PDF Unitization (Custom Split)

Use this template for repeatable, defensible AI review in Claira.

PromptingTemplate PickerPDF Unitization (Custom Split)

When you select this template, Claira displays the prompt configuration panel. Map the output to a review field in your case — typically a Memo field, since the output is a multi-line report ending in the unitization string.

UnitizationMemoclaira-unitization-02

What this prompt is for

This template splits a compiled PDF into separate documents according to a split instruction you write — by individual, by record type, by date of service, by chapter, email versus attachment. It returns the page-range string the Nuix Discover image viewer uses to split the file. Use it when the split turns on something the pages alone will not tell you; when the file is obviously a stack of distinct records with self-evident boundaries, use PDF Unitization instead.

Step-by-step in Claira

Open Prompting > Template picker in your case.
Select PDF Unitization (Custom Split) and replace the placeholder block with one or two plain sentences saying what each output document should be.
Map the output to a review field (a Memo field works best) and set the Scan as dropdown to Image so Claira reads the actual PDF pages.
Run the compiled PDF, review the proposed boundaries and any UNCERTAIN entries, then paste the unitization string into the image viewer's split dialog.

The prompt

TASK: PDF UNITIZATION

This PDF needs to be split into separate documents. Identify where each output
document starts and ends and output the page ranges needed to split the file apart.
You are drawing boundaries, not summarizing. Use physical PDF page numbers starting
at 1.

HOW TO SPLIT THIS FILE:

>>> REPLACE THIS BLOCK WITH YOUR SPLIT INSTRUCTION <<<

That instruction controls. The signals below only help you locate the boundaries it
calls for - they are not an independent reason to split.

SIGNALS THAT MARK A BOUNDARY:
- New letterhead, cover page, or title page
- A new email header block (From/To/Date/Subject) not quoted inside a thread
- A new date paired with a new author or addressee
- Pagination restarting, e.g. "Page 1 of 6" after "Page 4 of 4"
- A change in document type or format
- A slip sheet, exhibit divider, or fax cover sheet
- A change in the matter, entity, or individual the pages concern

DO NOT SPLIT ON, unless the instruction calls for it:
- A page break inside a continuing document
- A section, article, or chapter heading within the same document
- A signature, notary, schedule, or exhibit page belonging to the document before it
- A continuation page of a table or list
- Repeated letterhead on later pages of the same letter
- An earlier message quoted inside an email thread

RULES:
- Every page appears in exactly one range. No gaps, no overlaps, no page twice.
- Ranges ascend and cover page 1 through the last page.
- Assign a blank or separator page to the document it introduces, otherwise to the
  document before it.
- If no signal on a page starts a new document, it continues the previous one. Never
  split on a weak hint - keep the pages together and flag it.
- If a page fits no category the instruction describes, keep it with the document it
  follows and list it under UNCERTAIN.
- If the instruction cannot be applied to this file, say so rather than guess.

OUTPUT:

TOTAL PAGES: [n]

DOCUMENTS:
1. [start-end] | [label under the instruction] | [type] | [date, or "no date"]
2. [start-end] | ...

UNITIZATION STRING: [e.g. 1-4; 5-9; 10-22]
UNCERTAIN: [page and reason, or "none"]

Hyphen for a range, comma to join non-contiguous pages in the same document,
semicolon between documents. Account for every page exactly once.

Recommended customizations

Replace the placeholder with one or two plain sentences, for example: "One document per invoice." / "One document per named individual, all of that person's records together." / "Each email is its own document, each attachment its own document." / "Split the medical records out; leave the rest as one document."
If the file holds several people and several record types per person, say which dimension wins — "one document per person per record type" versus "one document per person, all types together."
Keep the output contract intact: the UNITIZATION STRING line is the deliverable, and downstream QC depends on the DOCUMENTS table and UNCERTAIN list.

Worked example

Input excerpt

Split instruction: "One document per named individual, all of that person's records
together." A 40-page PDF of medical records covering three patients, interleaved.

Expected output shape

A structured report listing total pages, one line per output document with its page
ranges, the label the instruction assigns (e.g. the person's name), type, and date,
then the unitization string (e.g. 1-9, 22-25; 10-21; 26-40) and any uncertain pages.

Troubleshooting

If boundaries look wrong, confirm the scan ran in Image mode — text mode loses pagination and the model cannot see page breaks.
If the string has gaps or repeats a page, re-run and remind the model every page must appear exactly once; Nuix Discover errors if the same page appears twice in one pass.
If the output says the instruction cannot be applied, the file may not contain what the instruction describes — rewrite the instruction around what is actually in the file, or fall back to PDF Unitization.

PDF Unitization (Custom Split)

On this page