For the documented 2024-11-30 API, searchable PDF output belongs to the prebuilt-read model. Requesting ordinary JSON text extraction is not the same as obtaining a PDF with embedded searchable text. Check the model, requested output and final retrieval operation separately.
This guide explains the supported workflow without running a billable analysis. Use an authorized Azure resource and approved documents for any implementation. A searchable PDF remains an OCR-derived document whose text needs review; it is not evidence that every character was recognized correctly.
Decide which artifact the recipient needs
A JSON response is useful for applications extracting text and structure. A searchable PDF is a document artifact that combines a visible page with recognized text for search. Choose the required result before wiring the response into storage or an email process.
Microsoft's Read model documentation identifies the searchable PDF capability, the prebuilt-read model and the output=pdf request choice for the specified API. Do not assume that another model supports the same output merely because it also performs OCR.
Keep Layout and Read distinct
Layout extraction serves structural analysis tasks. The searchable-PDF feature described here has its own model requirements. A request built for table extraction should not be relabeled as a searchable-PDF request without checking the documented capability.
Preserve the original scan
Store the received source according to the approved records process. Label the searchable derivative clearly. Adding recognized text should not silently replace an authoritative signed original or imply that the source has been substantively corrected.
Follow the asynchronous sequence
The initial Analyze request starts processing. Follow its operation reference until the service reports the appropriate completion state, then use the documented operation for retrieving the PDF artifact. Do not save a status response or JSON result under a .pdf extension.
The official add-on capability reference describes the searchable PDF output flow and its model scope. Use the version-matching API documentation when implementing authentication and result retrieval; avoid copying an older preview route without checking it.
| Step | What to verify |
|---|---|
| Analyze request | Correct API version, model and PDF output choice |
| Status handling | Completed success rather than accepted or running |
| Artifact retrieval | Actual PDF response rather than ordinary JSON |
| Destination save | Complete file with the intended content type and name |
Validate the resulting document independently
Open the saved artifact in an approved PDF viewer. Confirm every page is present and readable. Search for distinctive phrases from several pages and compare copied text with the visible scan, including punctuation and any reference numbers.
Test text, not only the extension
A file named report.pdf can still be a mis-saved error response or an image-only original. Confirm that it opens as a PDF and that representative page text is searchable. The destination action reporting success is not sufficient evidence.
Review difficult source regions
Include faint print, accents, small references and unusual layouts in the acceptance set where they occur in the real work. Do not claim a universal accuracy rate from a few clean samples. Keep uncertain values under the organization's review procedure.
Keep service cost and output capability separate
The documentation describes searchable PDF output as included for the specified Read model version without an extra output-generation charge. That does not make all analysis, storage, networking or workflow execution free. Check current pricing and the resource's commercial configuration before production use.
Do not publish an Azure key or signed source URL in a troubleshooting example. Use the approved authentication mechanism and keep resource details in the application's secure configuration.
Our OneDrive scanned-PDF OCR guide describes another searchable-document workflow. Compare the resulting artifact and its review needs rather than treating all OCR features as interchangeable services.
Also test the storage destination's handling of a failed retrieval with a harmless simulated response. It should retain a visible failure state rather than publish an empty or incorrectly typed file to the receiving library.
Keep a concise record of the API version, model, requested artifact and sample verification. Accept the integration when it retrieves a genuine searchable derivative with complete pages, preserves the source as required and handles incomplete or failed operations without distributing a mislabeled file.
