Azure Search AI is not providing accurate result

Question

Azure Search AI is not providing accurate result

Admin 50

Hi,

we are using azure search service with bring your own data as pdf.

we are used RAG functionality while creating index and importing pdf, but still not able to get the desired result.

Admin 50 Reputation points

2025-11-20T18:55:25.5233333+00:00

Yes we have followed the semantic keyword search and also used vector based search, added skillset to chuck the pdf. but still not getting accurate result.

we have lot of pdf which might go upto 5-10k to be indexed.
Praneeth Maddali 2,355 Reputation points Microsoft External Staff Moderator

2025-11-20T19:20:22.17+00:00
Hi @Admin

Thank you for reaching us regarding the issue with Azure AI Search not returning accurate results for your PDFs. In most similar cases, a few small adjustments lead to much better relevance right away.

To help you as quickly as possible, could you provide:

An example query

What question are you asking?

What answer do you expect?

What result are you getting?

PDF type Is it mostly plain text, or does it have many tables, images, scanned pages, headers/footers, or a complex layout?

Embedding model : In the portal>under Index > vector field > Is it text-embedding-ada-002, text-embedding-3-large, or something else?

Search type : • Pure vector only • Hybrid (vector + keyword) • Semantic ranker enabled? (Yes/No)

Chunk size & overlap – Or just note if you’re using the default from the wizard

Data volume – Approximate amount (e.g., 20 PDFs / 500 MB)

Ingestion method – “Import and vectorize data” wizard (one-click RAG) – Custom indexer/skillset

With these details, especially items 1–4, we can usually identify the solution in the next step. Looking forward to your responses to help resolve this quickly.
Praneeth Maddali 2,355 Reputation points Microsoft External Staff Moderator

2025-11-24T02:28:11.2133333+00:00

Hi @Admin

I'm just reaching out to see if your issue has been resolved or if you've had a chance to review my previous comment?

1 answer

Your answer

Admin 50 Reputation points

2025-11-20T18:55:25.5233333+00:00

Yes we have followed the semantic keyword search and also used vector based search, added skillset to chuck the pdf. but still not getting accurate result.

we have lot of pdf which might go upto 5-10k to be indexed.
Praneeth Maddali 2,355 Reputation points Microsoft External Staff Moderator

2025-11-20T19:20:22.17+00:00

Hi @Admin

Thank you for reaching us regarding the issue with Azure AI Search not returning accurate results for your PDFs. In most similar cases, a few small adjustments lead to much better relevance right away.

To help you as quickly as possible, could you provide:

An example query

What question are you asking?

What answer do you expect?

What result are you getting?

PDF type Is it mostly plain text, or does it have many tables, images, scanned pages, headers/footers, or a complex layout?

Embedding model : In the portal>under Index > vector field > Is it text-embedding-ada-002, text-embedding-3-large, or something else?

Search type : • Pure vector only • Hybrid (vector + keyword) • Semantic ranker enabled? (Yes/No)

Chunk size & overlap – Or just note if you’re using the default from the wizard

Data volume – Approximate amount (e.g., 20 PDFs / 500 MB)

Ingestion method – “Import and vectorize data” wizard (one-click RAG) – Custom indexer/skillset

With these details, especially items 1–4, we can usually identify the solution in the next step. Looking forward to your responses to help resolve this quickly.
Praneeth Maddali 2,355 Reputation points Microsoft External Staff Moderator

2025-11-24T02:28:11.2133333+00:00

Hi @Admin

I'm just reaching out to see if your issue has been resolved or if you've had a chance to review my previous comment?

Answer 1

To improve the accuracy of results when using Azure Search with your own PDF data and the Retrieval Augmented Generation (RAG) functionality, consider the following strategies:

Indexing Quality: Ensure that the content from your PDFs is being indexed correctly. The indexing process should convert the PDF content into a searchable format. If the content is not being extracted properly, it may lead to inaccurate search results.
Field Mapping: Review your field mappings in the index to ensure that they accurately reflect the structure of your PDF documents. Mismatched mappings can cause important content to be overlooked during searches.
Query Optimization: Analyze the queries you are using. Sometimes, refining the queries to be more specific or using filters can yield better results. Consider using both vector and keyword searches to enhance the search effectiveness.
Reranking Mechanisms: Implement reranking techniques to improve the relevance of the results returned. You can use language models or cross-encoders for reranking to evaluate the results based on their semantic relevance.
Semantic Ranking: Utilize Azure's semantic ranking capabilities, which leverage deep learning models to promote the most semantically relevant results. This can help in retrieving more accurate results based on user queries.
Testing and Iteration: Continuously test and iterate on your indexing and querying strategies. Monitor the performance and adjust your approach based on the results you observe.

By focusing on these areas, you should be able to enhance the accuracy of the search results provided by Azure Search with your PDF data.

References:

Share via

Azure Search AI is not providing accurate result

1 answer

Your answer