Markdown and tables

Many AI fetchers read a page as extracted text, not as a rendered screen. Anthropic's web fetch tool retrieves a URL's full text content, accepts only text, HTML and PDF, and does not support websites dynamically rendered with JavaScript. Trafilatura, an extraction library its documentation says is used by Hugging Face, IBM, NVIDIA, the Allen Institute for AI and others, drops recurring boilerplate, keeps paragraphs, titles, lists, quotes and code (tables optionally) and outputs text or Markdown. Content that exists only in images, canvas or client-rendered widgets therefore does not reach such pipelines, while HTML headings, lists and tables carry their structure through.