DAN AI — Supported Document Types

DAN AI ships 21 tuned document types organized into 6 categories. The type you pick drives which prompt is sent to the AI provider, so matching the right type significantly improves extraction accuracy.

If you pass an unknown type or omit it entirely, DAN AI falls back to general — a universal prompt that returns whatever structured key/value pairs it can recognize.

Finance

Type	What it's tuned for	Has line items?
invoice	Vendor invoices: vendor & buyer, invoice number, dates, currency, line items, subtotal/tax/total.	Yes
receipt	Printed retail / restaurant receipts: merchant, date, item lines, tax, payment method.	Yes
purchase_order	Buyer-issued POs: PO number, supplier, ship-to, item lines, terms.	Yes
expense	Reimbursable expenses: claimant, category, date, amount, currency, project code.	No
bank_statement	Multi-page bank statements: account holder, period, opening/closing balance, transaction rows.	Yes
payslip	Employee payslips: employee, pay period, gross/net pay, deductions, employer details.	Yes
tax_form	Tax filings & related forms: jurisdiction, form number, taxpayer, totals.	Sometimes
quote	Sales quotes / estimates: quoter, quote number, validity, item lines, totals.	Yes

Legal

Type	What it's tuned for	Has line items?
contract	Multi-party contracts: parties, effective date, term, renewal, payment terms, governing law.	No
nda	Non-disclosure agreements: parties, effective date, scope, duration, exceptions.	No
insurance_policy	Insurance policies: policyholder, insurer, policy number, coverage limits, premium.	No
certificate	Certificates of various kinds: issuer, holder, certificate number, issue & expiry dates.	No

Logistics

Type	What it's tuned for	Has line items?
delivery_note	Shipping / delivery notes: sender, recipient, tracking number, item lines.	Yes
shipping_label	Carrier labels: carrier, service, tracking number, weight, dimensions, addresses.	No

People & ID

Type	What it's tuned for	Has line items?
resume	Candidate resumes: contact info, education, experience, skills, certifications.	No (uses arrays)
id_card	Government IDs (passport, driver's licence, national ID): name, DOB, ID number, expiry.	No

Health

Type	What it's tuned for	Has line items?
medical_report	Diagnostic lab reports: patient, test date, panel name, lab rows with reference intervals.	Yes
prescription	Prescriptions: prescriber, patient, drug list with dosage, refills, date.	Yes

General

Type	What it's tuned for	Has line items?
general	Universal prompt — best when nothing else fits. Output schema varies by document content.	Sometimes
letter	Business / personal letters: sender, recipient, date, subject, body summary.	No
memo	Internal memos: from, to, date, subject, body summary.	No

Choosing the right type

A few rules of thumb:

• Always prefer a specific type over general. Even if you have to spread mixed documents across multiple types, you'll get better accuracy than throwing everything into the universal prompt.
• Use general for unknown shapes. It's also a useful diagnostic — when a specific type returns poor results, switching to general for one extraction reveals what the AI actually "sees" in the file.
• Receipts vs invoices — pick receipt for short printed thermal receipts and invoice for any A4-style document with a "Total Due" block, even if the seller calls it a receipt.
• Bank statements — multi-page statements work, but very long statements (50+ pages) may be truncated. If you can split them up by month before uploading, you'll get cleaner output.

Adding a custom prompt for a type

If your documents have a non-standard layout (e.g. invoices from a vendor that buries the total in the footer), you can override the built-in prompt for that document type. See Custom Fields and Templates.

Using the type through each ingestion path

The doc_type value works the same everywhere it appears:

• Dashboard upload — picked from the type dropdown in the upload modal.
• Inbox sync — set per ruleset under Inboxes → Manage Rules. Apply different types to different senders.
• REST API — passed as the doc_type field in the multipart request. See REST API.