The AI Generic Extraction action reads a document with AI and pulls out the fields defined on the Record's Record Type — header fields and line item fields — for any document type. It is the general-purpose sibling of AI Invoice Extraction: instead of a fixed invoice schema, it builds its extraction list from the Record Type you have already set up, so in the common case there is nothing to configure.


Use it whenever you want AI extraction on documents that are not invoices — applications, forms, statements, medical records, contracts, or any structured document.


AZURE ONLY: This is an AI action and runs only on Azure-hosted / AI-licensed sites.


How it differs from AI Invoice Extraction

Both actions read a document and store the extracted values on the Record, but they differ in where the field list comes from:

  • AI Invoice Extraction uses a fixed, built-in invoice schema (Vendor, Invoice Number, Invoice Date, line items, and so on). The fields are always the same whatever Record Type is in play.
  • AI Generic Extraction takes its field list from the current Record Type. Whatever header and line item fields you defined there are what it looks for, and you adjust that list with Extra Instructions and Exclude Fields.



Reach for Generic Extraction for any non-invoice document, and for invoices keep using AI Invoice Extraction so you get its tuned invoice schema for free.


The two-action pipeline

Extraction and indexing are two separate steps, joined by the data the first one writes:


  1. AI Generic Extraction reads the document and stores each value it finds on the Record as temporary data — header fields as Index_<name> and line item fields as IndexLine_<name>, where <name> is the Record Type field name (or the name you gave the field in Extra Instructions).
  2. The Generic Indexing action then maps those Index_* / IndexLine_* values onto the real Record Type fields. Turn on its Auto-Map and the matching names are applied for you.


Add AI Generic Extraction first, then Generic Indexing later in the same process. Generic Indexing makes no AI call, so that second step is free.


If the Record has no Record Type, there is no field list to build from — in that case only the fields you name in Extra Instructions are extracted, which is a perfectly valid way to use the action.


Options

The action's settings are grouped into three tabs.

Document(s)

  • Active Document — When on, the action extracts from the Record's active document if one is available.
  • Found Documents — Extract only the first matching document, or all documents that match. When Active Document is on, this applies only when there is no active document.
  • Cat/Sub/Name — Restrict which documents are extracted by Category, SubCategory and Name. Leave blank to match any.
  • Pages to Send — Limit which pages go to the AI. Blank sends all pages; 1 sends page one only, and ranges like 1-3 or lists like 1,2,5 also work. This matters more here than on most actions: extraction treats one document as one set of fields, so it does not split a large document across several AI calls the way the split and classification actions do. A long or high-resolution scan can exceed what a single call can carry and is refused outright. Naming just the pages that carry the data is how a thick document gets through, and it cuts the token cost on ordinary documents too.

AI Level

  • Quality — Maps to an AI model:
    • Economy — low token cost; best for fewer, clean digital PDFs. Economy does not support caching, so it suits smaller volumes.
    • Balanced — medium token cost; good for most document types.
    • Maximum Quality — high token cost; best for poor-quality scans, handwriting, or unusual layouts.

Test with your own documents and check the AI Workflow report to see which quality works best for them.

Advanced

  • Include Other Data — Off by default. Turn it on to capture every data element the AI finds, not just the Record Type fields and the ones you named — the "extract everything" mode. This can increase token counts sharply, so use it only when you need it and only on documents of a few pages.
  • Extra Instructions — Optional. Use it to add fields beyond the Record Type, or to give hints. Name each added field and say whether it is per-document (header) or per-line, for example "Also capture the policy number as PolicyNumber" or "Capture each line's part number as PartNumber (per line)". Header values are saved as Index_<name> and line values as IndexLine_<name>, the same as the Record Type's own fields, so Generic Indexing maps them the same way.
  • Exclude Fields — Optional. Record Type field names, one per line, that the AI should not be asked for at all. See below.
  • Date Format — How ambiguous dates (for example 05/06/2025) are interpreted. System Default inherits the global Default Date Format from System Options; MDY or DMY forces the interpretation for documents processed by this action.



Exclude Fields

This action reads every field on the current Record Type without being asked to, which is what makes it work with no configuration. Exclude Fields is how you take a field back out. Enter Record Type field names, one per line, and those fields are left out of the extraction entirely — the AI is never asked for them.


Because the field never reaches the AI, excluding it costs nothing and cannot come back wrong. That makes this the right place to remove a field you do not want read at all: a value a person fills in later, a field another workflow action already sets, a field the document simply never contains, or a sensitive field you would rather not hand to the AI. Fewer fields also means fewer tokens on every document the action processes.


Names must match the Record Type field name exactly, although capitalization does not matter. Wildcards and partial names are not supported — Invoice will not match InvoiceDate. Both header and line item fields are matched, so a name is excluded wherever it appears on the Record Type.


Exclude Fields only filters the fields that come from the Record Type. A field you name in Extra Instructions is still requested, so if you exclude a Record Type field and then describe it in Extra Instructions, it is extracted using your description instead of the Record Type's.


This is not the same as the Exclude From Auto-Map boxes on the Generic Indexing action. Those let a field be extracted and then keep the value from landing on the Record. Exclude Fields stops it being read in the first place. Use Exclude Fields when you do not want the value at all, and Exclude From Auto-Map when you want the value captured but placed on the Record some other way.


TIPS

  1. In the common case there is nothing to configure — set the Record Type up properly, add this action, then add Generic Indexing with Auto-Map on. The field names carry through from end to end.
  2. Field Header text on the Record Type is sent to the AI as the field's description, so a clear Header there improves extraction without any Extra Instructions at all.
  3. Name added fields clearly and consistently in Extra Instructions. Those exact names are what you type as the AI Field Name if you map them by hand in Generic Indexing.
  4. Say whether each added field is header or per-line. Per-line fields are stored as IndexLine_<name> and are mapped on the Line Item Fields tab of Generic Indexing.
  5. Trim the field list with Exclude Fields before reaching for a cheaper quality setting. Fewer fields is a cheaper call with no loss of accuracy on the fields you kept.
  6. If a large document fails, use Pages to Send rather than splitting the file by hand — extraction does not window across multiple calls.
  7. Start with Balanced quality; move to Maximum Quality only if fields are being missed on poor scans, and try Economy for clean, low-volume digital PDFs.



NOTE ON (AI): Reading the document with AI consumes tokens and may incur extra fees. Workflow tokens are billed separately from End-User or Admin tokens — check with your sales rep for more information.