Grounding -- the process of anchoring a model's responses in real, external data rather than memorized training weights -- has quietly become the most important unsolved infrastructure problem in production AI. Microsoft just made its biggest bet on solving it. At Build 2026, the company unveiled Web IQ, a suite of AI-native APIs that connects agents to live web intelligence, and it's already the engine running underneath both Microsoft Copilot and ChatGPT's web search.
Microsoft calls Web IQ "a search engine for AI systems." Where Bing helps people find web pages, Web IQ helps AI agents find information they can use while reasoning through a task. That distinction is more important than it sounds. Traditional search returns a ranked list of links. Web IQ returns something fundamentally different.
Passages, not pages
Web IQ does not return documents -- it returns passages and structured evidence objects. Models do not need documents; they need information, and documents are often a poor proxy for that. By operating at the level of passages, the system concentrates useful signal while eliminating irrelevant context, producing a much higher ratio of information to tokens.
The practical upshot of this design is captured in a single principle Microsoft repeats throughout its documentation: fewer tokens in, better answers out, lower cost per call. For agents that call a grounding API dozens of times per workflow, that efficiency compounds fast.
The service returns three key components for each query: a ranked result list, brief evidence snippets that directly answer the query intent, and provenance citations that include source URLs and titles. That structure means the output is ready to drop directly into an LLM's context window with no post-processing required.
A ground-up re-architecture
Microsoft announced Web IQ at Build 2026 as a set of AI-native grounding APIs that connect enterprise agents to fresh web information, positioning the service as part of its broader Microsoft IQ layer across Copilot, Foundry, and Copilot Studio. But the engineering story goes deeper than a product announcement.
Don't miss what's next in AI
Join 300,000+ engineers and researchers who get the signal, not the noise.
- Full access to in-depth AI research breakdowns
- Be the first to know what's trending before it hits mainstream
- Daily curated papers, repos, and industry moves