Connecting a source is the easy part. How well the chatbot answers depends on how your content is written — the same document can produce a great answer or a vague one depending on its structure.
Answer one question per section
A chatbot retrieves small chunks of your content to answer a question, not the whole document. A page that mixes pricing, setup steps, and troubleshooting in one long paragraph forces the chatbot to guess which part is relevant. Clear headings that each cover one topic make retrieval far more accurate.
Say the thing directly
"See the table below for details" works for a human skimming a page. It doesn't work for a chatbot, which retrieves text, not layout. State the actual answer in words near the heading it belongs to, even if you also have a table or diagram alongside it.
Keep it current, not just correct
An accurate-but-outdated doc is worse than an honest gap, because the chatbot will confidently repeat it. If a feature changed, update the doc that describes it — don't leave the old explanation around "for reference," since the chatbot has no way to know it's stale.
Don't be afraid of short pages
A focused one-paragraph FAQ answer often chunks and retrieves better than a single page trying to cover ten related questions. If a topic has several distinct sub-questions, separate pages or clearly separated headings usually outperform one dense page.
Let the gaps show up naturally
Turn on fallback and escalation in your widget settings so questions your docs don't cover get routed to a human instead of guessed at — the resulting conversations are a direct list of what to document next.
Good chatbot answers aren't a separate writing exercise from good documentation — they're the same thing, just held to a slightly higher standard of clarity.
Related reading
- Turning your docs into an AI chatbot — where this fits into the overall setup.
- Chunking and embeddings 101 — why formatting choices affect retrieval, not just readability.
- We pointed Lumen Chat at our own docs — this process, run on ourselves.