When you connect a website to Serochat, it indexes up to 25 pages that you choose. It does not crawl your site. Most competing tools advertise the opposite as a headline feature, so the limit deserves an explanation rather than a footnote.
The short version: a crawler cannot tell the difference between your returns policy and a promotion that ended in March. Your bot then answers from both.
What a crawl actually swallows
Point an automated crawler at a five-year-old website and it will faithfully collect the current help pages, and also the 2024 Christmas delivery notice, two earlier versions of the same FAQ that were never deleted, a careers page, the terms you replaced last year but left at the old URL, three near-identical product pages, and a blog post from a member of staff who no longer works there.
Everything on that list is public, so nothing has gone wrong technically. But retrieval does not know which page is authoritative — it knows which page matches the words in the question. The 2024 delivery notice matches “Christmas delivery” beautifully. Your customer gets a cut-off date that expired two years ago, correctly cited, from a page that genuinely exists on your site.
A wider index is not a better one. It is a larger surface for the wrong answer to come from.
Choosing the pages is the setup work
The 25-page limit forces a decision we think you should be making anyway: which pages on your site are the ones you would want quoted back to a customer. In practice that list is short.
- The delivery and shipping page, with current cut-offs.
- Returns, refunds and exchanges.
- Sizing, compatibility or specification guidance — whatever your product needs.
- Payment, account and order-status help.
- Opening hours, locations and contact routes.
- The three or four product or service pages that generate the most questions.
Discovery makes the choosing quick rather than tedious: Serochat can read your site’s sitemap and list up to 500 candidate URLs, and you tick the ones that belong. The reading list is yours; the tool just hands you the shelf.
If a page does not survive that selection, that is information too. A page you would not want a bot to quote is usually a page a customer should not be finding either.
What we give up
This is a real limit, not a cleverly disguised feature, and it costs something.
A large help centre will not fit in one website source. If yours runs to hundreds of articles, the website connector is the wrong route and a direct connection to Zendesk, Confluence or Notion is the right one — those read your help content as articles rather than as pages, at far higher volumes. The Starter plan includes one connected source, so a business with content in several places will want the Team plan for the extra sources rather than trying to squeeze everything through a URL list.
We also lose the convenience of pointing at a domain and walking away. Choosing 25 pages takes perhaps twenty minutes the first time. We think that twenty minutes is the most valuable part of the setup, because it is the moment someone who knows the business decides what the business officially says.
The bot can only be as trustworthy as its reading list. Ours is short on purpose, and you are the one who writes it.