Structure your content for better answers
Your bot finds answers by searching your content in small pieces. Clean, well-organised content gives it better pieces to work with, and uses fewer storage tokens.
Write content the bot can use
- One topic per page or entry. A page that covers returns, shipping and warranties at once is harder to match than three focused pages.
- Use clear headings and plain questions. Headings like "How long does delivery take?" match the way people ask.
- Say things explicitly. Put prices, limits and conditions in text, not only in images, tables of icons or PDFs of scans. The bot can't read text inside images.
- Fill gaps with manual content. If people keep asking something your site doesn't answer, add the answer as a manual data source rather than rewriting your website.
- Remove what's out of date. Old pages compete with new ones. Delete them from the data source's Content tab.
Split large sources
Put content in separate data sources when parts of it need different settings. For example:
- By section: docs, blog and product pages as separate website sources.
- By how often it changes: frequently updated pages on a daily sync, static pages on a weekly or manual sync.
- By page layout: sections with a different page structure can each have their own CSS selector.
Splitting lets you re-sync only what changed and makes problems easier to find. Each data source counts towards your plan's data source limit, so balance this against your plan.
Pick a CSS selector for your website
A CSS selector tells Chat Thing which part of each page to read, so menus, footers and cookie banners don't end up in your bot's knowledge. The default reads the whole page body and excludes the header and footer.
| Platform | Selector to try first |
|---|---|
| WordPress | article or .entry-content |
| Webflow | main |
| Squarespace | main |
| Shopify | main |
| Most other sites | main or article |
Themes vary, so check your own pages: right-click the main content in your browser, choose Inspect, and find the element that wraps the content you want. Test the selector on one page before syncing a whole site. See website: choose which part of each page is read.
Choose how to add web pages
| Method | Best for |
|---|---|
| Sitemap | Sites with a sitemap: the most complete and predictable option. |
| Crawl | Sites without a sitemap: follows links from a starting page. |
| Individual or bulk URLs | A known list of specific pages. |
A crawl adds up to 600 pages. Re-syncing doesn't find new pages, so when you add pages to your site, add them to the data source too (see website).
Keep knowledge fresh without wasting tokens
| Content | Suggested sync |
|---|---|
| News or blog with daily posts | Daily |
| Documentation and help centre | Daily or weekly |
| Product pages where prices or stock change | Daily |
| Marketing pages | Weekly |
| Content that rarely changes | Manual |
See keep your bot up to date to set a schedule.
Source-specific tips
- Notion: add top-level pages only. Their sub-pages are included automatically, so adding them again creates duplicates.
- YouTube: videos need captions. Auto-generated captions work; manual captions are more accurate.
Related
- Add your website
Add pages from your website to your bot by URL, sitemap or crawl, choose which part of each page is read, and access password-protected sites.
- Retrieval settings
Control how much of your knowledge your bot reads for each message - document relevance, context documents, enhanced retrieval, sources and chunk size.
- Keep your bot's knowledge up to date
Re-sync data sources by hand, on a daily, weekly or monthly schedule, or automatically when you publish in your CMS, and see what each sync changes.
Last updated