LLM-Generated Scrapers Lose DOM Context When HTML Is Converted to Markdown

When HTML is converted to Markdown for LLM consumption, the structural DOM metadata — CSS selectors and XPaths — is discarded, forcing developers to either re-query the LLM repeatedly for scraping logic or hand-code brittle selectors. This creates a token-cost and accuracy problem for anyone building LLM-assisted web scrapers at scale. Without DOM annotations preserved alongside readable content, LLMs cannot generate stable, reusable extraction code in a single pass.