A workflow for analysing documents you can check line by line
Ask for structured output with a location for every value — a table of field, value and the clause or page it came from. That turns document analysis into something you can audit in a minute instead of a paragraph you have to trust.
Document work is rarely 'what does this say'. It is 'what are the dates', 'how does this version differ' and 'which clause creates the obligation' — questions with checkable answers.
ChatOTL keeps the document and the questioning in one place: attach the file, ask in a thread, branch when a clause deserves its own investigation, and search your history later for what you found.
Ask for structure, with sources
- Request a table: field, value, and the clause or page number it came from.
- Ask explicitly what is missing or ambiguous rather than letting it guess.
- Ask for the three riskiest items and why each one is risky.
- Spot-check the values you will act on against the document itself.
Comparing two versions
Attach both versions in one message and ask for a change list grouped by materiality — what changes an obligation, what changes a number and what is only wording. Then check the material ones in the text.
How files reach the model
Attach up to 5 files per message at 15 MB each. Plain text, Markdown, CSV, JSON and source code are extracted and passed inline; PDFs and images are transcribed by a dedicated vision pipeline, and that text is provided to DeepSeek. Files are scoped to your account by row-level security.
Keep the analysis reusable
Work one document per thread, branch for deep dives into a single clause, and use keyword search across your message history to find a previous extraction instead of redoing it.
Where it fits
Contract intake
Extract parties, term, notice periods and amounts with clause references.
Version diff
Attach both drafts and get a change list ordered by materiality.
Report triage
Pull the figures that moved and the explanation given for each.
Example prompts
- Extract every date, amount and party as a table with the clause number for each row.
- Compare these two versions. Group changes into material, numerical and cosmetic.
- What does this document leave ambiguous? List the questions I should ask before signing.
Limitations
- Scanned or low-quality images can be misread; verify values against the original page.
- No web browsing, so it cannot look up a referenced standard, filing or counterparty.
- Output is not legal, financial or compliance advice and always needs human review.
Frequently asked questions
Can it analyse scanned PDFs?
Images and scanned PDFs are transcribed by a dedicated vision pipeline before DeepSeek sees the text, so legible scans usually work, but low-quality scans get misread. Always check extracted numbers against the page.
How do I trust an extraction?
Ask for the clause or page number alongside every value. An extraction with locations can be audited in a minute; one without cannot be checked at all.
Can I compare two contracts?
Yes — attach both in the same message and ask for the differences grouped by materiality rather than a flat diff.
Where are uploaded documents stored?
In private storage scoped to your account, with row-level security applied to every file, conversation and message you create.
Put one contract through it
Ask for a table with clause references and audit the rows.
Open ChatOTL