Five lessons, about an hour. A brand's knowledge base is where the answer to the fortieth "does this ship to Kenya" question should already live, written once and well, rather than reinvented every shift. This module is about building and keeping that document honest — because module 7's AI chatbot is only as good as what it's trained on.
The fifth time you type essentially the same answer to essentially the same question, that's the signal — not a coincidence, a document that doesn't exist yet. Waiting for a "good moment" to write it down means answering it a sixth, seventh and twentieth time by hand.
If you've answered it three times this week, it belongs in the knowledge base, not in your memory.
Not a rough note to yourself — a version another rep, or a customer directly, could read cold and understand. Written with a stopping point: the article answers the question fully and then stops, rather than trailing into every edge case that could theoretically come up, which is how help-centre articles become unreadable walls of text nobody finishes.
"Yes! We ship to Kenya via DHL, typically 8–12 working days, and duties are included in the checkout price so there's nothing extra on delivery." — turned into a help-centre article titled "Do you ship to Kenya?" with exactly that answer, so it's now a link, not a retype.
Think of a question you've personally answered more than once recently, anywhere. Write the documented version, with a clear stopping point.
A help centre nobody has audited in months quietly accumulates a specific kind of danger: articles that were correct when written and are now wrong, because the policy behind them changed and nobody updated the page. A customer who reads an outdated return window and is then declined trusts the brand less than one who never found an article at all.
Is this still factually correct against current policy? Is anything the brand now offers or does missing entirely? And — the one most audits skip — does this article actually match what reps are saying in real tickets, or has practice quietly drifted from what's written?
Help-centre article: "Returns accepted within 30 days." Actual current practice, found across recent tickets: reps have been quietly extending to 35 days for repeat customers without anyone documenting it as policy. The article isn't wrong exactly — it's just no longer what's actually happening, which is its own kind of problem.
Pick any real, publicly available help-centre article. Check it against the three questions and note anything that fails.
Nobody reads a help-centre article for fun. They arrive because something's wrong and they want it fixed, often mid-frustration — which means an article written like a calm reference manual, full of caveats and subclauses, fails the actual reader even when every word in it is accurate.
Lead with the direct answer in the first line. Follow with the steps, numbered, one action per line. Put exceptions and edge cases at the end, clearly separated, for the smaller number of readers who need them — not woven through the main answer where they slow everyone else down.
"Returns are generally accepted within a window of time, though this can vary depending on the item purchased and whether it was part of a promotional sale, and customers should be aware that certain conditions may apply before initiating..."
"You can return most items within 30 days of delivery. 1) Email us your order number. 2) We'll send a return label. 3) Refund lands in 3–5 days once received. Sale items are final — see exceptions below."
Rewrite a caveat-heavy paragraph — real or made up — into the stress-readable shape.
A macro and a help-centre article answering the same question are two copies of the same fact, and two copies drift apart the moment one gets updated and the other doesn't. A customer who reads the help centre and gets a different answer from a macro-based reply notices immediately, and it reads as the brand not knowing its own policy.
Whenever a policy changes, update the source article first, then treat every macro referencing that fact as needing a check — not an assumption it's still fine. A short running list of "what references this policy" for anything genuinely load-bearing (a return window, a shipping zone) makes this a five-minute task instead of a search-and-hope.
A chatbot trained on an outdated help centre will confidently repeat the outdated answer. Keeping the source documents current isn't just a tidiness habit — it's what module 7's chatbot supervision depends on.
List three places (macros, articles, a pinned message) that would need updating if your brand's return window changed tomorrow.
Documenting an answer well solves the support side of a repeated question. It doesn't solve the underlying cause if the repeated question is actually a symptom — five broken-pump complaints this month isn't a knowledge-base gap, it's a supplier problem that a founder needs to know about, and support is often the first place the pattern becomes visible at all.
Frequency rising rather than flat, a shared specific cause rather than five unrelated complaints that happen to look similar, and a cost to the business beyond support time — refunds, lost customers, reputational risk. A pattern report names the specific tickets, the shared thread, and a plain recommendation, not just a vague "we're getting a lot of complaints about this."
"Flagging: three broken-pump complaints this month (orders #2288, #2299, #2311), all the same 250ml rose clay line, all reported within days of arrival — that's up from zero last month. Recommend checking with the supplier on this specific batch before more ship."
Write a pattern report for three related complaints, real or made up, using the worked example's shape.
Bumi Naturals' help centre is supplied, along with its current actual policy and a batch of recent tickets. The return-policy article is stale, two articles are missing entirely, and one macro contradicts what the help centre says.
Whether you can treat a knowledge base as a living document that needs maintenance, not a one-time writing task — the same discipline that keeps a brand's self-service answers actually trustworthy.
You should now have: a habit of documenting repeated questions, an audit method, a stress-readable article shape, a sync-tracking habit, and a pattern-report template. Module 7 turns to handing this knowledge base to an AI chatbot as your junior — and catching it when it invents an answer the documents never actually gave it.