Azure Document Intelligence Python SDK Returns Data Only for First Page

Question

Azure Document Intelligence Python SDK Returns Data Only for First Page

Kiran Dhumma 25

Hi - I am encountering an issue while using the azure.ai.documentintelligence Python library to extract data from a PDF using page ranges.

My application processes the document page by page, and this workflow was working correctly until this morning. However, starting today, when I attempt to extract data for specific pages, the service returns result only for the first page, regardless of the page range specified.

There have been no code changes on my side. Has anyone else experienced this issue, or is there a recent service update or known limitation affecting page-range extraction?

Gowtham CP 7,725 Reputation points Volunteer Moderator

2025-12-17T07:27:32.3466667+00:00

Hi Kiran Dhumma

Thanks for the question.

1. SDK usage can cause only the first page to be returned If only the first page is returned, this is often related to how the Python SDK call is made. Unlike older Form Recognizer SDKs, in azure.ai.documentintelligence you must ensure the pages parameter is passed correctly in the analysis request and that your SDK version supports page ranges. The REST API supports this, but the parameter must be set in the correct part of the request.

2. Free tier limitation If your Document Intelligence resource is on the F0 (free) tier, the service analyzes only the first pages of a PDF regardless of the page range specified. Upgrading to S0 (standard) removes this limitation and enables full multi-page extraction. Reference: https://learn.microsoft.com/answers/questions/5569121/azure-document-intelligence-only-analyzing-2-docum

3. Document type matters Page-range behavior is best supported for PDF/TIFF files. For formats like DOCX or XLSX, the SDK may not treat content as page-based, which can appear as if only one page is processed. Reference: https://stackoverflow.com/questions/79475225/azure-documen-intelligence-python-sdk-doesnt-separate-pages

Could you please share the Python SDK code snippet you are using (including the client initialization and analyze call)? That will help confirm whether the request is being sent correctly.

I

Kiran Dhumma 25

Hi - @Gowtham CP here are the details FYR:

SDK & Version: azure-ai-documentintelligence>=1.0.2
Resource tier:
Code snippet:

import json
from typing import Dict, List, Any, Optional
from collections import defaultdict
import logging
from doc_processing.helpers import get_openai_client
from doc_processing.azure_key_vault import get_secret
from azure.ai.documentintelligence import DocumentIntelligenceClient
from azure.ai.documentintelligence.models import AnalyzeDocumentRequest
from azure.core.credentials import AzureKeyCredential
from doc_processing.constants import AZURE_OPENAI_DEPLOYMENT,INTERNAL_ERROR

# Configure logging
logger = logging.getLogger(__name__)



def get_document_intelligence_client():
    """Initialize Document Intelligence client with credential validation"""
    endpoint = get_secret("AZURE-DOC-INTELLIGENCE-ENDPOINT")
    key = get_secret("AZURE-DOC-INTELLIGENCE-KEY")
    
    if not endpoint or not key:
        raise ValueError("Missing Azure Document Intelligence credentials in environment variables")
    
    return DocumentIntelligenceClient(endpoint=endpoint, credential=AzureKeyCredential(key))

def analyze_form_with_page_range(document_url: str, retry_attempts: int = 3, retry_delay_seconds: int = 1, pages: str = None) -> dict:
    """
    Analyze document using Azure Document Intelligence
    
    Args:
        document_url: SAS URL of the document
        pkg_id: Package ID for logging
        file_id: File ID for logging  
        doc_id: Document ID for logging
        pages: Comma-separated page numbers (optional)
        
    Returns:
        dict: Extracted content, key-value pairs, and tables
    """
    if not document_url:
        raise ValueError("Document URL is required")

    print(f"[DOC_INTEL] Analyzing document at URL for pages: {pages if pages else 'all'}")
    logger.info(f"Analyzing document at {document_url} for pages: {pages}")

    try:
        client = get_document_intelligence_client()
        # delay_strategy = DelayStrategy.CreateExponentialDelayStrategy(initialDelay = default, maxDelay = default)
        # retry_policy = RetryPolicy(maxRetries = 3, )
        # client_options = DocumentIntelligenceClientOptions()

        analyze_request = AnalyzeDocumentRequest(url_source=document_url)

        for attempt in (1, retry_attempts):
            poller = client.begin_analyze_document(
                "prebuilt-layout",
                analyze_request,
                features=["keyValuePairs"],
                pages=pages
            )

            print(f"[DOC_INTEL] Analysis started, waiting for completion...")
            result = poller.result()
            # for i in result:
            #     print(i)
            content = result.content

            if content:
                key_value_pairs = []
                extracted_tables = []
                paragraphs = []

                if hasattr(result, "key_value_pairs") and result.key_value_pairs:
                    for kv in result["keyValuePairs"]:
                        key = kv["key"]["content"] if "key" in kv and kv["key"] else ""
                        value = kv["value"]["content"] if "value" in kv and kv["value"] else ""
                        confidence = kv.get("confidence", None)
                    
                        key_value_pairs.append({
                            "key": key,
                            "value": value,
                            "confidence": confidence
                        })

                # Extract tables (content only, no coordinates)

                if "tables" in result and result["tables"]:
                    for table_obj in result["tables"]:
                        row_count = table_obj.get("rowCount", 0)
                        column_count = table_obj.get("columnCount", 0)
                    
                        table_cells = [["" for _ in range(column_count)] for _ in range(row_count)]
                    
                        if "cells" in table_obj:
                            for cell in table_obj["cells"]:
                                row_index = cell.get("rowIndex", 0)
                                column_index = cell.get("columnIndex", 0)
                                table_cells[row_index][column_index] = cell.get("content", "")
                    
                        extracted_tables.append({
                            "rowCount": row_count,
                            "columnCount": column_count,
                            "rows": table_cells
                        })

                if "paragraphs" in result and result["paragraphs"]:
                    for para in result["paragraphs"]:
                        para_content = para.get("content", "")
                   
                        # Extract bounding regions (polygon and page number)
                        bounding_regions = []
                        if "boundingRegions" in para and para["boundingRegions"]:
                            for region in para["boundingRegions"]:
                                bounding_regions.append({
                                    "pageNumber": region.get("pageNumber"),
                                    "polygon": list(region.get("polygon", []))
                                })
                    
                        paragraphs.append({
                            "content": para_content,
                            "boundingRegions": bounding_regions
                        })

                print(f"[DOC_INTEL] Analysis completed - extracted {len(key_value_pairs)} key-value pairs and {len(extracted_tables)} tables")
                logger.info(f"Document analysis complete for {document_url} pages: {pages}")

                return {
                "content": content,
                "key_value_pairs": key_value_pairs,
                "tables": extracted_tables,
                "paragraphs": paragraphs
                }
            
            else:
                if attempt == retry_attempts:
                    print("[DOC_INTEL] Extraction failed Max retries reached.")
                    return False

                else:
                    print(f"[DOC_INTEL] Extraction failed for attempt:{attempt}")
                    print(f"[DOC_INTEL] Retrying in {retry_delay_seconds} seconds...")
                    # time.sleep(retry_delay_seconds)
    except Exception as e:
        logger.error(f"Analyze document with Azure doc intelligence failed: {e}")


## Execution

openai_client = get_openai_client()

page_range = 4
# counter = 1
json_state = {}

for i in range(1, page_range + 1):
    file_name = "extraction_op_v3." + str(i)
    content = analyze_form_with_page_range("<blob_url>", pages=str(i))
    
    with open(f"{file_name}.json", "w") as f:
        json.dump(content, f, indent=4)
    
    page_status = get_page_status(i, page_range)
    json_state = generate_structured_output(content, json_state, page_status, i)
    # counter += 1

with open("Std_output_v2.1.json", "w") as f:
    json.dump(json_state, f, indent=4)

txn_idx = extract_transaction_page_table_pairs(json_state)
# print(txn_idx)

Kiran Dhumma 25 Reputation points

2025-12-17T10:14:39.1+00:00

Any updates on the issue @Gowtham CP ?
Fremin Abreu Ortega 20 Reputation points

2025-12-17T16:10:14.28+00:00

I'm facing the same issue using C# sdk. It began happening without any code changes, yesterday the same code works fine...
Anshika Varshney 6,195 Reputation points Microsoft External Staff Moderator

2025-12-26T11:41:37.5266667+00:00
Hi Kiran Dhumma,

Thank you for sharing your feedback. I truly apologies if the initial support did not fully meet your expectations.

I’ve reviewed the existing comments and the SDK behavior that returns results only for the first page when a page range is requested. Below are targeted checks and fixes that have helped resolve this exact issue for others using azure.ai.documentintelligence:

1) Confirm SDK version and API

The page selection behavior changed compared to older Form Recognizer SDKs. Multi‑page extraction works reliably on recent builds.

Recommended: upgrade to the latest stable (e.g., azure-ai-documentintelligence >= 1.0.x).

Pin a version temporarily to verify a regression:
pip install "azure-ai-documentintelligence==<known-good-version>"

2) Pass pages correctly on begin_analyze_document

Ensure the pages argument is sent in the initial analyze call, using either a list of strings or a single range string. Passing pages elsewhere (or omitting it) often defaults the service to page 1 only.

from azure.ai.documentintelligence import DocumentIntelligenceClient from azure.core.credentials import AzureKeyCredential client = DocumentIntelligenceClient(endpoint=ENDPOINT, credential=AzureKeyCredential(KEY)) poller = client.begin_analyze_document( model_id="prebuilt-layout", # or your custom model ID analyze_request={ "urlSource": PDF_URL # or use 'content' for bytes }, features=["ocr"], # optional; include if your scenario needs OCR text pages=["2-4", "6"] # range + explicit page(s) ) result = poller.result()

Notes

Ranges like "2-4" and single pages like "6" are supported.

If you stream bytes: pass content=<bytes> and still provide pages in the same begin_analyze_document call.

3) Verify the result actually contains your requested pages

Sometimes a client-side mapping reads only the first element.

page_numbers = [p.page_number for p in result.pages] print("Returned pages:", page_numbers) # Expect [2, 3, 4, 6] for the example above ``

If you only see [1], the request didn’t honor pages (or a version/SDK mismatch exists).

4) Check model & feature combinations

Certain model/feature mixes can yield limited outputs:

Try prebuilt-layout first (broadest coverage).

If you use a custom model, confirm it supports your doc type across all requested pages.

Include features=["ocr"] when you need full text extraction and your docs are image‑heavy.

5) Quick isolation steps

Test a different multi‑page PDF to rule out file structure anomalies.

Re‑run with only two pages (e.g., pages=["2-3"]) to see if selection starts working.

If bytes input fails, try URL input (or vice‑versa) to isolate client encoding issues.

6) If it still returns only page 1

Confirm the installed version:
import print

Try the previous minor version (known-good) to rule out a recent change.

Post the SDK version and a redacted request snippet showing exactly where pages is set, along with the operation ID. That will let us pinpoint whether this is a client parsing issue or a service-side regression.

Please review it and let me know if there are any remaining questions or if further guidance is needed. I’ll be happy to assist.

Thankyou!

3 answers

Your answer

Gowtham CP 7,725 Reputation points Volunteer Moderator

2025-12-17T07:27:32.3466667+00:00

Hi Kiran Dhumma

Thanks for the question.

1. SDK usage can cause only the first page to be returned If only the first page is returned, this is often related to how the Python SDK call is made. Unlike older Form Recognizer SDKs, in azure.ai.documentintelligence you must ensure the pages parameter is passed correctly in the analysis request and that your SDK version supports page ranges. The REST API supports this, but the parameter must be set in the correct part of the request.

2. Free tier limitation If your Document Intelligence resource is on the F0 (free) tier, the service analyzes only the first pages of a PDF regardless of the page range specified. Upgrading to S0 (standard) removes this limitation and enables full multi-page extraction. Reference: https://learn.microsoft.com/answers/questions/5569121/azure-document-intelligence-only-analyzing-2-docum

3. Document type matters Page-range behavior is best supported for PDF/TIFF files. For formats like DOCX or XLSX, the SDK may not treat content as page-based, which can appear as if only one page is processed. Reference: https://stackoverflow.com/questions/79475225/azure-documen-intelligence-python-sdk-doesnt-separate-pages

Could you please share the Python SDK code snippet you are using (including the client initialization and analyze call)? That will help confirm whether the request is being sent correctly.

I
Kiran Dhumma 25 Reputation points

2025-12-17T10:14:39.1+00:00

Any updates on the issue @Gowtham CP ?
Fremin Abreu Ortega 20 Reputation points

2025-12-17T16:10:14.28+00:00

I'm facing the same issue using C# sdk. It began happening without any code changes, yesterday the same code works fine...

Answer 1

RAHUL SHEDGE 0

Could you please confirm if any resolution or fix has been implemented for this issue?

Kiran Dhumma 25 Reputation points

2025-12-18T13:35:07.3933333+00:00

No not yet, I am still facing this issue.

Answer 2

Kiran Dhumma 25

This issue has been resolved in latest update.

Answer 3

Hey Kiran Dhumma,

Thank you for reaching out on the Microsoft Q&A.

It looks like you’re running into an issue with the Azure Document Intelligence Python SDK, where it's only processing the first page of your PDF, even when specifying a range of pages. This can be frustrating, especially since you haven't changed your code.

Here are a few things to consider:
1. SDK Parameter Check: First, ensure that you are correctly using the pages parameter in you begin_analyze_document call. In the newer SDK versions, it's crucial that the pages parameter is set up correctly. Refer to the documentation for proper usage.
2. Subscription Tier: Since you're using the version 1.0.2 of the SDK, please confirm that your Azure Document Intelligence resource is not on the F0 (free) tier. The free tier only analyzes the first two pages of a PDF, regardless of the specified page range. If you are on the free tier, upgrading to the S0 (standard) tier will allow full multi-page extraction.
3. Check Document Type: Ensure that your document is indeed a PDF. Other formats like DOCX or XLSX may not support the same page-based processing as PDFs do.
4. Service Updates or Limitations: If this behavior started recently and there haven't been any code changes on your part, it could be worth checking the Azure updates or forums for any announcements about service changes that might impact the Document Intelligence processing.
5. Retry Logic: The retry logic in your code looks good, but it’s also important to check if any rate limiting is coming into play. This can sometimes affect results if repeated calls are made too quickly.
References:

I Hope this helps. Do let me know if you have any further queries.

If this answers your query, please do click Accept Answer and Yes for was this answer helpful.

User's image

Thank you!

Kiran Dhumma 25 Reputation points

2025-12-26T05:09:14.91+00:00

Hey @Anshika Varshney Read the existing comments before posting. Adding incorrect or irrelevant information is not acceptable. This thread has already been answered. Continuing to comment without reviewing what’s already been discussed is a waste of time and needs to stop.

Share via

Azure Document Intelligence Python SDK Returns Data Only for First Page

3 answers

Your answer