In a documentation context, deflection happens when a user finds the answer in the docs, a search result, or a docs chatbot instead of writing to support. The question still gets asked; it just gets answered without a human agent.
AI chatbots on docs sites can reduce tickets by answering in context before a user contacts support. Measure the result with conversation outcomes and follow-on ticket volume; grounding and source links improve trust but do not prove deflection on their own.
Deflection is measured with deflection rate: the share of questions resolved by self-service. Tracking it requires knowing both how many questions the assistant resolved and how many of those users still ended up in the support queue.