AI & TechArtificial IntelligenceBigTech CompaniesDigital MarketingDigital PublishingNewswireTechnology

Microsoft Bing Grounding APIs Now Power AI Agents

▼ Summary

– Microsoft announced Web IQ, a set of grounding APIs that allow AI agents to pull information from Bing’s search index, returning passages instead of full web pages.
– Web IQ uses a rebuilt retrieval stack based on the Bing index, designed for AI agents that search repeatedly under tight time constraints across multiple steps.
– Microsoft claims Web IQ scores higher than competitors on grounding satisfaction (GDSAT) based on 3,000 sample queries and reports sub-165ms response times at P95.
– Web IQ follows the same robots exclusion rules and publisher preferences that Bing already honors, and Microsoft is working on industry standards for AI web access.
– Microsoft has not announced general availability, pricing, or which AI platforms will use Web IQ, nor whether existing Copilot or Bing Chat grounding uses it.

Microsoft has officially unveiled Web IQ, a new set of grounding APIs designed to let AI agents tap directly into Bing’s search index for real-time, factual information. Dubbed “a search engine for AI systems,” Web IQ shifts the paradigm from helping humans find web pages to helping machines extract usable data while reasoning through complex tasks.

At its core, Web IQ reimagines how content is indexed, ranked, and selected. Instead of returning full web pages, the API delivers passages and structured evidence objects,essentially stripping away extraneous content so AI models only process the most relevant parts. This is critical because every token an AI model consumes adds both cost and latency. As Microsoft puts it, the goal is “fewer tokens in, better answers out, lower cost per call.”

The performance numbers are striking. Microsoft claims sub-165 millisecond response times at the 95th percentile, nearly 2.5 times faster than competing solutions, based on tests across five data centers. They also report that Web IQ scores higher on grounding satisfaction (GDSAT),a metric measuring freshness and trustworthiness,across 3,000 sample queries. On token efficiency, the system delivers the same quality with fewer tokens as the volume of search results grows.

Publisher controls remain a priority. Web IQ adheres to the same robots exclusion rules and publisher preferences that Bing already honors. Microsoft is also collaborating with the IETF and other industry groups to develop standards for how AI systems access web content.

Technically, Web IQ leverages Microsoft’s open-sourced embedding model to find relevant content, then uses additional models to rank and select passages. Importantly, these models are trained for how they’ll be used inside AI reasoning loops, not for standalone benchmark scores. For fast, large-scale search, the system extends DiskANN, Microsoft’s technology for searching massive indexes without loading everything into memory.

This launch follows a clear trajectory. In February, Bing Webmaster Tools added AI citation data; in March, it mapped grounding queries to cited pages; and at SEO Week, Microsoft previewed Citation Share. Those tools show publishers how AI systems use their content, and Web IQ is the engine that pulls that content in the first place. The shift from ranking full pages to ranking useful passages means that what makes a page rank well in traditional search may not align with what makes a passage valuable for grounding,a nuance Microsoft’s own grounding framework post explored earlier this year.

Looking ahead, Web IQ is accepting expressions of interest but has not announced general availability, pricing, or which AI platforms will use it. Microsoft has not clarified whether existing Copilot or Bing Chat grounding relies on Web IQ, or whether it operates as a separate system.

(Source: Search Engine Journal)

Topics

web iq api 98% ai grounding 95% token efficiency 92% search engine for ai 90% performance metrics 88% publisher controls 86% retrieval stack 85% diskann technology 83% embedding models 82% structured evidence objects 81%