Get Searchable PDF from Azure Document Intelligence Service

Question

Get Searchable PDF from Azure Document Intelligence Service

Fred 95

I'm trying to use python to convert a scanned PDF into a searchable PDF. I see here that it is supported but I am looking for a code example or sample on how to do this. Where I am struggling is the request seems to require a request_id but I am unclear on how to get that.

I have tried using the document intelligence service. The below code is not working

from loguru import logger
from azure.core.credentials import AzureKeyCredential
from azure.ai.documentintelligence.models import (
    AnalyzeDocumentRequest,
)
from azure.ai.documentintelligence import DocumentIntelligenceClient
import os
def extract_pdf(
    file: str,
    # tables: DBTables,
    use_cache: bool = True,
    text_model: str = "prebuilt-read",
) -> str:
    """creates markdown formatted text"""
    logger.info("=======Starting Text Extraction=======")
    filename, file_extension = os.path.splitext(file)
    out = filename + ".ocr.pdf"
    if use_cache:
        logger.debug("Checking cache")
        if os.path.isfile(out):
            logger.debug("Results found in cache, done")
            with open(out, "r") as f:
                return f.read()
        logger.debug("Results not in cache, extracting")
    logger.debug("running poller")
    credential = AzureKeyCredential("MYKEY")
    analysis_client = DocumentIntelligenceClient(
        "MYENDPOINT", credential
    )
    with open(file, "rb") as f:
        poller = analysis_client.begin_analyze_document(
            text_model,
            AnalyzeDocumentRequest(bytes_source=f.read()),
            output=["pdf"],
        )
    result = poller.result()
    print(result)
    with open(out, "wb") as f:
        f.write(result)
    return out

Any guidance or a working example would be appreciated

Anonymous

2025-09-26T00:42:10.2+00:00
Hello Fred

The issue you’re facing is due to two key points:

The code currently writes the analysis result object (AnalyzeResult) directly to a .pdf file. This object contains structured JSON data, not the actual searchable PDF content.

The request_id mentioned in the documentation refers to the resultId of the long‑running operation. This ID is returned in the Operation-Location header when you start the analysis.

To generate a searchable PDF using Azure Document Intelligence (v4.0), the correct process is:

Step 1: Start the analysis with the Read model (prebuilt-read) and request PDF output by setting output=pdf (REST) or output=["pdf"] (SDK).

Step 2: Poll the URL provided in the Operation-Location header until the status is succeeded. This URL contains the analyzeResults/{resultId}.

Step 3: Download the actual searchable PDF by calling the Get Analyze Result PDF endpoint using that resultId. Save the returned bytes to a .pdf file.

For reference:

Analyze Document (REST API)

Get Analyze Result PDF (REST API)To generate a searchable PDF using Azure Document Intelligence (v4.0), the correct process is:

Step 1: Start the analysis with the Read model (prebuilt-read) and request PDF output by setting output=pdf (REST) or output=["pdf"] (SDK).

Step 2: Poll the URL provided in the Operation-Location header until the status is succeeded. This URL contains the analyzeResults/{resultId}.

Step 3: Download the actual searchable PDF by calling the Get Analyze Result PDF endpoint using that resultId. Save the returned bytes to a .pdf file.

For reference:

Analyze Document (REST API)

Get Analyze Result PDF (REST API)

Using the Python SDK: After calling begin_analyze_document(..., output=["pdf"]), note that poller.result() returns an AnalyzeResult (not PDF bytes). To get the PDF:

Retrieve the operation URL via poller.continuation_token().

Extract the resultId from that URL.

Call DocumentIntelligenceClient.get_analyze_result_pdf(model_id, result_id) to download the searchable PDF bytes.

Helpful resources:

Azure AI Document Intelligence Python SDK README

Using the Python SDK: After calling begin_analyze_document(..., output=["pdf"]), note that poller.result() returns an AnalyzeResult (not PDF bytes). To get the PDF:

Retrieve the operation URL via poller.continuation_token().

Extract the resultId from that URL.

Call DocumentIntelligenceClient.get_analyze_result_pdf(model_id, result_id) to download the searchable PDF bytes.

Helpful resources:

Azure AI Document Intelligence Python SDK README

Hope it helps!

Thank you

1 answer

Your answer

Answer 1

Deleted

This answer has been deleted due to a violation of our Code of Conduct. The answer was manually reported or identified through automated detection before action was taken. Please refer to our Code of Conduct for more information.

2 deleted comments

Comments have been turned off. Learn more