AI Chatbots That Don't Embarrass You
A practical guide to integrating AI assistants into client sites — what works, what doesn't, and why most are broken on day one.
A practical guide to integrating AI assistants into client sites — what works, what doesn't, and why most are broken on day one.
Most AI chatbots on agency sites are doing more damage than good. They hallucinate prices. They give confidently wrong answers about services. They make the brand look like it's bolted-on technology rather than thoughtful integration.
Here's how to ship one that actually helps.
Before any code, decide one thing: is the bot answering questions or capturing leads?
These need different architectures.
Answering questions (FAQ-style support): needs reliable retrieval from a known document set. You're not letting the AI improvise — you're letting it summarize sourced material.
Capturing leads (qualify and book): doesn't need an LLM at all most of the time. A 3-question flow with conditional branching is more reliable, faster, and cheaper.
Most agencies build the wrong one for the use case.
Three patterns we see repeatedly:
Our standard agency chatbot pattern:
We tell clients: a chatbot is a feature, not a strategy. It should reduce easy questions so the team can focus on serious leads — not replace the team.
Thinking about adding AI to your site? Talk to us first — we'll help you decide if you actually need it.
Tell us about your project. We reply within 24 hours with thoughtful questions, not generic responses.