The Document Intelligence Analyze API uses one-based page numbers for its pages parameter. A printed page label is not necessarily the same as the page's position in the uploaded PDF. Check that distinction before interpreting missing fields as an OCR failure.
This guide refers to the documented 2024-11-30 API. It explains request design and validation without making a paid service call. Model, input-format and account limitations still apply, and actual processing should use approved documents and an authorized Azure resource.
Map the PDF's physical page positions
Open a harmless sample and count pages from the beginning, including covers and unnumbered inserts. A document whose printed “Page 1” appears after two cover sheets has a different physical position. Use that position when defining the API selection.
Microsoft's Analyze Document REST reference describes one-based page numbers and range syntax, such as a continuous interval or a comma-separated selection. The parameter chooses what to analyze; it does not rename the page numbers printed inside the document.
Build a page map before the request
Record physical position, visible label and business role for the sample pages. For example, distinguish cover, header page, continuation and appendix. Keep this map nonconfidential and small; it is a diagnostic aid rather than a copy of the full source document.
Keep multi-page business content together
If a model needs a complete invoice or form, selecting only its first visible header can omit the relevant totals or continuation table. Choose a range around the intended business document, not merely the first page where its title appears.
Confirm model and input support
Review the actual model and file type before assuming the page-selection behavior is universal. The official Read model documentation describes supported inputs and page-unit considerations. Its notes for Office or HTML inputs differ from ordinary scanned-PDF page handling.
Do not confuse request pages with output collections
The request selection tells the service which source pages to process. The response contains analysis data that must be interpreted according to the returned schema. Do not assume the first element of a returned array means physical page one of the original file.
Preserve the source reference
Record the approved source identity and selected range with the analysis attempt. If a revised PDF adds a cover sheet, the old range can point at different content. A filename alone may not distinguish those revisions.
| Value | Meaning to verify |
|---|---|
| Physical PDF position | Page counted from the start of the file |
| Printed label | Text placed on the page by the author |
| Request range | Source positions chosen for analysis |
| Returned page metadata | The response's description of processed content |
Test a deliberately distinguishable sample
Use a short PDF whose pages contain different harmless headings. Select one known physical page, inspect the completed result and confirm its content. Then test a multi-page range and a supported noncontiguous selection if the workflow requires one.
Check the completed operation
An accepted asynchronous request is not yet a successful analysis. Follow the documented operation status and inspect the completed output before judging the selected content. Keep authentication details and any signed source URLs out of logs.
Inspect excluded context consciously
Page selection can reduce unnecessary work, but it may also exclude information required to interpret the result. Confirm that the chosen range supplies the fields and context the business process needs. Do not promise an accuracy or cost improvement without measuring the actual workflow.
Handle document revisions safely
When the source layout changes, repeat the page map and selection test. A static range appropriate to one supplier's pack can become wrong when a cover, appendix or continuation page is introduced. Route uncertain layouts to review rather than processing an arbitrary default range.
For organizing resulting records, our SharePoint metadata guide can help define source and review fields. Include the original document reference and analyzed range where the approved traceability process requires them.
Selected source pages and requested output artifacts are different API choices. Compare Document Intelligence Searchable PDF: Read Output.
Keep a compact acceptance note with API version, model, file type and selected physical pages. The goal is an explainable mapping between the intended source content and the completed result, rather than assuming the page number printed in a footer is the API's index.
