Retrieval Solution

Running the solution produces a summary such as:

The website discusses the changing landscape of software development ...

Trying test case number two gives:

The website introduces AI Hero, and it emphasizes my experience as a member of the XState core team...

Looking at the prompt, the background data is wrapped in XML tags. There is a <url> tag and a <content> tag. The <content> tag is especially important because the raw content may include markdown formatting. Having a clear start and end tag lets the LLM know exactly where the content begins and ends. Without this, the raw content could throw it off.

The rules included here are:

The prompt also passes in the conversation history and adds an output format section: return only the summary.

This approach shows how to handle adding retrieved data into your prompts. With multiple websites, you could even consider wrapping them in a <website> tag or moving the URL inside the website tag <website url="${url}">.

XML makes the structure expressive and explicit, clearly marking the boundaries.