Writing Knowledge-Base Entries an Auto-Responder Can Actually Quote

A knowledge base can be useful to a person, but it may not work well for an auto-responder. When people read a document, they often skim, infer information, and jump around. An auto-responder, on the other hand, quotes directly from the entry, which can result in losing important context. Writing entries for an auto-responder requires a different approach than writing traditional documentation, and it begins with breaking down the information into smaller, more granular pieces rather than focusing on style.

Before polishing any entry, it's important to run a simple test: if the auto-responder lifted one paragraph from the entry and sent it alone, would it fully answer the question it matched? If the paragraph depends on information from another part of the entry or continues a thought that is split across sections, the answer is no. The customer would receive a fragment that sounds confident but explains little.

Every paragraph intended for outbound replies should be able to stand on its own. If it can't, it should be rewritten until it can, or moved into internal notes that the auto-responder is not allowed to quote.

When writing entries for an auto-responder, each entry should have a single purpose: to answer one specific question. Each paragraph inside the answer should carry a single fact. Comparing two versions of a shipping entry, for example, can illustrate this point. The first version reads: "Standard shipping takes several business days, express costs extra, and oversized items ship separately."

A customer asking about express delivery would receive a single paragraph that contains only a portion of the answer, and the auto-responder would have to guess which part to include. By breaking down the entry into three separate paragraphs, each paragraph becomes a quotable unit, and the auto-responder can select the appropriate one.

Atomic entries, which are broken down into individual facts, also age better. When a price or transit time changes, you only need to edit one entry instead of untangling a long policy essay. The rest of the library remains untouched. During a walkthrough of desktop auto-reply setup, a guide explicitly describes this approach: product lists are converted into atomic fact cards covering item, price, spec, and policy.

Any draft reply that doesn't match a verified card is blocked from sending. This safety check is only possible once the entries themselves are granular.