Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more

Update: Microsoft has already retired the Bing Search APIs. The shutdown took effect on August 11, 2025, and new customer sign-ups were discontinued. The often-repeated claim that Microsoft is killing the API “later this year” is now outdated.

Microsoft’s announced direction is Grounding with Bing, particularly Grounding with Bing Search in Azure AI Foundry Agent Service. It is not a renamed endpoint for the old API. The former Bing Web Search API returned structured search results to an application; the replacement gives web-derived context to an AI agent, which then generates the response.

What changed when the Bing Search API was retired?

The standalone Bing Search APIs were decommissioned on August 11, 2025. Microsoft also stopped accepting new customers, so creating a new instance of the old Bing Web Search API is no longer a migration option.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

The distinction matters because many migration guides describe Grounding with Bing Search as though it were the same service behind a new URL. It is not. A traditional search API generally follows this pattern:

  1. Your application sends a query.
  2. Bing returns structured JSON containing webpages, titles, snippets, URLs, and related fields.
  3. Your code decides how to rank, display, store, or process those results.

Grounding with Bing Search follows a different pattern:

  1. An Azure AI agent receives a user request.
  2. The model decides whether it needs web information and calls the grounding tool.
  3. Bing supplies relevant web-derived chunks to the model.
  4. The model writes the final answer and includes citations.

Raw Bing results are not exposed to the developer or end user. Microsoft’s documentation says the response includes references to consulted websites and a link to the Bing query, but it does not provide the underlying grounding content as raw search output.

Microsoft’s replacement is called Grounding with Bing

Microsoft’s current product page uses the broader name Grounding with Bing. It lists two separate offerings:

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
Offering Purpose Where it is used
Grounding with Bing Search Retrieves current information from the wider web for an AI-generated answer Azure AI Foundry Agent Service, Azure AI Search, and Foundry Knowledge
Grounding with Bing Custom Search Grounds answers against a restricted set of websites or domains Azure AI Foundry Agent Service, Azure AI Search, and Foundry Knowledge

In Azure AI Foundry Agent Service, the capability can appear as a resource or tool. In Azure AI Search and Foundry Knowledge, it is presented as a Web Knowledge Source.

For new agent projects, Microsoft’s documentation also uses the label Web Search tool. Do not confuse that newer recommended terminology with the older Grounding with Bing Search (classic) workflow. The classic documentation applies to Microsoft Foundry classic agents, which Microsoft says are deprecated and scheduled for retirement on March 31, 2027.

How much does Grounding with Bing cost?

Microsoft currently lists both Grounding with Bing Search and Grounding with Bing Custom Search at:

  • $14 per 1,000 transactions
  • 150 transactions per second maximum
  • 1 million transactions per day maximum

A transaction is not necessarily one user question. In Azure AI Foundry, billing is based on the number of grounding tool calls made during an agent run. The model can call the tool more than once if it decides it needs additional information or context.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

For example, if an agent makes three Bing grounding calls while answering one question, that run can consume three transactions. More calls can also mean higher latency. Microsoft says the count can be inspected from the run-step details.

The cost of a single call is relatively small, but high-volume applications should model the number of tool calls per run rather than multiplying the user-question count by a single search price.

What developers can and cannot build with it

It works well for current-information answers

Grounding with Bing Search is intended for agents that need information from the current web. Examples include an assistant that explains recent news, checks current product information, or answers questions whose facts may have changed since the model was trained.

It is not a general replacement for raw search results

If your application used the old Bing API to render a search-results page, collect URLs, extract snippets, build a crawler pipeline, or store Bing’s JSON response, Grounding with Bing Search is not a drop-in replacement. The application receives the model-generated answer and required references, not the raw search payload.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Microsoft’s product page also describes Grounding with Bing as an add-on for selected Microsoft products. It does not present it as a general-purpose standalone search API whose output can be freely retrieved by arbitrary applications.

It is not intended to summarize an entire webpage

The tool returns relevant chunks rather than handing the model a complete page for deterministic processing. The model chooses which portions to use. Microsoft therefore does not recommend it for summarizing an entire webpage from beginning to end.

Use Custom Search for a controlled domain set

Grounding with Bing Search is designed for broad web retrieval, not for restricting searches to particular domains. If an answer must come only from approved sites—for example, a company’s documentation, government websites, or a curated publisher list—use Grounding with Bing Custom Search instead.

Important deployment restrictions

Several Azure requirements can affect whether the service is usable in a real project:

What’s actually slowing this PC down?

Pick the symptom - the matching free tool is one click away.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
  • Subscription: Grounding with Bing Search is available only with paid and pay-as-you-go Azure subscriptions. Sponsored subscriptions and subscriptions based on free credits are not eligible to deploy it.
  • Permissions: Creating the resource requires Owner or Contributor permissions on the subscription or resource group.
  • Resource group: Microsoft recommends placing the Bing resource in the same resource group as the Azure AI Agent, AI Project, and related resources.
  • Network: The service requires normal network access. It does not work with agents using a VPN or Private Endpoints.
  • Compliance: Customer data sent through Grounding with Bing Search is transferred outside the Azure compliance boundary. It does not have the same data-processing terms, processing-location guarantees, compliance standards, or certifications as Foundry Agent Service.

That compliance point should be reviewed before sending confidential prompts, regulated data, customer records, or internal documents to the service. A private Azure network design does not automatically make Bing grounding private.

Creating the resource from Azure

For a code-first workflow or deployment template, first ensure that the Bing resource provider is registered:

az provider register --namespace 'Microsoft.Bing'

The caller needs permission to perform /register/action. That permission is included in the built-in Contributor and Owner roles.

After the resource is created, attach or select the grounding capability in the relevant Azure AI Foundry Agent Service project. The exact portal labels depend on whether the project uses the current Web Search tool or the classic agent experience, so avoid treating screenshots for “Grounding with Bing Search (classic)” as instructions for a new architecture.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Optional search controls

The classic Grounding with Bing Search documentation describes several parameters that can refine the web lookup:

Parameter Accepted examples and behavior
count Defaults to 5 and supports up to 50 results. The actual number may be lower, results may overlap, and the model may not use every result. It affects webpage results only.
freshness Day, Week, Month, a date range such as 2025-01-01..2025-01-31, or a single date such as 2019-02-04.
market Use a value such as en-US. Microsoft recommends specifying it because unsupported values fall back to an internally mapped market.
set_lang Prefer a culture code such as en-US. Invalid or unsupported values can fall back to English.

These controls are useful, but they do not turn the service into deterministic structured search. The model still decides how to use the returned context.

Citations are part of the application’s job

Applications with a custom interface must display both kinds of references Microsoft exposes:

  • Website URLs: available through the response’s annotations.
  • Bing query URLs: available from the agent run-step details.

The documented Python pattern for inspecting run steps is:

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
run_steps = project_client.agents.runs_steps.list(
    run_id=run.id,
    thread_id=thread.id
)
run_steps_data = run_steps["data"]
print(f"Last run step detail: {run_steps_data}")

For displaying a Bing query citation, Microsoft recommends replacing the endpoint in the query URL with www.bing.com, producing a URL in this form:

https://www.bing.com/search?q={search query}

Do not silently discard the citations when building a chat front end. The generated answer depends on web content, and users need to be able to inspect the cited sources and query.

Model support and latency caveat

The current classic documentation says the classic grounding tool works with Azure OpenAI models supported by Agent Service except gpt-4o-mini, the 2024-07-18 model, and GPT-5 models. That is a caveat for the documented classic integration, not a blanket statement about every Foundry agent or newer Web Search implementation. Check the support matrix for the specific tool and API version you plan to deploy.

Latency is also different from calling a search endpoint directly. End-to-end time includes model preprocessing, the Bing grounding request, and model post-processing. An agent may perform repeated grounding calls during one run, making both response time and charges variable.

Free tools Windows power users keep installed

One-click scans. No signup required.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

What existing Bing Search API users should do

  1. Identify the old API dependency. Find applications, Azure resources, keys, SDK packages, and deployment templates that reference Bing Search APIs.
  2. Decide whether you need search data or an AI answer. If your code needs raw results, URLs, snippets, or indexing input, Grounding with Bing is not equivalent. Evaluate another search provider or a different Microsoft product that exposes the data you require.
  3. Choose the current Microsoft path for agent answers. For broad web context, evaluate Grounding with Bing Search or the current Web Search tool. For approved domains, evaluate Grounding with Bing Custom Search.
  4. Check subscription and network eligibility. Confirm that the subscription is paid or pay-as-you-go and that the agent does not depend on a VPN or Private Endpoint.
  5. Review data handling. Treat the transfer outside the Azure compliance boundary as a design and governance issue, not merely a networking detail.
  6. Instrument tool calls. Inspect run steps so you can measure transactions per run, repeated searches, citation behavior, and latency before estimating production cost.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Support on Ko-Fi

FAQ

Is the Bing Search API being retired later this year?

No. Microsoft retired the Bing Search APIs on August 11, 2025, and discontinued new customer sign-ups. Statements describing a future shutdown are outdated.

Is Grounding with Bing Search the same as the old Bing Search API?

No. The old API returned structured search results directly. Grounding with Bing Search supplies relevant web-derived context to an AI model, which generates the final answer. Raw grounding content is not returned to the application.

How much does Grounding with Bing Search cost?

Microsoft lists a price of $14 per 1,000 transactions. A transaction is based on a grounding tool call during an agent run, so one user question can consume multiple transactions.

Can Grounding with Bing Search search only my company’s websites?

It is not designed for restricted domain searches. Use Grounding with Bing Custom Search when the answer must be based on a defined set of websites.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Does Grounding with Bing work with a VPN or Private Endpoint?

No. Microsoft’s documentation says the service works only with agents that have normal network access; VPN and Private Endpoint configurations are unsupported.

Can I use it to retrieve Bing’s raw JSON results?

No. The service exposes the generated response and required citations, not the raw Bing grounding payload. Applications that need structured search data require a different solution.

The Bottom Line

Microsoft’s Bing Search API retirement is no longer a future event: it happened on August 11, 2025. Grounding with Bing is Microsoft’s replacement direction for AI agents, but it solves a different problem. It provides web context for an LLM and citations for the resulting answer; it does not provide a new standalone Bing search endpoint or raw JSON feed.

For a new Azure agent, evaluate the current Web Search tool or Grounding with Bing Search, use Custom Search for domain restrictions, and budget for tool calls rather than just user questions. For applications built around direct result retrieval, plan a separate search migration instead of assuming Microsoft’s agent grounding feature will preserve the old API behavior.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Sources: Microsoft Bing Search API retirement announcement; Microsoft Grounding with Bing product page; Grounding with Bing Search documentation.

Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.