Knowledge base
The knowledge base is the ceiling. No configuration makes an assistant answer something that is not in it.
Why it matters
Teams debug assistant quality by adjusting prompts and settings when the actual cause is almost always coverage. If the answer is not in the knowledge base, no amount of tuning will produce it.
- Refusals on questions that are genuinely undocumented
- Time spent tuning configuration instead of writing content
- Stale pages producing confident, wrong, cited answers
What belongs in it
Anything you want answerable: help articles, policy pages, product documentation, and uploaded documents for material that is not published. What matters is coverage of the questions people actually ask, which the gap report tells you directly.
Stale content is worse than missing content
A missing answer produces a refusal, which is honest. An outdated page produces a confident, cited, wrong answer — which looks more trustworthy precisely because it has a source attached.
Structure affects retrieval
One topic per page, headings phrased as questions, and the answer stated near the top all improve retrieval measurably — and none of them make the content worse for a human reader.
Common questions
- How much content do I need to start?
- Enough to cover your common questions. A small, accurate, current knowledge base outperforms a large stale one, because the stale one produces wrong answers with citations.
- Should I write content specifically for the assistant?
- Write it for humans, structured well. The practices that help retrieval — direct answers, clear headings, one topic per page — are ordinary good writing.
Related terms
From the blog
All posts- BuildingKnowledge ops for a two-person teamA support bot on a tiny team survives on refusal lists, ten-minute recrawls, and docs edits tied to real tickets — not on a KM platform nobody maintains.Read
- BuildingAgencies: one bot, many client knowledge basesHow agencies should run AI support across clients: separate knowledge bases, domain locks, and evaluation per property.Read
- MeasuringDeflection rate is the most overstated number in support automationCounting bot conversations as deflected tickets overstates the result. The honest version is a before-and-after on ticket volume, read next to satisfaction.Read
Answer honestly. Capture the rest.
Point Matter Chat at your site and see what it can — and can't — answer. It's honest about both.
No credit card. 2 minute setup.
Every answer cites the source it came from. When there isn't one, it says so — and hands the visitor to you.
Installs on the tools you already run.