Microsoft Web IQ frequently asked questions

Answers to common questions about using Microsoft Web IQ. For API details and request examples, see the API reference.

Service and deployment

What is Microsoft Web IQ?

Microsoft Web IQ is a suite of AI-native web search and grounding APIs that gives LLMs, AI agents, and applications access to current, relevant web information. This enables more accurate, up-to-date AI answers with sources users can verify. Web IQ powers leading frontier AI assistants, including Microsoft Copilot and ChatGPT.

Web IQ is available through REST at https://api.microsoft.ai/v3 and through MCP at https://api.microsoft.ai/v3/mcp. Each Web IQ AppID can call Web Search, News, Images, Videos, and Browse. Auto (beta) is enabled for an AppID on request: open a Web IQ Support request.

What is web grounding, and when does an AI agent need it?

Web grounding gives AI access to the latest web information so it can answer using current facts beyond its training data, with cited sources users can check. It helps most with frequently changing information, such as market developments, competitor activity, news, and regulatory updates.

Is Web IQ an LLM? Does it generate the final answer?

No. Web IQ is not an AI model or LLM. It's a suite of AI-native web search and grounding APIs that gives LLMs, AI agents, and applications access to current, relevant web information.

Web IQ doesn't generate the final AI answer. It retrieves relevant information from the web, and your chosen LLM, AI agent, or application logic uses that evidence to generate a response.

What makes Web IQ different from other web search APIs and grounding services?

Most web search APIs force a tradeoff: fast or accurate, cost-efficient or complete. In Microsoft's published comparison, Web IQ delivers the highest grounding satisfaction, the fastest speed, and the fewest tokens per query. It's the same platform that AI assistants like Copilot and ChatGPT rely on every day, proven at global scale and ready for any AI agent or app to build on.

Web IQ is also differentiated by how it accesses content. It takes a responsible approach to web access, bringing together the open web and structured data with enterprise-grade security and protections built in.

Web IQ works with the AI stack you already use. You can use your preferred AI model and connect through the REST API or Model Context Protocol (MCP), without being tied to a single model or platform.

Source: Web IQ benchmark results (Quality, Efficiency and Speed sections).

What can I build with Web IQ?

Web IQ powers AI agents and applications that need fresh, external information from the public web. Internal teams can use it for competitive intelligence, monitoring, regulatory research, sales intelligence, and content optimization. Customer-facing experiences can use it for research, discovery, recommendations, and specialized agents.

It's best suited for scenarios where information changes frequently, spans many external sources, and needs to be grounded in current, attributable web content.

Which Web IQ API fits my use case?

Use the Web Search API to find relevant public pages and passages for your model, the News API for current events, the Images and Videos APIs for visual and video results, and the Browse API when you already know the URL. Check the API reference for the capabilities and limits available to your account.

Can I use Web IQ with my existing AI model and agent framework?

Yes. Web IQ lets you add web information to the AI model and agent workflow you choose. REST APIs and MCP give you flexibility in how you connect it. Check the SDK and MCP integration documentation for supported configurations.

What does integration look like?

Your AI agent or application calls Web IQ for web evidence, then passes selected results to its model to generate a response. You control how the information is selected, used, and cited. Follow the Quickstart to connect your first workflow, or get started with the Playground.

Can Web IQ retrieve private company information?

No. Web IQ retrieves public-web information. For enterprise context, Microsoft offers Work IQ, Fabric IQ, and Foundry IQ, each serving a different role, from organizational context to business data and enterprise knowledge.

How does Web IQ relate to Foundry IQ, Work IQ, and Fabric IQ?

Microsoft IQ brings together four integrated layers that power enterprise intelligence across Microsoft. Web IQ provides information from the public web. Work IQ focuses on work signals, Fabric IQ on business data, and Foundry IQ on organizational knowledge. These products address different information needs for AI agents and applications and are most powerful when used together.

Is Web IQ an Azure resource?

You deploy a Web IQ resource in your Azure subscription. The resource handles metering and billing. Your application's identity in Web IQ is its Web IQ AppID, created when you connect the resource in the Web IQ portal.

The service itself is Microsoft-hosted and multi-tenant. Web retrieval and ranking run in Microsoft infrastructure, and your application calls api.microsoft.ai over HTTPS. The Azure resource doesn't host Web IQ and isn't in the request path.

What's the difference between a Web IQ resource and a Web IQ AppID?

A Web IQ resource is the Azure resource you create in your Azure subscription, in the Azure portal or with the Azure CLI. Web IQ usage is metered and billed to your subscription through it. The resource isn't in the request path: your application calls api.microsoft.ai directly.

A Web IQ AppID is your application's identity in Web IQ. It's created when you connect a Web IQ resource in the Web IQ portal. Everything about how your application uses Web IQ is set on the AppID: which APIs it can call, its rate limit (15 QPS by default), its API keys, and the Entra application (client) IDs bound to it. Don't confuse it with the Entra application (client) ID of your app registration: that ID identifies your application to Microsoft Entra ID when it gets a token, and you bind it to a Web IQ AppID.

Connecting a resource creates an AppID, so most customers start with one of each. If you need more than one AppID, for example one per environment or workload, your Microsoft contact can tell you the options.

Where is Web IQ deployed?

Web IQ is Microsoft-hosted and available through the public https://api.microsoft.ai/v3 endpoint. Your application calls the API, and Web IQ performs the web retrieval server-side.

Are there dedicated APIs for finance, sports, or weather?

Not today. There's no dedicated finance, sports, or weather API. For these and other subjects, use Web Search, which returns relevant crawled content. That content is unstructured page text rather than a structured feed.

Access and onboarding

How do I move from evaluation to production?

1. Complete your evaluation and confirm the service meets your requirements.
2. Create a Web IQ resource in a paid Azure subscription, then sign in to the Web IQ Customer Site and connect the resource. This creates your production AppID, which starts at the default limit of 15 queries per second (QPS).
3. To authenticate with Microsoft Entra ID, bind your app registration's Entra application (client) ID to that AppID in the Web IQ Customer Site.

Your use of Web IQ is governed by the Web IQ Terms of Use, in the context of your existing Microsoft Customer Agreement.

Which Azure subscriptions can I use?

Web IQ requires a paid Azure subscription. Azure credits don't work for Web IQ.

Which APIs can my application call?

Each Web IQ AppID can call Web Search, News, Images, Videos, and Browse. Auto is enabled for an AppID on request: open a Web IQ Support request. To see which APIs your AppID can call, sign in to the Web IQ portal, go to Profile Management, select your AppID by name in the drop-down at the top left, and check Allowed Services.

Which address should I use to sign in?

Use the Microsoft sign-in address that has access to Web IQ: the address Microsoft enabled for you, or the account you used to connect your Web IQ resource in the Web IQ Customer Site. This is normally the address you use for Azure, Teams, or Office, which may differ from your mailbox address.

To confirm it, sign in to Azure portal and check the address in the top-right corner.

Why does sign-in say that I do not have access?

First, confirm that the address shown in the top-right corner of Azure portal exactly matches the address that has access to Web IQ. Organizations often use one address for Microsoft sign-in and another for email.

If the addresses match, your organization may block cross-tenant guest access to the Microsoft Corporation tenant. Ask your IT administrator to allow that access.

Where can I get help?

Visit the Microsoft Web IQ support site for product assistance. For questions about permitted use, privacy, storage, or display, consult the current Web IQ Terms of Use and your applicable customer agreement.

API authentication

What authentication methods are supported?

Web IQ supports:

• Microsoft Entra ID bearer tokens through the OAuth 2.0 client credentials flow. This is recommended for production workloads.
• API keys sent in the x-apikey request header.

All API traffic uses HTTPS.

How do I configure server-to-server authentication?

Register an application in Microsoft Entra ID and use the OAuth 2.0 client credentials flow. Bind the app registration's Entra application (client) ID to your Web IQ AppID in the Web IQ Customer Site, acquire an access token, and send it as Authorization: Bearer <access_token>.

See Authentication for the complete setup steps.

Can I manage my own API keys?

Yes. You can create as many API keys as you need from the Web IQ Customer Site. There is no charge for creating keys.

Does Web IQ use Azure RBAC for API authorization?

No. Web IQ is a Microsoft-hosted service that you call over the internet. The Web IQ resource you deploy in your Azure subscription handles metering and billing, but it isn't in the request path, so there's no service resource for Azure RBAC roles to attach to.

Web IQ authorizes requests per AppID. Each AppID has its own content-eligibility profile and quota, and unauthorized calls return HTTP 403. To separate teams or workloads, for example with different access policies or quotas, use multiple AppIDs.

Can one Entra application (client) ID call more than one Web IQ AppID?

Yes. A single Entra application (client) ID can be authorized to call one or more AppIDs. To choose the AppID, pass it as a query parameter, for example https://api.microsoft.ai/v3/search/web?APPID=<your-web-iq-appid>. The API resolves the AppID like this:

• Your Entra application (client) ID is bound to one AppID: if you don't specify an AppID, or you specify that AppID, the request succeeds (HTTP 200). If you specify a different AppID, the request returns HTTP 403.
• Your Entra application (client) ID is bound to several AppIDs: you must specify one. A missing or blank AppID returns HTTP 400; one of your bound AppIDs succeeds (HTTP 200); any other AppID returns HTTP 403.
• Your Entra application (client) ID isn't bound to any AppID: the request returns HTTP 401.

API key authentication (x-apikey) is unaffected. Each API key is bound to a single AppID, so APPID is never needed or accepted for key-authenticated calls.

What authentication should a hosted agent use: Managed Identity or AgentID?

Either works. Managed Identity and AgentID are both Entra ID identities, and Web IQ accepts any Entra identity whose Entra application (client) ID is bound to your Web IQ AppID from the Profile Management page of the Web IQ Customer Site.

Which one you use depends on where the agent is hosted:

• Managed Identity fits agents that run on Azure infrastructure, such as Foundry, App Service, Container Apps, AKS, Functions, or VMs. System-assigned and user-assigned identities both work.
• AgentID fits Foundry or Copilot Studio-hosted agents that expose an agent identity.
• Agents hosted outside Azure, for example on your own Kubernetes cluster or another cloud, can use a standard Entra application registration with a client secret or federated credential.

In every case, the call uses the OAuth 2.0 client credentials flow and sends the token as Authorization: Bearer <access_token>. See Authentication for the setup steps. An API key in the x-apikey header remains the simpler option for prototypes.

Networking and security

Does Web IQ support Private Link, private endpoints, VNet integration, or IP allowlisting?

No. None of these are supported today. Web IQ is accessed through a single public endpoint, https://api.microsoft.ai/v3, over HTTPS with Entra ID or API key authentication. There's no IP-based access control at the API endpoint.

The Web IQ resource in your Azure subscription handles metering and billing but isn't in the request path, so Azure network controls such as Private Link, VNet integration, Azure Firewall, NSGs, and hub-and-spoke topologies have nothing on the Web IQ side to attach to. Identity (Entra ID) integration is Azure-aligned; the network layer is not.

On your side, allow the api.microsoft.ai hostname in your egress controls.

Does traffic from Azure to Web IQ stay on Microsoft's network or the Azure backbone?

This isn't guaranteed. Clients connect to the public https://api.microsoft.ai endpoint, and traffic is secured in transit with HTTPS.

Web IQ API traffic terminates at Azure Front Door, which supports TLS 1.2 and TLS 1.3 with these cipher suites:

• TLS 1.3: TLS_AES_256_GCM_SHA384, TLS_AES_128_GCM_SHA256
• TLS 1.2: TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384, TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256, TLS_ECDHE_RSA_WITH_AES_256_CBC_SHA384, TLS_ECDHE_RSA_WITH_AES_128_CBC_SHA256

Will Zscaler or another egress inspection tool interfere with Web IQ?

It should not interfere when configured to allow connections to api.microsoft.ai. This is a standard allow-list action, not a broad bypass. The same approach applies to other secure web gateways such as Netskope.

Can I put Web IQ behind Azure API Management?

Yes. You can place Azure API Management or another gateway in front of Web IQ in your own Azure subscription for observability and governance, and there are no known blocking limitations. API calls go from your caller to https://api.microsoft.ai/v3 over TLS with an API key or Entra ID token, so anything you put in front of that traffic is your own infrastructure in your tenant. Microsoft doesn't publish a reference architecture for these deployments; your cloud architecture team owns that design.

Under the Web IQ Terms of Use, response content can't be cached or stored to build a derivative dataset, regardless of what the gateway does. Operational telemetry such as status, latency, headers, and trace IDs may be retained, for example in Application Insights or Log Analytics.

What security controls are available?

• Authentication: Entra ID (OAuth 2.0 client credentials) or an API key.
• Authorization: per-AppID access to APIs. Unauthorized calls return HTTP 403.
• Rate limiting: per-AppID quota enforcement. Requests above the limit return HTTP 429.
• Transport: HTTPS only.

Network policy at your perimeter is your responsibility. If you have specific architecture requirements you'd like assessed, contact Web IQ Support.

Rate limits and API keys

Is there a rate limit?

Yes. Each Web IQ AppID has a default limit of 15 queries per second (QPS). A Web IQ resource usually has one AppID, created when you connect the resource. The Web IQ Customer Site displays the limit.

The limit is shared across all API keys and bound Entra application (client) IDs for that AppID. Requests above the limit return HTTP 429.

Can I increase my rate limit?

Yes. Each Web IQ AppID starts at 15 QPS. To request a higher limit, open a Web IQ Support request with your expected traffic and use case.

How many API keys can I create?

You can create as many keys as your application needs. Key creation is free, but all keys for one AppID share its rate limit.

Privacy and data handling

Does Web IQ log API queries?

Yes. API requests are logged for operational purposes: service reliability, debugging, and quality improvement. This is standard for Microsoft services.

• Access is restricted to authorized Microsoft engineering and operations personnel and is governed by Microsoft's internal access control policies. Data in logs isn't accessible to other customers or third parties.
• Logs are stored in Microsoft's production infrastructure and are subject to Microsoft's standard security controls, including role-based access control and audit trails.
• Logs are encrypted in transit (HTTPS) and at rest using Microsoft's standard encryption practices.

Microsoft operates as a data controller under the Microsoft Privacy Statement.

How long are queries retained?

Retention of query telemetry is governed by Section 4 of the Web IQ Terms of Use, which subjects Microsoft's processing to the Microsoft Privacy Statement. There is no separate Web IQ-specific retention commitment beyond what the Privacy Statement describes.

Can I turn off query logging for regulated or sensitive workloads?

Query text is captured as part of standard request telemetry, and its processing is governed by the Microsoft Privacy Statement and Section 4 of the Web IQ Terms of Use.

If you have a specific regulatory or compliance driver, Web IQ offers Zero Query Logging, which suppresses query text from telemetry. It isn't a self-serve feature; each use case is reviewed and approved individually, and it can apply to all your applications or only to specific ones. To request it, contact your Microsoft account team and share:

• The regulatory or compliance framework that applies (for example, GLBA, HIPAA, GDPR, or APPI).
• A short description of your deployment scenario.

What data residency commitments does Web IQ provide?

None today. Web IQ doesn't make region-specific data residency or in-region processing commitments, and cross-region routing to the nearest available endpoint is expected behavior. Privacy is covered by the Microsoft Privacy Statement.

Is Web IQ covered by the Azure Data Protection Addendum or EU Data Boundary?

No. Section 4.1 of the Web IQ Terms of Use excludes the service from the Microsoft Products and Services Data Protection Addendum.

Instead, Microsoft and the customer act as independent data controllers for any Personal Data processed under the Terms of Use (Section 4.4). Microsoft's processing is governed by the Microsoft Privacy Statement (Section 4.2), and you're responsible for giving appropriate notice to your end users.

For international transfers, the Terms of Use incorporate Standard Contractual Clauses where required (Section 4.6.1). That transfer mechanism travels through the Terms of Use, which sit under your Microsoft Customer Agreement, not through the Data Protection Addendum.

Web IQ does not make EU Data Boundary, region-specific data residency, or in-region processing commitments today.

Does Web IQ detect or mask personal data in queries?

No. Web IQ doesn't perform server-side detection or masking of personal data in inbound queries. If a workload may send personal or sensitive data, scrub it before calling Web IQ, for example with Azure AI Language PII Detection as a preprocessing step. That way nothing sensitive leaves your environment, and you keep full control over what reaches the API.

Does Web IQ support regulated industries?

Yes. Web IQ supports a range of enterprise and regulated use cases. If you have specialized compliance, governance, or industry requirements, contact Web IQ Support.

Content safety and coverage

What content safety controls are applied?

Web IQ applies crawl-time quality and safety classification, adult-content filtering, and per-customer content eligibility rules. SafeSearch is enforced server-side at strict by default. Web Search results also include an isAdult field for additional client-side handling.

Does Web IQ protect an LLM from prompt injection in retrieved content?

Microsoft Web IQ retrieves and ranks content from the web; it does not execute, interpret, or act on instructions embedded in retrieved content. Web IQ complements this retrieval behavior with Bing's existing quality, trust, spam, and malicious-content screening.

Protection against indirect prompt injection, jailbreak attempts, and instruction-following attacks is primarily the responsibility of the downstream model and orchestration layer. Treat prompts and retrieved content as untrusted inputs and apply layered controls such as Prompt Shields, Azure AI Content Safety, system-prompt isolation, least-privilege tool access, grounding validation, and output validation. Web IQ should not be treated as the sole enforcement point for prompt-injection mitigation.

How is web content processed before Web IQ returns it? Is it sanitized for prompt-like instructions?

Microsoft Web IQ does not operate directly on raw page HTML at retrieval time. During indexing, web content is normalized into a structured document representation designed to preserve meaningful content and document structure. Passages returned by Web IQ are selected from that representation, which is what enables more consistent extraction and retrieval across the API surface. Bing's existing quality, safety, spam, and malicious-content screening also applies as part of the indexing pipeline.

Web IQ can't guarantee that returned content is free of scripts or malware, and this normalization isn't a guarantee that prompt-like text is identified or removed. Treat retrieved content as untrusted input and apply the model- and orchestration-layer protections described in the previous answer.

Does Web IQ return hidden or non-user-facing page content?

Microsoft Web IQ attempts to return content that is visible through normal user interaction, including content revealed through ordinary page controls or client-side rendering. This is not guaranteed. Page structure and implementation vary, so applications should not assume that every returned passage was visible in the page's initial state or that every hidden element will always be excluded.

How does Microsoft maintain index quality?

Pages are evaluated for quality, authority, and trustworthiness. Malicious, spam, and phishing content is detected and excluded, and the index is continuously refreshed. It's the same index Microsoft's own AI products depend on.

How does Web IQ respect publisher content preferences?

Web IQ is designed to honor supported publisher controls, including applicable robots.txt directives and other publisher signals recognized by Microsoft.

Is there a published list of excluded sites or blocked categories?

No. Exclusion controls include robots.txt compliance, legal and compliance removals (such as DMCA notices and court orders), spam, malware, and phishing filtering, and adult-content filtering. The lists themselves are proprietary and continuously updated. If you have concerns about specific categories of content appearing, or not appearing, in results, contact Web IQ Support to discuss whether configuration of your Web IQ AppID addresses those requirements.

What content does Web IQ cover, and how fresh is it?

Web IQ continually retrieves fresh information across web pages, news, and image and video results to help your AI agent or application answer questions beyond its model's training data.

Availability and freshness vary by API, source type, and use case; there is no single freshness guarantee across all content. Consider the recency requirements of your workload when choosing sources and retrieval settings.

Which languages and markets does Web IQ support?

Web IQ draws from the global web, but coverage varies by language, market, topic, and source type. Review the Supported languages and regions page for the codes each API accepts, and assess coverage against the information your intended users need.

Can I configure domain allow or deny lists?

For individual domains or small lists, use standard query operators such as site:example.com to include a domain and -site:example.com to exclude one. Larger configurable lists are not currently exposed as a customer-managed feature.

Why am I not seeing LinkedIn results?

LinkedIn results are currently excluded.

What happens when Browse is called for an excluded URL?

Browse returns HTTP 403. The underlying exclusion list is not exposed.

How can I make my website discoverable to Web IQ?

Web IQ is built on the same web index Microsoft uses for its own AI products. Use standard web-discovery practices such as sitemaps, IndexNow, Bing Webmaster Tools, and an appropriate robots.txt policy. There is no separate Web IQ submission path.

Query behavior and result quality

Should I use natural-language or keyword queries?

Both work. Natural-language queries often work well with contentFormat=passage, while keyword queries are useful for precise lookups such as company names, tickers, or specific topics. Avoid overly long or compound queries; split multiple information needs into separate API calls.

What is the maximum query length?

The maximum is 1,000 characters, and longer queries are rejected. Shorter, focused queries generally produce better results: if a query contains a lot of context, retrieval may not focus on the right signal.

Can I filter Web Search results by date?

Web Search doesn't offer a date-range filter today, and we recommend avoiding a hard date filter unless you truly need one. Freshness is already factored into ranking based on the query, and adding a hard date constraint almost always costs relevance. For news scenarios, use the News API rather than filtering Web Search results by date.

lastUpdatedAt is the date the publisher last updated the content, not the date Web IQ last checked the page, so an older date doesn't mean the page wasn't checked recently. It's returned in the response, it isn't a request filter, and not every result includes it. You can use it to sort or filter results in your application.

Can Web Search handle a prompt that contains several questions?

Each Web Search request handles one query. If a prompt contains several questions or intents, split it into separate queries in your application, call Web Search for each, and combine the results before generating the answer.

Can Web IQ support multi-step agent workflows?

Yes. An agent can call Web IQ whenever it needs additional web evidence during a task. Retrieval latency is particularly important in workflows that make several calls before producing an answer.

How should my AI agent or application handle retrieved evidence?

Use evidence relevant to the question and track which sources support the answer. Keep retrieved content separate from instructions that control the agent. When sources conflict or evidence is incomplete, make that uncertainty clear rather than filling gaps with unsupported statements.

How does grounding quality compare?

Microsoft reports the highest grounding satisfaction in its published comparison, with a score of 79.05. Results depend on the query set, model, and evaluation method, so validate quality on your own scenarios.

Source: Web IQ benchmark results. 3K Global, Blind Queries sampled from prod, Config: 10 results, 10K chars per result (or equivalent).

Why do results differ from Bing.com?

Web IQ is optimized for LLM consumption, factual density, passage quality, and low latency. Bing.com is a consumer experience with additional ranking signals, personalization, visual answers, and content that may not be redistributable through an API. Some ranking differences are expected.

Can identical requests return different results?

Some variation is expected because the index is continuously refreshed and updates propagate across data centers. Results for identical requests made close together should remain broadly stable. If you see significant variance on identical queries within seconds, contact Web IQ Support.

Why does a query return fewer results than expected?

Check the raw Web IQ response before counting results. The most common cause is a client-side setting or downstream pipeline step that filters out valid results before they're measured. Also verify maxResults, content eligibility, SafeSearch behavior, and any site: or -site: operators in the query.

Content formats

What does each result return: passages or the full document?

Web Search and News Search return each result's content in the content field. With contentFormat set to passage, you get the parts of the document most relevant to your query. With text, html, or markdown, you get the full document: the structured version of the page that Web IQ builds during indexing. Passages work for every application; the full document formats need full document access. Images and Videos return result details such as titles, URLs, and thumbnails, not page content. See Passages and content formats.

What is passage, and why use it?

Set contentFormat=passage and the API returns, for each result, only the segments of the document that are most relevant to the query, not the full document. The full document is processed on the Web IQ side, and what comes back is a query-targeted extract, so your LLM doesn't have to skim past unrelated paragraphs to find the relevant sentence.

• Token savings: the LLM receives a fraction of the document instead of all of it.
• Better factuality: with less off-topic content in the context, the model is less likely to anchor on a tangential paragraph.

Set contentFormat=passage on every Web Search and News request. Browse also supports passage: send a query and Browse returns the parts of the page most relevant to your query.

Which format should I use?

Use passage for grounding. It's the recommended format and works for every application. Use text, html, or markdown when you need the whole document, for example to summarize or extract from a complete page. These formats need full document access.

Should I set contentFormat and maxLength on every request?

Yes. Set both on every Web Search and News request so you always get the format and length you expect.

How long can the content field be?

Up to maxLength characters. Without full document access, passages are at most 3,000 characters per result, even if maxLength is higher. With full document access, content can be up to maxLength, up to 500,000 characters. Full documents longer than maxLength are cut off.

Why do I get HTTP 403 (AuthContentFormatForbidden) when I set contentFormat to text, html, or markdown?

In Web Search and News Search, text, html, and markdown return the full document and need full document access. Without it, those formats return HTTP 403 with the error code AuthContentFormatForbidden. passage works for every application.

How do I get the full document as text, html, or markdown?

Open a Web IQ Support request. Applications that already use html or text keep working as today.

What does Browse return?

Browse returns content for a URL you choose, up to maxLength characters. With contentFormat set to passage and a query, it returns the parts of the page most relevant to the query. This is the default format unless your application has full document access. With full document access, it can also return the whole document as text, html, or markdown. See the Browse API reference.

Why do I get HTTP 400 from Browse when I don't send a query?

A passage request requires a query that isn't blank. passage is the default format unless your application has full document access, so include query in every Browse request, up to 1000 characters. See Passages and content formats.

How should I tune maxResults and maxLength to manage context window and token costs?

Web IQ can prioritize passages relevant to a query instead of supplying unnecessary page context. This can reduce the amount of text passed to your model and lower input-token costs. Benefits depend on your model, prompts, query mix, and retrieval settings.

For most grounding scenarios, more results with shorter passages give better output than fewer results with longer passages for roughly the same total token budget. A good starting point is maxResults=10 with maxLength around 3,000 characters in passage format. Benchmark with your own queries and model.

Why don't Web Search and News Search return a snippet field?

This is by design. Both return content as a query-dependent extract (passage) instead of a separate, query-independent snippet.

Response content

What is the maximum maxResults value for Web Search?

Web Search supports up to 50 results per request.

Which request parameters can I configure?

Web Search exposes maxResults, language and region, contentFormat (passage, text, html, or markdown), maxLength, and safeSearch. See the API reference for the full parameter list.

What is the difference between lastUpdatedAt and crawledAt?

lastUpdatedAt is the publisher-reported content update date, returned by Web Search, News, Videos, Images, Browse, and Auto when available. crawledAt is the time Web IQ last fetched the page, returned by Web Search, Browse, News, and Auto. Coverage of both fields varies by result.

What is Browse Live Crawl and when should I use it?

Browse returns content from the Bing web index, which is continuously refreshed as our crawlers observe changes across the web. This is how web search works in general: search results are served from an index rather than by fetching every URL live on every request, and it is what gives Browse its speed and consistency.

Browse supports Live Crawl for URLs that aren't available in the index. Set liveCrawl: "fallback" on the request to request a live fetch when the indexed copy is missing. Live crawling is subject to publisher politeness restrictions and availability; fallback doesn't guarantee an actual crawl. If indexed content is available, Browse returns it. Omitting the parameter (or setting liveCrawl: "none") returns indexed content only, without live crawling. See the Browse API reference for full parameter details.

Live Crawl is available by default on Web IQ Browse and is provisioned with a modest per-second capacity ceiling, separate from your overall Browse QPS. This ceiling is shared across all API keys under the AppID. Most Browse workloads do not need Live Crawl and are best served by the standard indexed path; use Live Crawl when a URL isn't available in the index.

Can Browse retrieve JavaScript-rendered pages?

Yes. Use renderDynamicPages: true together with liveCrawl: "fallback". See the Browse API reference for the full parameter set and response details. Test representative URLs before you depend on this behavior in production.

Can Browse extract online PDF files?

PDFs hosted online are not a documented Browse scenario today. If you have a specific PDF use case, contact Web IQ Support with a representative URL and the behavior you expect, and you'll get a definitive answer.

Billing and legal

What does Web IQ cost?

Web IQ uses consumption-based, per-call pricing. Pricing depends on the APIs enabled for your account, your usage, and your commercial terms. See Pricing for current rates.

Does Web IQ include an SLA, RTO, or RPO commitment?

No. Microsoft Web IQ does not include a service level agreement. The Web IQ Terms of Use do not commit to uptime, latency, Recovery Time Objective (RTO), or Recovery Point Objective (RPO) targets, and SLA terms are not added for individual customers.

Do I need to display citations?

Yes. Exhibit A, Section 1 (LLM Use) of the Web IQ Terms of Use requires source attribution and a link back to each source used for the displayed LLM content, near that content.

• Cite only the sources actually used to generate the answer. If Web IQ returned 10 results and your model used 3, display those 3.
• Citations must stay visible with the answer and be reachable with a single action. Don't show them only briefly, hide them behind multiple screens, or require users to ask for sources separately.
• For constrained interfaces such as cars, wearables, voice-first devices, or kiosks, use a persistent QR code that resolves to a page listing every citation used for the current answer, a companion app that shows the answer with clickable citations, or an on-screen list the user can open directly.

You may store citations as part of an individual end user's chat history for up to 30 days. See the caching question below.

How much source content can my AI agent or application display?

Exhibit A of the Web IQ Terms of Use limits displayed Grounding Content from any single source to the lesser of 200 characters or the maximum specified by the website owner or webmaster. For voice or audio experiences, apply the same limit to spoken output. Additional restrictions apply to summaries and display. Review the Terms of Use before designing excerpts; this limit is not permission to reproduce entire source pages.

Can image and video results be used to ground an LLM?

Image and video metadata and alt text are included in the definition of Grounding Content. However, the Web IQ Terms of Use prohibit using actual images or videos from results for grounding, including them in LLM output, or creating derivative images or videos from them. Review the terms for your intended implementation.

Can I cache or store Web IQ content?

As a general rule, no. The Web IQ Terms of Use prohibit copying, storing, caching, archiving, or creating a database of Content, except under narrow exceptions (Section 3.2.15 and Section 4 of Exhibits A and B):

• End-user chat history (up to 30 days): Content may be stored on a non-aggregate, per-end-user basis for an individual end user's chat history for up to 30 days. For LLM Use (Exhibit A) this covers Content in general; for Search Use (Exhibit B) it covers image and video URLs.
• Atomic Action transient processing: Content may be held in a transient processing state only as strictly necessary to complete a single logical request or workflow initiated by an LLM (an "Atomic Action", which may involve multiple internal steps or agentic interactions). Content must be deleted promptly when the Atomic Action completes.
• Your own work product: to the extent permitted by copyright law, Content may be copied or stored as an integrated part of your own work product, as long as the work product isn't created to substitute for the Services (for example, building a database of Content, indexing links for crawling, or offering a separate search).

Outside these exceptions, the Terms of Use don't allow derivative databases, persistent caching, or long-term archival of Content.

Where can I read the Web IQ Terms of Use?

The current terms are published at Microsoft Web IQ Terms of Use. They apply together with your existing Microsoft Customer Agreement.