Microsoft launched Web IQ at Build 2026, a set of grounding APIs that replace the now-retired Bing Search API with search infrastructure built specifically for AI agent consumption. The system sits on Bing’s global index but redesigns everything above it for machines that read differently than people, according to The AI Economy.

“We needed to be able to cleanly break from the convention and decisions we made to optimize for a human, to make it work excellently for AI systems,” Tim Frank, Microsoft’s corporate vice president for monetization, commerce, and the AI economy, told The AI Economy. “The name is intentionally separate so people don’t confuse the stuff designed for humans versus what is explicitly designed for AI.”

How It Differs From Bing

The core design change: Web IQ returns passages and structured evidence objects rather than ranked links. Traditional search optimizes for ten results on a screen. Agents have no viewport.

“They don’t have a limit of ten because they don’t have a viewport on the screen,” Frank explained to The AI Economy. “For the agent, you actually could give many more links because it’s not very expensive for the agent to consider 15 or 1,500 items. The cost is in how many tokens you give it.”

Microsoft’s official announcement describes the system as purpose-built for multi-step agentic workflows where agents “retrieve repeatedly, reason over evidence, adapt to new information, and operate inside tight latency budgets.” The company rebuilt the stack from indexing and retrieval through ranking, passage selection, and orchestration to align every layer around inference-time grounding.

Under the hood, Web IQ runs a small set of in-house models trained not for standalone benchmark scores but for how their outputs perform inside another model’s reasoning. One model converts web content into embeddings for semantic matching. Others read, rank, and select passages worth handing to an agent. The result: a set of Microsoft models decides what evidence other companies’ models get to reason over.

Publisher Licensing

The content licensing model is inference-only. Publisher content is fetched when an agent needs it, used, and traced back to its source. Nothing gets absorbed into model weights. “None of these licenses are for model training,” Frank told The AI Economy. Web IQ inherits Bing’s commitment to robots.txt, publisher controls, and access preferences, according to Microsoft’s blog post.

Microsoft says it is working through IETF and industry forums to help develop interoperable standards for agent-era content access.

Performance Claims

Microsoft claims Web IQ leads competitors on grounding satisfaction (completeness, freshness, and authority), returns responses in under a sixth of a second (2.5x faster than peers), and reduces token consumption through passage-level evidence. The company did not name the competitors it benchmarked against, and the tests were run internally, as The AI Economy noted.

Enterprise Positioning

Web IQ joins Microsoft’s IQ product family alongside Work IQ (workplace activity data), Fabric IQ (infrastructure inventory), and Foundry IQ (development context). Frank used a Windows 11 enterprise upgrade scenario to illustrate the stacking: Web IQ provides market and support information, Fabric IQ shows deployed hardware inventory, and Work IQ reveals usage patterns, according to The AI Economy.

The service is currently available to a limited group of enterprise customers developing production AI workloads that require real-world grounding, with priority given to those working with Microsoft account teams.

The Infrastructure Control Point

The strategic position is worth noting. Any company building agents that need real-time web information faces a narrow set of options: Google, Microsoft, and a handful of smaller index operators. Web IQ makes Microsoft’s pitch explicit: use their infrastructure for the evidence layer, and their models curate what your models get to see. For agent builders who need web grounding but lack their own index, that trade-off may be straightforward. For those concerned about a single vendor controlling the evidence pipeline for AI reasoning, it raises questions about concentration that the industry has not yet resolved.