Native .docx vs converted Markdown

Native .docx ingestion in Gemini is fine for ad-hoc "summarise this Word file" prompts. It is the wrong tool for the job when you want to: (a) keep a re-usable text version of the document, (b) hand-correct any conversion errors before feeding it to multiple prompts, (c) cross-reference the document with text from other sources in the same conversation, or (d) chunk it for semantic retrieval. For all of those, Markdown is the right primitive.

The 1M-token window is wasted on raw .docx — XML overhead consumes context that could otherwise hold more documents. Converted Markdown packs 30-50% more usable content per token.

AI Studio and Vertex workflow

Convert each Word document on Word to Markdown, attach the .md files to an AI Studio session. Combine with PDFs (PDF for Gemini), web pages (URL for Gemini), audio (Audio for Gemini), and videos (Video for Gemini) — Gemini cross-references the unified context as one source set.

Pattern: the giant compliance corpus

Compliance teams maintain dozens of Word policies that need cross-referencing on every audit query. Convert all of them once, store the .md files, attach the relevant subset to each Gemini session. The 1M window holds the entire corpus; queries like "which policies conflict on data retention" become tractable single-shot prompts.