How the AI answers
The AI answers from your Knowledge Base and nothing else. It does not browse the internet, it does not borrow from other businesses on Jiviq, and it does not improvise prices, policies, or promises. When your Knowledge Base does not contain the answer, it says so instead of guessing.
This page explains how that works, and where the questions it could not answer end up.
How does Jiviq work?
A chat box sits on your website. When a visitor asks something, the AI assistant:
- Searches your Knowledge Base — the website pages and documents you imported (Import your website, Upload documents).
- Writes a reply using only what it found. No outside sources.
- Checks the reply before sending. Every claim in the draft is compared against the material it found. A claim that is not supported does not go out.
- Sends the reply — or admits it does not know. If the check fails, the visitor gets an honest “I don’t have that” instead of a confident wrong answer.
Your team sees every conversation in the admin app at app.jiviq.com and can step in at any time — see Talking to a human. Whether the AI sends replies itself or only drafts them for your team depends on your AI mode — see AI modes.
Visitors see this happening honestly. While the AI works, the chat shows the real stage it is at: “Searching the docs…”, “Writing a reply…”, and “Checking every claim against the docs…”.
The answer check
Nothing the AI writes is sent unchecked. After drafting a reply, a second verification step compares the draft, claim by claim, against the Knowledge Base sections it retrieved. If any claim is unsupported — a number that isn’t in your material, a policy it inferred, a step it made up — the draft is discarded.
This is the reason a reply can take a few extra seconds. It is also the reason the AI will sometimes tell a visitor it does not know something that feels adjacent to what you imported: close is not the same as supported.
When it does not know, it says so
A visitor who asks something your Knowledge Base does not cover gets a plain admission, for example:
“I couldn’t find that in the knowledge base, so I won’t guess. Ask me anything else, or I can pass this to the team.”
No invented answer, no vague deflection. And when the assistant can’t answer, your team is alerted (“Assistant had no answer”) so a person can pick the conversation up. Any visitor can also ask for a person at any time — the AI’s opening line tells them so: “Want a person instead? Just ask.”
What topics can you help me on?
Visitors ask the AI this, and the honest answer is: the topics in your Knowledge Base. The AI’s range is exactly what you have imported — services, prices, hours, policies, how-tos. It will help with anything covered there and say so plainly when something is not.
That means you control the range. If visitors keep asking about something the AI declines, the fix is to add that answer to your Knowledge Base — which is exactly what the next section is for.
Unanswered questions are kept, not lost
Two things happen when the AI cannot answer:
- The conversation reaches your team. When the AI hands off, the conversation is flagged in your inbox — use the Escalated filter on the Chats page to see them — and your team is notified by email and push.
- The missed question is collected for you. Jiviq gathers questions the assistant could not confidently answer into a queue called Assistant gaps in the admin app sidebar. Each entry shows the visitor’s question; press Answer, write the answer, and Add to knowledge base — the assistant uses it from the next conversation onward. The same question asked by many visitors appears once, not once per visitor. You can also Dismiss entries you don’t want to answer.
If you don’t see Assistant gaps in your sidebar, contact support and we’ll switch it on for your account.
This is the loop that makes the assistant better every week: it tells you exactly what it is missing, and you teach it in one step.
Visitors always know it is an AI
The AI never pretends to be a person. At first contact, every visitor reads a disclosure line — “I’m {assistant name}, {your business}‘s AI assistant. Want a person instead? Just ask.” — and AI replies carry a visible AI tag next to the assistant’s name (once per run of consecutive AI messages), even if you gave the assistant a human-sounding name.
What the AI never does
- It never answers from outside your Knowledge Base.
- It never sends a reply that failed the answer check.
- It never uses another business’s content. Your material answers your visitors only.
- It never poses as a human. AI replies are always labeled as AI.
If something here doesn’t match what you’re seeing, tell us at support@jiviq.com — this page is meant to be checkable.