Here's a fast way to check whether a page of documentation will make a good chatbot answer, before you connect anything: pick the question it's meant to answer, then try to answer it yourself using only two sentences pulled directly from the page. If you can't, your chatbot won't be able to either.
Try it on a real page
Open any page in your docs and ask: what question is this actually answering? Now find the two sentences that answer it. Not a summary in your own words — the literal sentences already on the page.
Passes the test:
"Refunds are available within 30 days of purchase for annual plans. Monthly plans are not eligible for refunds but can be canceled at any time."
Two sentences, direct, no scavenger hunt required. A chatbot retrieving this chunk has exactly what it needs.
Fails the test:
"We want every customer to have a great experience with our product, and we're always working to improve. Our policies are designed with fairness in mind. For more information about specific scenarios, please refer to the table below or contact support."
No sentence here actually states a policy. A chatbot retrieving this chunk has nothing to answer from except vague reassurance — which is exactly the kind of page that produces a hedge, a guess, or an unnecessary escalation instead of a real answer.
Why two sentences, specifically
It's not an arbitrary number — it roughly matches how much a well-formed retrieved chunk can usefully carry for one specific question. If the real answer takes five sentences of buildup before it says anything, that's a sign the answer is buried, not that the topic is inherently complicated. If it can't be stated in two sentences no matter how you rewrite it, that's a sign the actual answer doesn't exist on the page yet — a gap to go fill, not a formatting problem to fix.
Running this across a whole doc set
You don't need to test every page by hand. A faster version: point your chatbot at your own docs and ask it the real questions customers ask. Every vague, hedging, or wrong answer it gives back is the two-sentence test failing on some page — just discovered automatically instead of manually.
The takeaway
Good chatbot-ready writing and good human-readable writing aren't different skills, just the same one held to a slightly stricter standard: say the actual answer, in as few sentences as it honestly takes, near the heading it belongs to. Everything else in writing docs AI chatbots answer well from is really just this test applied page by page.
Related reading
- Writing docs AI chatbots actually answer well from — the fuller set of formatting practices this test is a shortcut for.
- Chunking and embeddings 101 — why chunk size and clarity affect retrieval, not just readability.
- We pointed Lumen Chat at our own docs — running this test at scale, on real content.