Fix
Web sources import clean, readable text
Sources you added from a web search sometimes came in as raw page code instead of the article. Now they import the actual readable text, so chat and citations quote real content.
When you searched the web and imported a result as a source, some pages came in as raw page code — stylesheet rules and stray menu text — instead of the article you wanted. In chat, that meant the assistant would "read" a page of styling code, and citations would show it back to you as if it were the source.
That is fixed. Imported web sources now keep the readable text and drop the page's behind-the-scenes code (styles, scripts, and other non-content). For results that came from a web search, we now import the exact passage the search was based on, so the source is short, on-topic, and quotable instead of a full page of navigation and boilerplate.
This applies to new imports. If you have an older source that came in as code, delete it and add it again to pick up the clean version.