Footer Code, save. Back in Asyntai, point the crawler at your Webflow domain and wait a minute or two while the assistant reads every published URL. Upload private PDFs — pricing ladders, playbooks, brand guides. Write three or four behavioral rules in plain English. Color-match the widget to your design system. Publish the Webflow site from the Designer. Run a handful of test questions from the live URL to confirm tone and accuracy. Switch on lead notifications. From that point forward, the Webflow AI assistant quietly does the conversational work your contact form was doing badly — guiding visitors toward the information they came for, recognizing the signs of a qualified prospect, and handing you a named lead with the whole conversation attached.