Contracts are highly structured documents: numbered clauses, defined terms, exhibits, signature blocks. PDF buries that structure behind a fixed layout. Markdown puts it back — making AI-assisted contract review actually work.
Legal contracts are written for a particular kind of reading: jump to clause 7.3, cross-reference defined terms in Exhibit A, compare two redlines side-by-side. PDF preserves the visual layout but loses the structure underneath — text extraction yields a flat stream where clause numbers float free of their content and exhibits interleave with the main body. Convert to Markdown and the structure comes back: numbered clauses stay grouped, defined terms preserve their formatting, exhibits become separate sections. For AI review (Claude or ChatGPT asked to summarise risks, flag deviations from a template, or compare two versions), Markdown input is the difference between useful answers and vague paraphrasing.
Drop the contract PDF on t0md, get the Markdown. For solo review: paste into Claude with a prompt like "flag any clause that's unusual versus a standard NDA" — the model can reference clauses by number because the structure survived. For redlining: convert both versions, diff the Markdown in any text-diff tool, and the changes are line-by-line readable instead of page-by-page screenshots. For high-volume review (procurement, M&A diligence), pipe contracts through t0md's API at ingest, then feed Markdown into your review pipeline.
Signature blocks come through as text (names, titles, dates if typed). Exhibits become their own sections — t0md treats them as continuations of the document rather than separate files.
Scanned image-only PDFs aren't supported yet — OCR is on the roadmap. If your contract was generated digitally (e.g. via DocuSign, Adobe Sign), the underlying PDF has text and converts cleanly.
Files are processed in memory and discarded immediately after conversion — t0md never persists them. For higher compliance bars, self-host t0md and route your contracts to your own instance.