You have just enabled the “Use information from the web” setting in your Copilot Studio agent. It feels like a simple toggle, a way to let your AI agent access the open internet. Yet a specific question lingers: is this system hitting the same infrastructure as the consumer search engine you use every day? Or does it operate on a separate, private crawler?
This distinction is not a minor technicality. It defines exactly what your agent can and cannot see. To answer that, we need to look past the user interface and examine the underlying architecture. Specifically, we must understand how the Bing index functions within the agent’s retrieval pipeline. Is it a real-time snapshot of the web, or a query into a pre-existing dataset? And how does that compare to the scoped, verified data you might add through specific AI search sources? The answer changes how you configure your knowledge base and ultimately how your Copilot web answers are grounded.
Grounding with Bing Search: how Copilot retrieves web answers
The “Use information from the web” setting is the specific mechanism that allows a Copilot agent to pull data from the open internet. When enabled, this feature uses Grounding with Bing Search to answer queries by searching all public websites that have been indexed by Bing. This means the agent does not have its own independent crawler; instead, it relies entirely on the existing infrastructure of the Bing index to find relevant snippets.
Querying the Existing Index
It is a common misconception that this feature performs a real-time crawl of the internet. It does not. The system queries the existing Bing index, which implies that for any content to be retrieved, it must already be indexed. If a page is new, blocked, or excluded from Bing’s index, the Copilot agent will not see it. This dependency on the pre-existing index means that the freshness and scope of Copilot web answers are directly tied to how thoroughly Bing has cataloged the web.
Orchestration Requirements
A critical technical prerequisite is that this setting requires generative orchestration to be enabled on the agent. You cannot use web grounding in classic orchestration mode. This distinction matters because generative orchestration changes how the agent processes information, allowing it to filter and interleave sources more dynamically. If your agent is currently running on classic orchestration, you will need to switch modes to access these AI search sources. This architectural choice defines the foundation of the system, separating simple, rule-based responses from grounded, web-aware generation.
Web Search vs. public website knowledge sources
The distinction between the two retrieval methods comes down to scope and control. When you enable the “Use information from the web” setting, the agent accesses the global Bing index. This means the search queries span every public website that Bing has indexed. It is a broad, unbounded search designed to capture general knowledge, news, and external facts. In contrast, a configured public website knowledge source is scoped. You explicitly list the domains the agent is allowed to query. The system restricts results to those specific URLs, ensuring the agent only pulls from sources you have vetted and approved.
A key architectural detail is how these two sources interact when a user asks a question. The system does not run them sequentially. Instead, it executes both the broad web search and the scoped domain searches in parallel. This concurrent execution ensures that the agent gathers all relevant data simultaneously, rather than waiting for one source to finish before querying the other. This parallel approach is critical for maintaining low latency in Copilot web answers, as it prevents a bottleneck where the faster source would sit idle while the slower one completes.
Once the searches complete, the results are not kept separate. The architecture interleaves the data from the global index with the results from your specific domains. The LLM receives a blended context, mixing broad internet facts with your controlled corporate data. This blending allows the agent to answer questions that require both general knowledge and specific internal or brand-related details. For example, if you ask a question about a new industry regulation, the agent can combine the general summary from the Bing index with specific implementation details from your compliance website. This strategy ensures the final answer is well-rounded, balancing external context with the precise information you deemed authoritative. However, this blending also means you must trust that the external data retrieved from the global index is accurate, as the system does not filter out ungrounded web results by default unless specific moderation settings are applied.
Why this architecture matters for AI search sources
The dependency on the Bing index creates a direct link between traditional search engine optimization and AI agent capabilities. For businesses, this means that presence in the Bing index is not optional if you want your content retrieved by the “Use information from the web” setting. If your site is not indexed by Bing, your agent will not find it, regardless of your brand’s visibility on other platforms. This makes indexing status a prerequisite for AI-driven discovery, not just a side effect of SEO efforts.
The trade-off of scope versus control
Choosing between a broad AI search source and a scoped internal one involves balancing coverage with accuracy. A broad source accesses the entire public web, offering wider reach but less control over data quality. In contrast, a scoped source limits results to verified, specific domains, ensuring reliability but reducing breadth. This distinction is critical for teams managing customer-facing agents, where unverified web data can introduce inconsistencies into Copilot web answers.
A nested dependency on Bing infrastructure
Public website knowledge sources also rely on Bing to process query inputs, even when restricted to specific websites. This creates a nested dependency on Bing’s infrastructure. Whether you use the global web search or a scoped source, the underlying mechanism remains the same: Bing executes the search. This architectural consistency means that optimizing for Bing’s architecture benefits both types of sources, making it a central focus for any strategy aiming to enhance AI visibility.
Common questions about Copilot Bing index dependencies
Managers often ask whether the underlying search infrastructure for Copilot agents mirrors their existing consumer search strategies. Understanding these dependencies helps clarify which content is visible to AI-driven responses and how retrieval prioritization works in practice.
Does Copilot use a different index than Google?
No. Copilot’s web search relies entirely on the Bing index. If your content is indexed by Bing, it becomes a candidate for retrieval by the agent. If it is not, the content remains invisible to this specific feature, regardless of its ranking in other search engines. This distinction is critical for AI search sources strategy, as visibility in one ecosystem does not guarantee visibility in another.
Can I force Copilot to use a specific URL?
You can add specific public websites as knowledge sources to restrict and prioritize search results to those domains. However, the underlying search mechanism still leverages Bing’s infrastructure to find and retrieve the content. This means you are not bypassing the index; you are simply narrowing the scope of what the model can access from within the Bing index.
Is the data from the Bing index real-time?
The data is as current as the Bing index itself. There is no separate, independent real-time crawler for this feature. It mirrors the state of the public Bing search index, meaning if your site has been recently updated but not yet re-crawled by Bing, the agent will not reflect those changes. For teams managing Copilot web answers, it is essential to account for the inherent latency of any major search engine’s indexing cycle.
The dependency on the Bing index changes how we should view web visibility in the AI era. When an agent retrieves information via the “Use information from the web” setting, it is not browsing the live internet in real-time; it is querying a specific, pre-built database. This technical constraint means that content existing outside that index is effectively invisible to these AI-driven systems, regardless of how well it performs in other search environments.
For decision-makers, this shifts the conversation from a narrow SEO tactic to a strategic prerequisite. If you build agents or rely on AI-generated answers to represent your brand, your presence in the Bing architecture is no longer optional—it is the entry ticket. As these AI search sources become standard in customer-facing workflows, understanding the underlying data dependencies will be just as critical as understanding the user experience itself.
