Blog · October 15, 2026

Why AI assistants want your documents in Markdown

If you have pasted a contract into an AI assistant and watched it misread the table, or attached a PDF and got an answer that missed the second column, you have met the problem this article is about.

Why Markdown works better

Large language models are trained on enormous amounts of text, and a great deal of that text, documentation, README files, forum posts, notes, is Markdown. The models know its conventions deeply: # is a heading, - starts a list item, pipes make a table, ** is emphasis. When you give a model Markdown, the structure of your document arrives as structure the model understands. Headings scope the sections, lists stay lists, and a table’s columns stay columns.

A PDF, by contrast, arrives as whatever text the tool on the other end managed to extract, often one line per visual line, with tables flattened into word soup and columns interleaved. A Word file is better but still has to be converted by the assistant’s own pipeline, which you cannot see or correct. Markdown is the form in which you control what the model reads.

Markdown is also compact. A page of Markdown is a few hundred tokens; the same page as extracted PDF text with layout artefacts can be twice that, and the model’s context window is finite.

What to convert

• Reference material you ask about repeatedly: contracts, policies, manuals, specifications. Convert once, keep the .md beside the original, paste or attach the Markdown.
• Spreadsheets you want summarised. A Markdown table with a header row is something a model can reason over; a screenshot of Excel is not.
• Scanned letters and forms. OCR them to Markdown first; then the assistant gets text, not a picture it has to squint at.
• Folders. If you are building a knowledge base for an agent to search, convert the whole folder; plain text is what every indexing tool wants.

The privacy point

The obvious way to get a document into an assistant is to upload the file. For anything sensitive that means the file sits on a server you do not control, possibly for a long time. Converting on your own machine first, then pasting only the parts you need, keeps the original where it is and lets you redact before anything leaves. This is the reason Anymark converts locally and never has a cloud component: the conversion step should not be the leak.

A simple workflow

1. Convert the document on your computer (Anymark: double-click it, save, you have a .md).
2. Read the Markdown once. Fix the odd heading, delete what the assistant does not need.
3. Paste or attach the .md. Tell the assistant what it is (“this is our supplier agreement; the table in section 4 is the price list”).
4. Keep the .md with the original. Next time you will not need to convert.

The result is better answers, lower cost per question, and documents that stayed on your disk.