AI Widgets

What's included

Build

• Widget scoped to one specific job, decided in a short discovery call before anything gets built
• Custom React or vanilla JS component that embeds with a single script tag, with an iframe option if you want isolation
• Server-side API proxy so your keys never reach the browser
• Streaming responses, so the widget shows output as it generates instead of spinning for eight seconds
• Structured output where the answer drives the interface, using JSON parsed into components rather than a wall of text
• Mobile layout and dark mode handling
• Fallback behavior for API failures, rate limits, and timeouts, so the widget degrades instead of disappearing

Prompt and model work

• Prompt design and iteration against real inputs from your audience
• Model selection based on what the task actually needs, since most widgets run fine on a fast cheap model
• Guardrails against off-topic use and prompt injection, including the people who will try to get free homework help out of your ROI calculator
• Input validation and output constraints so the widget can't say something that creates a problem for you
• Caching on repeated inputs to cut cost

Lead capture and data

• Gated results, progressive capture, or fully open, depending on what your funnel needs
• Lead push into HubSpot, Zoho, or whatever CRM you run, with UTM values and click IDs preserved
• Event tracking into GA4 and PostHog covering starts, completions, drop-off by step, and submissions
• Conversion event wired to your ad platforms so paid traffic can optimize toward it

Operations

• Rate limiting per session and a monthly spend cap
• Abuse protection and bot filtering
• Admin view for prompt changes without a redeploy
• White-label build if you're deploying this under your own brand across client sites

Why it matters

Most AI widgets on marketing sites are chatbots wearing a costume. They take an input, produce a paragraph, and give the visitor nothing they couldn't have found in ten seconds elsewhere. The ones that work are narrow. A pricing estimator that asks four questions and returns a number. A qualifier that tells someone quickly whether your service fits their situation. An assessment that scores their setup and shows what's broken.

The specificity is what makes the lead worth having. Someone who answered six questions about their ad spend before handing over an email is a different prospect from someone who downloaded a PDF. Tracking matters as much as the build here, because a widget you can't measure is a demo you're paying to host.