What is knowledge
Knowledge is the material the agent reads when it builds an answer. Pages collected from your site, product information, and notes you enter yourself all go here.
The agent does not automatically know your whole site. It knows only what is registered as knowledge. It cannot read pages that are on your site but not registered, and it cannot get policies that are not on your site from anywhere.
When a visitor asks "How much is shipping?", whether the agent can answer depends on whether your shipping policy is in its knowledge. This is why knowledge registration affects answer quality more than installation does.
Three ways to register
| Method | What it reads | Limit | Processing |
|---|---|---|---|
| Full site collection | Follows pages within the site, starting from the address | Up to 1,000 pages, up to 15 minutes | Asynchronous, one at a time |
| Single URL | One specified page | 400KiB per page | Completes immediately |
| Direct text entry | The pasted text as is | 400KB | Accepted immediately, takes time to appear in search |
For content such as shipping fees or exchange rules that is scattered across your site or missing entirely, entering text directly is the reliable option. However, knowledge entered as text does not appear in the list and cannot be edited or deleted after registration. For content that may change, use site collection.
Full site collection and re-collection use that day's collection runs. A run is deducted even if collection fails.
Diagnosis is for reference
The first step of the registration wizard is a site diagnosis. It reads the site and shows an SEO score (0 to 100) and issues found. If the score is 50 or lower, suggestions for improving search and AI visibility are shown as well.
The diagnosis result does not block collection. Even if it says the site could not be read, you can try collecting with "Register anyway", and you can also skip the diagnosis and register right away.
When registered knowledge is used in answers
When collection finishes and the source status becomes "Ready", that knowledge is used in answers. To confirm it is actually used, enter a question in the "Search check" tab. Up to 5 related knowledge snippets are shown.
| Status | Description |
|---|---|
| Pending | Registered but not yet collected |
| Collecting | Being read. Waits if an earlier collection is running |
| Ready | Used in answers |
| Failed | Could not be read |
| Update failed · answering from previous collection | Re-collection failed, but answers continue from the previous collection |
Change the knowledge and the answers follow
When your site content changes, "Re-collect" reads it again from the start. When it completes, the existing knowledge is replaced all at once with the new collection, and answers use the updated content from the next conversation.
If collection fails or is canceled, the previous collection stays in place and answers are not interrupted.
The same address cannot be registered twice. Entering an address that already exists shows "This address is already registered. Re-collect it from the source list instead". This is because duplicate pages push other knowledge out of answers.
Product information
When a site is collected, product information is organized separately and builds up in the "Products" tab. This list is the material for product cards in answers. For businesses without products, an empty tab is expected.
On a Cafe24 store, web pages are not collected. Cafe24 product information is received through the API instead. Products you register or edit are reflected as is, and products that are not displayed or not on sale are not brought in.
On a directly installed site, product pages must meet certain conditions for products to be organized.
The Premium plan includes 3,000 pages per month of automatic crawl learning.
Next steps
- For registering and managing knowledge on screen, see Knowledge management.
- For the conditions product pages must meet, see Preparing your site.
- When answers differ from your site content, see Inaccurate answers.