Scoring Profiles ignored if ask question by using chat completions

Question

Scoring Profiles ignored if ask question by using chat completions

Hi,

I am building chatbot on top of Azure AI search RAG.The problem I am facing is;

I have applied Freshness function for Scoring profile - It does work on AI search Query section but When I ask by using Chat Completion I just completely ignore the scoring profile and also, format doesn't allow to separately mention to impose scoring profile.
I have mentioned boostedreranking in the semantic_configuration as well.

I don't know what's happening?Thanks,
Sadaf

User's image

Praneeth Maddali 1,110 Reputation points Microsoft External Staff Moderator

2025-10-22T13:05:51.8766667+00:00

Hi @Zaheer. Sadaf (Enterprise Services)

Thanks for reaching us regarding the issue Scoring Profiles ignored if ask question by using chat completions we started checking on it we will provide our suggestion ASP
Praneeth Maddali 1,110 Reputation points Microsoft External Staff Moderator

2025-10-22T13:50:23.1366667+00:00
Hi @Zaheer. Sadaf (Enterprise Services)

The reason scoring profiles (e.g., freshness function) appear ignored in Chat Completions with Azure AI Search RAG is that the On-Your-Data integration does not expose scoring_profile or scoring_parameters in the chat completion request body. By default, semantic ranking is performed after initial scoring, which can override your scoring profile boosts.

To resolve this:

Set a default scoring profile (e.g., "freshness-profile") at the index level to automatically apply it to all queries.

Update your semantic configuration to include
"rankingOrder": "boostedRerankerScore".
This ensures that freshness boosts are applied after semantic reranking. Be sure to use API version 2025-08-01-preview or later for full support.

If you require query-time scoring parameters (such as tagging or distance), consider implementing a custom RAG flow using the Azure Search SDK or REST API to explicitly specify scoringProfile and scoringParameters, and then pass those results to the Large Language Model (LLM).

Note: Field weights will not affect the final ranking after semantic reranking. Only scoring functions (e.g., freshness) can influence the final order when boostedRerankerScore is enabled.

References:

https://free.blessedness.top/en-us/azure/search/index-add-scoring-profiles

https://free.blessedness.top/en-us/azure/search/semantic-how-to-enable-scoring-profiles

https://free.blessedness.top/en-us/azure/ai-services/openai/use-your-data

Kindly let us know if the above helps or you need further assistance on this issue.

Please "upvote" if the information helped you. This will help us and others in the community as well.
Zaheer. Sadaf (Enterprise Services) 70 Reputation points

2025-10-22T14:09:46.0966667+00:00

Hi,

Yes,I have already enabled this in my semantic configurations:"rankingOrder": "boostedRerankerScore" and marked scoring profile as default in index.

Ideally when the query goes to Azure AI search engine from chat completion then it should automatically apply boosting reranking on it.

Final output which I get from the chat completion is semantically with no boosting applied.

Chat Playground in AI foundry is also giving me unboosted results.

The format of chat completion which I pasted in the above in question doesn't allow you to mention any scoring profile separately as parameter like we do in azure AI Search Json format.

Regards,
Sadaf Zaheer
Praneeth Maddali 1,110 Reputation points Microsoft External Staff Moderator

2025-10-22T16:22:21.49+00:00

Hi @Zaheer. Sadaf (Enterprise Services)

Thanks for confirming your configuration. You’re correct — even with the default scoring profile and boostedRerankerScore enabled, the On-Your-Data / Chat Completions integration does not currently pass scoringProfile or scoringParameters` to the search engine. This is why the freshness or other boosts are not applied in the final output.

Workaround: To ensure your scoring profile is honored, you can implement a custom RAG flow:

Perform the search query manually using the Azure Search REST API or SDK, specifying your scoring profile and any parameters.

Retrieve the top results from Search.

Pass these results explicitly to the Chat Completions messages array as context.

This approach guarantees that the search scoring (including freshness boosts) is applied before the content reaches the LLM.

Also, for future support, you may consider submitting a feature request to allow scoringProfile in Chat Completions queries.

Reference :

https://free.blessedness.top/en-us/azure/ai-foundry/openai/references/azure-search?tabs=python

https://free.blessedness.top/en-us/azure/search/semantic-how-to-enable-scoring-profiles

https://free.blessedness.top/en-us/azure/search/semantic-how-to-configure?tabs=portal

https://free.blessedness.top/en-us/python/api/overview/azure/search-documents-readme?view=azure-python

https://free.blessedness.top/en-us/azure/ai-foundry/agents/how-to/tools/azure-ai-search?tabs=keys%2Cazurecli

Kindly let us know if the above helps or you need further assistance on this issue.

Please "upvote" if the information helped you. This will help us and others in the community as well.
Zaheer. Sadaf (Enterprise Services) 70 Reputation points

2025-10-24T08:28:29.7433333+00:00

Hi,

Thanks for the work around,however it would become a bit tricky interms of Json return from Chat complete which later be consumed by App(Chat bot).

I've noticed one thing that Open AI Chat Completion refers to the old api version from 2023 of AI Search Service which is not supported for the Scoring Profile activation.

Can you please help me to do a work around on that - though I couldnt find any control over there?Thanks

Regards,
Sadaf
Praneeth Maddali 1,110 Reputation points Microsoft External Staff Moderator

2025-10-24T13:51:42.6333333+00:00
Hi @Zaheer. Sadaf (Enterprise Services)

Thank you for your detailed follow‑up.

You are correct that the On‑Your‑Data and Chat Completions integration currently uses an internal Azure Cognitive Search API version (2023‑07‑01‑Preview) that doesn’t fully support query‑time scoring profiles, scoring parameters, or rank fusion with scoring profiles. This limitation causes scoring boosts like freshness to be ignored, even when "rankingOrder": "boostedRerankerScore" and a default scoring profile are set.

Root Cause

The chat completion endpoint does not expose or pass scoringProfile and scoringParameters fields in the search plugin section.

The internal API used by the model for "on‑your‑data" retrieval still calls an older search API version.

Semantic ranking (vector + hybrid rerank) overrides scoring profile ranks.

Current Workarounds

Option 1: Custom RAG Flow (Recommended) You can enforce scoring profiles by separating the steps:

Call Azure Cognitive Search REST API or SDK directly (using the 2025‑08‑01‑Preview API version or later).

Specify "scoringProfile": "freshness-profile" and related parameters.

Pass the retrieved documents (with applied boosts) to the Chat Completions API as context via messages or data_sources.

This ensures the scoring logic (freshness boost, etc.) is honored before semantic reranking from the model.

Option 2: Update to Newest Azure AI Foundry SDK If you’re using SDK‑based RAG (e.g., azure-ai-projects Python SDK), ensure it targets a search dependency with API version 2025‑08‑01‑preview or higher. You can override this version by defining a custom search connection in code:

urces": [ { "type": "azure_search", "parameters": { "endpoint": f"https://{search_service}.search.windows.net", "api_version": "2025-08-01-preview", "authentication": { "type": "system_assigned_managed_identity" }, "index_name": search_index, "scoring_profile": "freshness-profile", "semantic_configuration": search_semantic_config } } ] }

While the public Chat Playground may still reference an older API, the SDKs and REST custom integrations already allow using the new API version.

References

Azure OpenAI On‑Your‑Data OverviewAzure Cognitive Search Scoring ProfilesSemantic Search and Boosted RerankingAzure AI Foundry: Azure Search IntegrationAzure Search REST API (2025‑08‑01‑Preview)

https://free.blessedness.top/en-us/azure/search/index-add-scoring-profiles

https://free.blessedness.top/en-us/azure/search/semantic-how-to-enable-scoring-profiles

https://free.blessedness.top/en-us/azure/ai-foundry/openai/references/azure-search?tabs=python

https://free.blessedness.top/en-us/rest/api/searchservice/

At this time, Chat Completions does not support specifying scoring profiles directly in extra_body. Microsoft’s internal roadmap includes alignment of the On‑Your‑Data plugin with the new API versions in upcoming releases.

Please try the Custom RAG flow or explicit search connection override workaroundKindly let us know if the above helps or you need further assistance on this issue.

Please "upvote" if the information helped you. This will help us and others in the community as well.
Zaheer. Sadaf (Enterprise Services) 70 Reputation points

2025-10-24T16:42:42.8666667+00:00

Hi,

Thanks for coming back to me on this.

Regards,
Sadaf Zaheer

Your answer

Praneeth Maddali 1,110 Reputation points Microsoft External Staff Moderator

2025-10-22T13:05:51.8766667+00:00

Hi @Zaheer. Sadaf (Enterprise Services)

Thanks for reaching us regarding the issue Scoring Profiles ignored if ask question by using chat completions we started checking on it we will provide our suggestion ASP
Praneeth Maddali 1,110 Reputation points Microsoft External Staff Moderator

2025-10-22T13:50:23.1366667+00:00

Hi @Zaheer. Sadaf (Enterprise Services)

The reason scoring profiles (e.g., freshness function) appear ignored in Chat Completions with Azure AI Search RAG is that the On-Your-Data integration does not expose scoring_profile or scoring_parameters in the chat completion request body. By default, semantic ranking is performed after initial scoring, which can override your scoring profile boosts.

To resolve this:

Set a default scoring profile (e.g., "freshness-profile") at the index level to automatically apply it to all queries.

Update your semantic configuration to include
"rankingOrder": "boostedRerankerScore".
This ensures that freshness boosts are applied after semantic reranking. Be sure to use API version 2025-08-01-preview or later for full support.

If you require query-time scoring parameters (such as tagging or distance), consider implementing a custom RAG flow using the Azure Search SDK or REST API to explicitly specify scoringProfile and scoringParameters, and then pass those results to the Large Language Model (LLM).

Note: Field weights will not affect the final ranking after semantic reranking. Only scoring functions (e.g., freshness) can influence the final order when boostedRerankerScore is enabled.

References:

https://free.blessedness.top/en-us/azure/search/index-add-scoring-profiles

https://free.blessedness.top/en-us/azure/search/semantic-how-to-enable-scoring-profiles

https://free.blessedness.top/en-us/azure/ai-services/openai/use-your-data

Kindly let us know if the above helps or you need further assistance on this issue.

Please "upvote" if the information helped you. This will help us and others in the community as well.
Zaheer. Sadaf (Enterprise Services) 70 Reputation points

2025-10-22T14:09:46.0966667+00:00

Hi,

Yes,I have already enabled this in my semantic configurations:"rankingOrder": "boostedRerankerScore" and marked scoring profile as default in index.

Ideally when the query goes to Azure AI search engine from chat completion then it should automatically apply boosting reranking on it.

Final output which I get from the chat completion is semantically with no boosting applied.

Chat Playground in AI foundry is also giving me unboosted results.

The format of chat completion which I pasted in the above in question doesn't allow you to mention any scoring profile separately as parameter like we do in azure AI Search Json format.

Regards,
Sadaf Zaheer
Praneeth Maddali 1,110 Reputation points Microsoft External Staff Moderator

2025-10-22T16:22:21.49+00:00

Hi @Zaheer. Sadaf (Enterprise Services)

Thanks for confirming your configuration. You’re correct — even with the default scoring profile and boostedRerankerScore enabled, the On-Your-Data / Chat Completions integration does not currently pass scoringProfile or scoringParameters` to the search engine. This is why the freshness or other boosts are not applied in the final output.

Workaround: To ensure your scoring profile is honored, you can implement a custom RAG flow:

Perform the search query manually using the Azure Search REST API or SDK, specifying your scoring profile and any parameters.

Retrieve the top results from Search.

Pass these results explicitly to the Chat Completions messages array as context.

This approach guarantees that the search scoring (including freshness boosts) is applied before the content reaches the LLM.

Also, for future support, you may consider submitting a feature request to allow scoringProfile in Chat Completions queries.

Reference :

https://free.blessedness.top/en-us/azure/ai-foundry/openai/references/azure-search?tabs=python

https://free.blessedness.top/en-us/azure/search/semantic-how-to-enable-scoring-profiles

https://free.blessedness.top/en-us/azure/search/semantic-how-to-configure?tabs=portal

https://free.blessedness.top/en-us/python/api/overview/azure/search-documents-readme?view=azure-python

https://free.blessedness.top/en-us/azure/ai-foundry/agents/how-to/tools/azure-ai-search?tabs=keys%2Cazurecli

Kindly let us know if the above helps or you need further assistance on this issue.

Please "upvote" if the information helped you. This will help us and others in the community as well.
Zaheer. Sadaf (Enterprise Services) 70 Reputation points

2025-10-24T08:28:29.7433333+00:00

Hi,

Thanks for the work around,however it would become a bit tricky interms of Json return from Chat complete which later be consumed by App(Chat bot).

I've noticed one thing that Open AI Chat Completion refers to the old api version from 2023 of AI Search Service which is not supported for the Scoring Profile activation.

Can you please help me to do a work around on that - though I couldnt find any control over there?Thanks

Regards,
Sadaf
Zaheer. Sadaf (Enterprise Services) 70 Reputation points

2025-10-24T16:42:42.8666667+00:00

Hi,

Thanks for coming back to me on this.

Regards,
Sadaf Zaheer

Share via

Scoring Profiles ignored if ask question by using chat completions

Your answer