Microsoft's Web IQ matches 88% of Copilot citations when queried with fanout, Rudzki finds

Microsoft has made Web IQ publicly available, and the company says it is the retrieval system behind Copilot's and ChatGPT's web grounding. Tomek Rudzki tested that claim and posted results on LinkedIn on 6 November.

Across 500 prompts and more than 1,000 fanout queries, 88% of the sources Copilot cited appeared in Web IQ's top 10 results. 78% sat in the top 5. Rudzki calls the sample small but says the pattern is clear.

The method matters. Coverage only reaches 88% when Web IQ is queried with Copilot's own search query fanout, meaning the expanded set of queries the assistant generates from a user prompt. Query Web IQ with the raw user prompt instead, and coverage falls to 25%.

For ChatGPT, 20% of sources appeared in Web IQ's top 20 on the same day. Rudzki attributes the gap to ChatGPT running its own index and blending Google, Web IQ and its own scraping. Web IQ tracks Copilot closely. It does not track ChatGPT closely.

What Rudzki says it is useful for

Two checks. Whether your content is in Bing's index at all, and whether it is a candidate for Copilot's answers. He describes it as the cheapest way to see what Bing's retrieval thinks of a page for a given query.

What we think

The fanout dependency is the part worth sitting with before anyone treats Web IQ as a Copilot diagnostic. A citation audit that uses raw prompts will understate Copilot coverage by roughly three quarters on these numbers, which would send teams chasing a retrieval problem they do not have. Anyone running this should reconstruct the fanout queries first, verify the pattern on their own URLs, and treat Web IQ results as one input into a view of Copilot presence rather than the view itself. The ChatGPT number is the other half of the lesson: a single retrieval API will not stand in for assistants that build their own indexes.