Skip to main content

Pasted text and YouTube

Two source types for content that is not a page and not a file.

Pasted text

Paste content straight into your chatbot's knowledge base. Give it a title so you can identify it later, then the text itself.

This is the most underrated source type, and often the one that fixes a misbehaving chatbot.

When to use it

Content that is not published anywhere. Your returns process, your delivery timescales, your opening hours over a holiday period — the things staff know but the website never spelled out.

Answers to questions your site dances around. If people constantly ask "do you deliver to Scotland?" and your site only says "UK-wide delivery available", write the explicit answer and paste it in.

A correction. When the crawler produced a garbled version of an important page, paste a clean version rather than fighting the crawl.

Content behind a login. Anything the crawler cannot reach, you can paste.

Writing text that answers well

Write it the way you would answer a customer, not the way you would write marketing copy.

Good:

Do you offer refunds? Yes. You can return any unused item within 30 days of delivery for a full refund. Email support@example.com with your order number and we will send a prepaid returns label. Refunds are processed within 5 working days of us receiving the item.

Less good:

Our customer-first philosophy means we go above and beyond to make sure you are delighted with every purchase.

The first contains facts that can be retrieved and quoted. The second contains nothing a visitor can act on.

Question-and-answer format works best

Writing pasted text as explicit Q&A pairs matches the shape of what visitors actually type, which makes the right chunk far more likely to be retrieved.

YouTube

Point SiteMesh at a YouTube video URL and it indexes the video's transcript.

Useful for:

  • Product demos and walkthroughs
  • Recorded webinars and training sessions
  • Tutorial content you never wrote up as text

Requirements

The video must have captions available. SiteMesh reads YouTube's transcript — it does not transcribe audio itself. If a video has no captions, whether auto-generated or uploaded, there is nothing to index and the source will error.

To check: open the video on YouTube and look for the transcript option. If you can see a transcript there, SiteMesh can read it.

What to expect

Transcripts are spoken language. They are conversational, they include filler, and auto-generated ones misspell product names and technical terms. The chatbot can still answer from them, but the quality is below well-written text.

For anything important — pricing stated in a video, a process explained verbally — write the key points out as a pasted text source as well. It takes two minutes and produces noticeably better answers.

Videos are re-indexed on refresh

A YouTube source can be refreshed like any other, which re-reads the current transcript. Useful if you replaced the captions with a corrected version.

Keeping content fresh