A support chatbot should only say what you approved.
Public chatbot answers come only from verified public-safe collections. The bot refuses when there is no approved source. Citations go out through Wiki Connect or the Answer Playground.
Keep support chatbot knowledge scoped to verified public-safe collections. Refuse when there is no source, and cite answers through Wiki Connect or the Answer Playground.
The refund question is where support chatbots break.
A support chatbot that can see every Google Drive folder will eventually invent a refund, a shipping exception, or a warranty the company never published. The model is not the root cause. The pile of unreviewed documents is.
Refund policy answers are the outbound wedge. Customers ask a high-stakes question. The bot either cites the approved public policy or it guesses. Guessing is how you get a chatbot invented refund in a transcript that finance cannot unwind.
The same failure shows up on billing, cancellation, and SLA questions. If the collection includes draft pages, internal macros, or last quarter's policy, the bot will retrieve something that looks close and speak with confidence.
What to do before the bot talks to customers
Treat support chatbot knowledge as a governed public surface, not a folder you pointed a model at.
- Step 1
Import the public policy, not the whole drive
Start with the refund, billing, and help pages customers are already allowed to see. Wiki can import a URL so the source is the same page you already publish.
- Step 2
Verify before the chatbot may use it
A source sitting in a collection is storage. A verified source with an owner is something a public bot may cite.
- Step 3
Keep it in a public-safe collection
Internal macros, escalation notes, and draft exceptions stay in internal collections. The public chatbot never gets a master key.
- Step 4
Attach an AI Application policy
Permission-aware retrieval means this chatbot may use the public-safe collection and must refuse when nothing approved matches the question.
- Step 5
Test the refund question in the Answer Playground
Ask the real customer questions before launch. Inspect citations. Confirm the bot refuses when the policy does not cover the case.
How Wiki keeps support chatbot knowledge public-safe
Wiki is the management layer for verified knowledge, permissions, and proof. These are live capabilities, not a planned helpdesk app.
Verified sources
Documents go through review before they are allowed for AI use. Draft and unverified pages stay out of answers.
Public-safe collections
Customer-facing applications stay scoped to collections you mark public-safe. Internal notes stay internal.
AI Application policies
Each chatbot gets a policy for which collections it may retrieve. Permission-aware retrieval enforces that policy at answer time.
Answer Playground
Test questions and inspect citations before customers see the answers. Save the refund, cancellation, and billing cases you care about.
Wiki Connect
An HTTP API for policy-constrained, citation-backed answers from your own bot. If nothing is approved, the API can refuse instead of guessing.
Knowledge Health
A workspace score for review cadence, coverage, and whether knowledge is actually ready for AI.
Questions teams ask first.
- Can Wiki stop a support chatbot from hallucinating?
- No tool can guarantee zero hallucinations. Wiki requires source-backed answers, refuses when nothing approved matches, and keeps unverified or internal content out of public bots. That is the control you actually have.
- How does a chatbot refund policy stay accurate?
- Import the public refund policy URL, verify it, put it in a public-safe collection, attach the chatbot's AI Application policy, then test the refund questions in the Answer Playground. Serve the cited answer through Wiki Connect, or refuse.
- Does Wiki replace my existing support chatbot?
- No. Wiki governs the knowledge the chatbot may use. Connect your own bot through the Wiki Connect HTTP API, or test answers in the Playground first.
- What happens when there is no source?
- The application should refuse. A missing policy is not a prompt problem. It is a knowledge gap you can see in the Playground and close with a verified source.
Essays, tools, and the product.
From the blog
The Complete Guide to Citation-Backed AI Support
Customer-facing AI should show where the answer came from, know when it lacks an approved source, and escalate when documentation cannot support a reliable answer.
Why Your AI Chatbot Keeps Giving Different Answers
Inconsistent chatbot answers usually come from fragmented sources, stale docs, conflicting policies, and missing governance, not random model behavior.
RAG Is Not a Source of Truth
Retrieval can find context. It does not decide whether that context is current, approved, allowed for the audience, or safe to cite.
Tools
Your AI is already talking.
Make sure it knows what it is talking about.