All articles

AI & Content

Knowledge Base

Upload documents the store AI uses to answer customer questions.

Overview

Assistant → Knowledge Base (/en/settings/knowledge-base) stores documents the WhatsApp / inbox AI uses to answer customer questions.

The page at a glance

  • Status strip — what the AI knows: documents and pages (ready / in queue / stuck / need re-index / failed), product information (up to date / changed / not indexed), approved Q&A and answers awaiting review, and your last test. "In queue" counts only sources that are really being indexed; a source still pending after 6 hours shows as Stuck with Retry. Products that were never indexed show as Not indexed, never as processing.
  • Tabs — Sources (files, web pages and text; search and pages on the server, no 100-item limit), Q&A (suggested answers to review and approved answers), Products (the AI index status of your catalog), Crawls and Test.
  • + Add knowledge — one entry for file, URL, text and crawl. Import from site opens the Migrate website page.
  • Edit text — text sources can be edited. Save stores the text (status Needs re-index); Save and re-index also re-indexes it.
  • Cost before you click — Test, Re-fetch / Re-index / Retry and Index products show an estimated credit cost next to the button. Opening the page, searching and paging never spend credits; the charge is what the ledger records when the job runs.
  • Deleting — Delete on the Sources tab removes a source only from the knowledge base. Products cannot be deleted on this page: Remove from AI index only takes a product out of the AI search. Delete products on the Products page.

Which information does the AI use first?

  1. Live store data — products, prices, stock, bookings and orders are always read live from your store.
  2. AI instructions and procedures — your rules, tone and return or cancellation steps.
  3. Knowledge base — documents, web pages, text and approved Q&A are searched together and the most relevant passage is used. Approved Q&A does not outrank documents.
  4. Product index — product descriptions the AI can look up; never used for prices or stock.

Prices, stock and bookings always come from live store data, even if a document says otherwise.

Ways to add content

  1. Migrate website — its own page: Store settings → Migrate website (also from Welcome, Products → Import → From a website, the Pages AI import and Import from site here). Confirm the store's main language and currency first. The steps run on one page, each with live progress and result counts: what to import → site address → access (Chrome extension, only when the site blocks our servers) → discovery → then:
    1. Products → the products import runs right on the page (prices into the catalog, not as KB docs). Large catalogs use a trial sample first.
    2. Pages, blog, categories & menus → copied as they are (rule based, no credits); every row shows its result. Product PDPs and docs/articles are not listed here.
    3. Knowledge → imports remaining docs / help / articles into the Knowledge Base only (skips products & storefront pages). Large sets use a trial sample too. KB docs feed AI answers and blog research — they do not create product pages from documentation.
    • Paste your homepage; we resolve sitemap.xml, wp-sitemap.xml, Rank Math indexes, and Sitemap: lines in robots.txt. If nothing is found, paste a direct sitemap URL.
    • Optional AI instructions apply across the flow (same idea as Products/Pages import notes).
    • Multilingual sites: we prefer English, then your store's default language, and skip other-locale duplicates.
  2. File upload — PDF, DOCX, TXT, MD, XML (max 20 MB each; multiple files allowed). Files are stored securely in OrderInChat.
  3. URL — Paste a page or PDF URL. We download the page/file, save a snapshot in OrderInChat, then index the extracted text. The original URL is kept only as a reference; the AI uses the saved snapshot.
  4. Manual text — Paste content directly.
  5. Crawl / sitemap (advanced) — One-shot Import Sitemap for power users. Prefer the Migrate website wizard for a full WordPress move.
  6. AI site import (Store settings → Pages) — Same Pages dock the wizard opens in step 2. Extra marketing/docs content stays in KB for the AI — it is not cloned into dozens of menu links.

Do not put shipping rates or tax rules only in KB — configure Store settings → Shipping and tax settings so checkout stays correct.

Large sites: one pass may not cover everything

A single import or crawl has soft caps so the job finishes reliably. That means:

  • You may see fewer URLs than your whole WordPress/Rank Math sitemap lists.
  • Multilingual duplicate URLs are skipped so the same page is not ingested in every language at once.
  • This is expected. You did not “lose” the rest of the site — leftovers can always be added later.

How to add remaining pages later

  1. Open Assistant → Knowledge Base.
  2. Add missing URLs one by one, or start another crawl / paste a sitemap section.
  3. Wait until status is ready — the WhatsApp / inbox AI will use the new content even if those URLs are not storefront menu pages.

You do not need to know every leftover URL in advance. Use your old sitemap, Google Search Console, or “pages the AI still cannot answer” as a checklist.

Storefront images vs KB crawl: Crawling many URLs does not mean copying one image per page into the shop. AI site import uses a small cap of main photos for Home/Shop design. See Store Pages.

Videos: KB may note video embeds in text; we do not download MP4/WebM into storage during crawl. Upload later via Media Library or keep YouTube/Vimeo links.

File size vs what the AI can use

The 20 MB limit is an upload / storage cap. It does not mean the AI reads the whole file in one reply.

After upload we:

  1. Extract text (or OCR scanned PDFs via vision when needed)
  2. Distill and split into searchable sections
  3. Index up to about 2 million characters of extracted text per file (extra text is truncated)
  4. At chat time, retrieve the most relevant sections for the customer’s question (not the entire document)

Dense spreadsheets, huge XML dumps, and scanned PDFs can hit content limits before the 20 MB file cap. Prefer clear policy / FAQ text over megabyte-scale dumps when possible.

Usage

  1. Add documents relevant to your store policies and FAQs
  2. Wait for indexing status to become ready
  3. Ask a test question in Inbox / WhatsApp
  4. Use Re-sync on a URL-sourced document to re-download, replace the saved snapshot, and re-index

Quality over quantity

  • Write FAQs the way customers ask (“Do you ship to Cyprus?”, “How do I cancel?”)
  • One clear document per topic beats ten overlapping marketing pages
  • After a policy change, re-sync or delete the old version — do not leave two conflicting docs indexed
  • Crawl leftovers are fine for coverage, but fix contradictions before blaming the chat model

Weak or inconsistent customer answers? See Improve store AI answers.

Important

  • Booking availability, prices, and live catalog should come from your live Products and Bookings — not only KB text
  • Keep KB updated when policies change (re-sync URL docs or re-upload files)
  • Avoid contradicting Products, AI Instructions, or Procedures (same policy → one number/source)
  • Usage consumes AI credits (see Billing)
  • After import, AI answers do not depend on the original website staying online — content lives in your saved snapshot
  • Reply length and how much context fits in one turn follow the AI model you selected for that channel (see AI Models / AI Instructions)

Need more help?

Chat with the Help assistant about OrderInChat, or open My Support Tickets while signed in (free trial included).