January 8, 2026 · 5 min read
A Client Asked for a Chatbot, What They Actually Needed Was a Better FAQ Page
A client wanted an AI chatbot for support deflection; a quick look at their actual support tickets showed the real problem was a buried, outdated FAQ page nobody had touched in two years.
A client came to me wanting an AI chatbot on their support page. The pitch was reasonable on its face — support volume was up, response times were slipping, and a chatbot sounded like the modern fix. Before scoping it, I asked for three months of support ticket transcripts, which is a habit I'd recommend before anyone builds anything AI-shaped.
What the tickets actually showed
About 60% of tickets were questions with answers that already existed on the site — in an FAQ page that hadn't been updated in two years, buried four clicks deep in a footer link, written in a format that answered questions nobody was currently asking. Customers weren't failing to find an answer because no AI was there to help them search. They were failing because the actual content was stale and badly organized.
A chatbot on top of a bad FAQ page is just a more expensive way to fail to answer the same question.
What we built instead
We rewrote the FAQ page based on actual ticket frequency — the most common questions moved to the top, answers were rewritten in the customer's own phrasing pulled straight from how they'd asked the question in tickets, and we added a simple search box with fuzzy matching. No LLM involved. Ticket volume for FAQ-coverable questions dropped by roughly 35% in the following quarter.
- Pull real support data before scoping any AI feature — the request someone makes is rarely the diagnosis.
- A chatbot adds a conversational layer on top of content; if the content underneath is wrong, the chatbot just says the wrong thing more confidently.
- A plain, well-organized FAQ page is cheaper, more auditable, and in this case more effective than the thing that was asked for.
We did eventually revisit the chatbot idea, about six months later, once the FAQ was solid and the remaining ticket volume was genuinely the kind of nuanced, account-specific question a chatbot could help triage. It shipped faster and worked better because it had something real underneath it.