Microsoft retires Bing Search APIs: what developers need to know about the transition to Azure AI Agents

Microsoft retired the Bing Search APIs on August 11, 2025, decommissioned every existing instance, and pointed customers to Grounding with Bing Search in Azure AI Agents, now part of Microsoft Foundry Agent Service.

This article covers what replaced the APIs, how the Foundry Web Search and Grounding with Bing tools differ from the old endpoints, the compliance changes, and the steps to move an app across.

Microsoft Retires Bing Search APIs
A grounded agent answer lists the Bing search query and numbered source links beneath the response. (Image: Microsoft)

The end of Bing Search APIs and a forced migration

The retirement was total: existing Bing Search API instances stopped working, and new sign-ups closed. Any app still calling the old endpoints needs a new design, not a new key.

  1. List every service that called a Bing Search API endpoint and what it did with the results.
  2. Decide whether an AI agent answer with citations can replace the raw result list your app used.
  3. Create a Microsoft Foundry project and deploy a supported Azure OpenAI model.
  4. Start with the Web Search tool, which needs no separate Bing resource.
  5. Move to Grounding with Bing Search only if you need count, freshness, market or set_lang, or a non-OpenAI model.
  6. Update your interface to show the website citations and the Bing query link that every response returns.

Why is Microsoft doing this?

Microsoft's lifecycle notice gives no business reason beyond the redirect. What it does say is where search now lives: inside AI agents that use Bing results to ground a model's answer.

Treat Bing web search as an agent tool from now on, not as a standalone results API. Microsoft's own migration path, documentation and pricing all sit under Foundry Agent Service, and the retirement notice names Grounding with Bing Search as the replacement.

What is “Grounding with Bing Search” in Azure AI Agents?

Grounding with Bing Search lets an agent pull real-time public web data into a model's response, which gets past the model's knowledge cutoff.

Stage What happens
Query formulation The agent spots an information gap and writes search queries from the user's input
Search execution The grounding tool sends the queries to Bing and gets results back
Information synthesis The agent works the results into its answer
Source attribution The answer cites the source URLs it used

The classic Foundry agents that first carried this tool are deprecated and retire on March 31, 2027, so build on the new agents API.

Grounding with Bing Search resource option in the Azure portal Add menu
Grounding with Bing Search sits beside Bing Search and Bing Custom Search under Add in Bing Resources. (Image: Microsoft)

Comparing the old and new models

Need Bing Search APIs (retired) Foundry Web Search Grounding with Bing Search
Status Retired August 11, 2025 Generally available Generally available
Bing resource Your own API key Managed by Microsoft You create a Grounding with Bing resource
What you get back Raw search results A model answer with citations A model answer with citations; no raw content
Tuning Query parameters user_location, search_context_size count (default 5, max 50), freshness, market, set_lang
Models Not applicable Azure OpenAI models Azure OpenAI and other Foundry models sold by Azure
Restrict to chosen domains Bing Custom Search Yes, with a Bing Custom Search instance Grounding with Bing Custom Search (preview)

Critical analysis: Notable strengths

The new model does add things the old endpoints never offered. The four strengths below map to documented features.

Strength Documented feature behind it
Single platform Search, models and agents share one Foundry project
Current answers Real-time web data past the model's knowledge cutoff
Admin control RBAC and Azure Policy can allow or block the tools
Room to grow Same tools across Python, C#, JavaScript, Java and REST

1. Unified AI experience for enterprises

Teams already on Azure OpenAI gain the most: web grounding is one more tool on an agent they already run. Web Search needs no roles beyond Foundry project access, and Grounding with Bing Search attaches through a project connection.

Azure AI Agents unified experience with Create an agent in Microsoft Foundry
The Foundry home page offers Create new for projects and Create an agent for agent building. (Image: Microsoft)

2. Enhanced real-time web access

Use the freshness parameter on Grounding with Bing Search when answers must reflect recent pages. It filters results by Day, Week, Month or a date range, much like Bing's own search operators and filters, so the agent can be held to what Bing discovered in that window.

3. Stronger governance and compliance posture within Azure

Governance is stronger for who can use the tools, not for where the data goes. Admins can control access with RBAC, unregister the Microsoft.Bing provider and block resource creation by policy, but query data still leaves the Azure compliance boundary.

4. AI-agent ecosystem and future potential

Build new search features on the current Foundry agents API rather than the classic one. Web Search also backs deep research agents, and Microsoft Agent Framework exposes a Bing grounding tool, while classic agents retire in March 2027.

Azure AI agent ecosystem shown in the Foundry Agents playground setup pane
The playground setup pane holds the agent name, model deployment, instructions and a Knowledge section. (Image: Microsoft)

Challenges and pain points

The costs of the move are just as concrete. Each row below is a documented limit, not a forecast.

Pain point What the documentation says
More setup Foundry project, model deployment, roles and a project connection
Data leaves Azure The Data Protection Addendum does not apply
Preview pieces Grounding with Bing Custom Search is in preview
No raw results Developers and end users cannot access the raw content

1. Increased complexity and steeper learning curve

Requirement Detail
Azure subscription With rights to create resources
Contributor role On the resource group while you create the Bing resource and get its keys
Foundry Project Manager role To create the project connection
Foundry User role To create and run agents
Foundry project and model A project endpoint and a deployed model
SDK azure-ai-projects (Python), Azure.AI.Extensions.OpenAI (C#), @azure/ai-projects (JavaScript)

The Foundry roles were renamed from Azure AI User, Azure AI Owner and Azure AI Project Manager, so older scripts may still use those names.

2. Data sovereignty and compliance concerns

Check with your compliance team before sending regulated data through any Bing grounding tool. Data sent to Grounding with Bing flows outside the Azure compliance and Geo boundary, and using it waives elevated Government Community Cloud commitments such as data sovereignty.

3. Dependency on preview and unsupported features

Feature Stage
Web Search Generally available
Grounding with Bing Search Generally available
Grounding with Bing Custom Search Preview
Classic Foundry agents Deprecated, retiring March 31, 2027

Not every model supports the Bing tools, so check Microsoft's tool support by region and model before you pick a deployment.

4. Disrupted workflows and a swift sunset

Old workflow What changes
Show a list of search results Not possible; the tool returns a model answer, not raw results
Store or reprocess results Raw content is not available to developers
Custom result layout You must display the citations and Bing query URL as Microsoft provides them
Simple per-call pricing Usage-based grounding charges on top of model usage

Navigating compliance and data privacy

Grounding with Bing's use and display requirements apply to every interface that shows a grounded answer.

  1. Read the response's arguments field to get the website URLs and the Bing query URL.
  2. Show both the source links and the Bing query link in your interface.
  3. Rewrite the query URL's endpoint to www.bing.com, giving https://www.bing.com/search?q={search query}.
  4. Keep both references in the exact form Microsoft provides.
  5. If your project is network-secured, note that these tools act as public endpoints and ignore VPNs and private endpoints.

Developer sentiment and community response

The practical complaints from developers line up with the documented gaps. These are the questions a team has to answer before it can migrate.

Question teams ask Answer from the documentation
Can we still get a plain results list? No; raw content is not returned
Do we need Azure OpenAI? Web Search supports Azure OpenAI models; Grounding with Bing Search also supports other Foundry models
Does it cost extra? Yes, grounding tools have their own usage-based charges
Can we limit results to our sites? Yes, through a Bing Custom Search instance

Strategic implications for Microsoft and the broader cloud AI ecosystem

Expect web search on Azure to be sold as part of the agent stack rather than as a separate API. Web Search is now the recommended starting tool, and grounding also powers Copilot Studio public website knowledge sources and the Azure OpenAI Responses API web search.

Guidance for developers and IT decision-makers

  1. If you just need current web answers, add the Web Search tool to a prompt agent.
  2. If you need tuning parameters, create a Grounding with Bing Search resource in the Azure portal.
  3. Find its keys under Resource Management > Keys.
  4. Add a connection for it in your Foundry project and record the connection name or ID.
  5. Set BING_PROJECT_CONNECTION_NAME for SDK samples or BING_PROJECT_CONNECTION_ID for REST.
  6. Create the agent with the BingGroundingTool and send a test prompt with tool_choice="required".

Activate the Contributor role only while you create the resource, using Microsoft Entra Privileged Identity Management, and turn it off afterwards.

Migrate to Grounding with Bing Search by adding Knowledge to an agent
Knowledge is where an agent gets data sources for grounding responses; select Add to attach one. (Image: Microsoft)

Grounding with Bing Search not working

Agent does not use web grounding

The tool is not configured, or the model does not support it.

  1. Confirm the tool is in the agent definition.
  2. Set tool_choice="required" to force it.
  3. Check that your model deployment supports the tool.

No citations in the response

The model answered without the search results.

  1. Tell the agent in its instructions to always cite sources.
  2. Use tool_choice="required".

Search results are not relevant

The query did not capture what the user meant.

  1. Improve the agent instructions on how to build queries.
  2. Set market and set_lang on the Bing tool.

Tool blocked by administrator

The organisation disabled web grounding tools.

  1. Ask your Azure administrator to enable access.
Grounding with Bing Search not working: check the Bing resource exists
Filter the resource group for bing to confirm the Bing Resource is deployed before testing the agent again. (Image: Microsoft)

Recommendations for Microsoft

Microsoft should give developers a documented path for apps that need raw search results, not only agent answers. The retirement removed that use case entirely, and neither Web Search nor Grounding with Bing Search returns raw content.

Looking forward: The new normal for search-as-a-service

Date Event
August 11, 2025 Bing Search APIs retired and instances decommissioned
August 31, 2026 Legacy SOAP and POX Bing Webmaster APIs retire
March 31, 2027 Classic Foundry agents retire

The Bing Webmaster API is a separate product from the Bing Search APIs; its REST API stays available.

Bing Search API retirement FAQ

When were the Bing Search APIs retired?

Microsoft retired the Bing Search APIs on August 11, 2025. Every existing instance was decommissioned, and the product closed to new customers on the same date.

What replaces the Bing Search APIs?

Microsoft points customers to Grounding with Bing Search in Azure AI Agents, now Foundry Agent Service. The Foundry Web Search tool is the recommended starting point because it needs no separate Bing resource.

What is the difference between Web Search and Grounding with Bing Search?

Web Search uses a Microsoft-managed Bing resource and supports Azure OpenAI models. Grounding with Bing Search needs your own Bing resource, adds count, freshness, market and set_lang parameters, and supports more Foundry models.

Can I get raw search results from Grounding with Bing Search?

No. Developers and end users cannot access the raw content. The response contains the model's answer, citations to the websites used and a link to the Bing query.

Does Grounding with Bing Search keep data inside Azure?

No. Data sent to Grounding with Bing flows outside the Azure compliance and Geo boundary, and the Microsoft Data Protection Addendum does not apply to it.

Does Grounding with Bing Search cost extra?

Yes. Web Search, Grounding with Bing Search and Grounding with Bing Custom Search are charged on top of standard Azure OpenAI usage. Microsoft publishes the rates on its web grounding pricing page.

Is the Bing Webmaster API being retired too?

Only its legacy SOAP and POX APIs, which retire on August 31, 2026. The Bing Webmaster REST API remains, and Microsoft advises moving to it.

Philip Celasco

Philip is a Texas-based technology writer and IT administrator at Techdows.com with more than 10 years of experience creating practical content for everyday users and professionals. He specializes in web browsers, particularly Chromium-based platforms such as Google Chrome, Microsoft Edge, Brave, and Opera. Through his work as an IT administrator, Philip has hands-on experience managing devices, configuring browser policies, troubleshooting software and network issues, and helping people resolve problems that affect productivity and security. His articles are based on practical testing and real-world technical experience. He covers browser settings, extensions, performance problems, privacy controls, security features, and Windows troubleshooting. Outside work, Philip enjoys the quieter side of life in Texas and stepping away from the screen when he can. He has two kids, two cats and loves to play golf with his mother during the weekends.

Leave a Reply

Your email address will not be published. Required fields are marked *