Microsoft shipped Web IQ at its Build 2026 conference on June 2, putting a Bing-powered grounding layer directly in the hands of AI agents rather than human searchers. Microsoft puts the system at about 2.5 times the speed of the closest alternative. Both Microsoft Copilot and OpenAI’s ChatGPT already run on it.
Web IQ is not a renamed version of the Bing Search API that shipped with the first wave of LLM products. Jordi Ribas, who leads Search and AI at Microsoft, told Search Engine Land the stack was rebuilt from the ground up: how it indexes and retrieves documents, how it ranks and selects passages, and how it orchestrates calls all changed. The goal was to align every layer around what agents need at inference time, not what a person needs at query time.
That distinction matters for anyone who tracks how content gets cited. Ranking, Ribas said, matters less for agents than it does for people. What an agent needs is fast, clean passage extraction: pull the right text from a document, package it, and return it with minimal token overhead. The stated aim is to push fewer tokens into the model, return a stronger answer, and lower the cost of each call.
Agents also query differently. A human searcher runs one query and reads results. An agent fans out across multiple queries in a single task loop. Web IQ’s architecture accounts for that fan-out pattern by minimizing cost per call, making repeated high-frequency querying economically viable for the platforms using it.
Access is limited at launch. Microsoft and OpenAI are the confirmed production users. A broader rollout to Azure customers is planned, with no specific date disclosed. Organizations that want early access can register interest through Microsoft’s Web IQ site.
The announcement does not include independent benchmarking of the speed claim. The figures are Microsoft’s own measurements, and no third-party comparison has been published.
For SEO and GEO teams, the operational question shifts in the next 90 days. If the grounding layer that powers Copilot and ChatGPT answers optimizes for passage extraction rather than traditional ranking signals, then the content properties that drive agent citations are different from those that drive blue-link rankings. Clean heading structure, self-contained paragraphs, and extractable factual sentences become retrieval levers, not just readability improvements. Teams should audit their highest-traffic pages for passage extractability and begin tracking referral patterns from Copilot and ChatGPT separately from organic search, so they hold a baseline before Web IQ scales to more Azure customers.
Search Engine Land, reporting by Barry Schwartz, published June 2, 2026.
