What an AI knowledge base does in AiRep24
The knowledge base is the approved business context the assistant uses when it answers a visitor. It is not a public model's general knowledge and it is not a guarantee that every question can be resolved automatically.
Its job is to connect a natural-language question with relevant website or store information, then return a grounded answer or a real next-step link.
Sources for websites and Shopify stores
Website installations can start with readable public pages and available sitemap links, including services, pricing, policies, FAQs, and help content. Content behind a login, on a private network, or missing from the public site needs another approved input.
Shopify installations can use customer-facing product, variant, inventory, collection, policy, page, market, and other connected store context available to the installation.
Review, correct, and rescan knowledge
Owners can inspect imported knowledge, correct gaps, add approved notes, and remove outdated material. When the source website or store changes, rescan and retest the affected questions.
Keep ownership with the business team that understands the policy or product fact. The assistant should not decide which conflicting source is authoritative.
Set up with representative customer questions
Start with one public URL or a connected store, review the resulting source set, and test real questions from sales and support. Include short queries, follow-up questions, misspellings, policy exceptions, and cases that should go to a person.
Launch only after the answers, links, and operator route behave as expected on the target website.
What happens when knowledge is incomplete
If the sources do not contain a reliable answer, the safer path is to ask for clarification, point to an approved contact route, or notify a configured operator with the available conversation context.
Review unresolved questions as content gaps. Add or correct the business source first, then rescan and test again.
Measure answer quality against your baseline
Track repeated-question volume, unresolved intents, time to a usable answer, links selected, and cases sent to an operator. Compare those measures with the workflow before launch.
Results depend on source quality, question mix, traffic, configuration, and the team's follow-up process; avoid treating a generic percentage as a product guarantee.
