Product knowledge sources
Written by the CustomerSaid team · Updated Oct 14, 2026 · 2 min read
Product knowledge is context about what your product is and does. With it, the Analyst can explain features, check whether your site already covers a customer request and make better suggestions.
Who can do this: people with the product context permission, normally the Owner.
Add a source
Open Settings › Product context and add one of:
- A landing page. CustomerSaid reads the page and follows links on the same site.
- A knowledge base or documentation site.
- Manual text. Paste up to 200,000 characters.
URL sources crawl up to 50 pages on the exact hostname you give. A subdomain is a separate source. CustomerSaid uses your sitemap and llms.txt file when they exist.
What happens next
Each source shows a status: queued, scanning, ready, failed or disabled. When ready, open it to see the extracted pages, add a page role (landing, docs, changelog, pricing, other) and a product area, and read the passages the AI uses. AI also groups page summaries into editable knowledge themes.
Sources refresh on their own when they are older than seven days. You can rescan any time, or delete a source.
What it cannot be used for
Product context never counts as customer evidence. The Analyst labels it as a product source and will not use it to claim demand, number of customers, revenue or priority.
Safety and politeness
- Only public HTTPS pages are read. Private networks and internal addresses are blocked.
- The crawler identifies itself as
CustomerSaidProductContextBot, respectsrobots.txtand sends no cookies. - Pages are limited to 2 MB. Scripts and forms are stripped, and only plain text is stored.
- Page text is treated as data. Nothing on a page can instruct the AI.
Limits
Creating or rescanning sources is limited to 10 per workspace per day, and scans run one at a time.
Credits
Scanning uses AI credits. See AI credits and usage.
Troubleshooting
My source failed. The error is shown on the row. Common causes are a blocked robots.txt, a non-public page or a redirect to a different host. Fix the cause, then rescan.
Few pages were found. Add a sitemap, or add pages you care about as manual text.
A JavaScript-heavy site returned little. Pages with little static text are rendered in a browser. If a site still blocks it, paste the content as manual text.
Still stuck?
Ask the Analyst inside the app. It answers from your own evidence and from these guides.