What Is a Grounding Query? Bing's New Unit of AI Visibility, Explained
- Patrick Duggan
- 10 minutes ago
- 3 min read
A grounding query is the reformulated search question an AI assistant writes to itself — automatically and invisibly — when it decides it needs to go read the live web before answering a user. When you ask Microsoft Copilot a messy, conversational question, it does not paste your exact words into a search box. It rewrites your intent into one or more cleaner, machine-optimized queries, runs those against Bing's index, reads the results, and uses what it finds to ground its answer in current facts. Each of those rewritten queries is a grounding query.
The term started appearing in front of publishers in February 2026, when Bing Webmaster Tools shipped its new AI Performance report. For the first time, site owners could see not just whether Copilot cited their pages, but the actual reformulated queries Copilot generated on its way to citing them. That is a genuinely new window. Until grounding queries were exposed, the entire process by which an AI assistant decided what to read was a black box.
Why grounding queries are different from search keywords
A normal search keyword is what a human types. A grounding query is what a machine writes after interpreting what a human meant. The two are often very different. A person might ask Copilot something rambling and personal; the model distills that into a crisp, almost librarian-like query — and that distilled query is what your content has to match to get pulled into the answer.
This flips a piece of SEO logic on its head. You are no longer only optimizing for how humans phrase things. You are optimizing for how a language model rephrases human intent into a retrieval query. Content written in clean, declarative, answer-shaped language — definitions, direct statements of fact, structured lists — tends to match grounding queries far better than clever marketing copy does, because the model's rewritten query is itself clean and declarative.
What grounding queries tell you that citations alone do not
A citation count tells you the model used your page. The grounding queries tell you why — what intent the model was trying to satisfy when it reached for you. If you see grounding queries you are winning, that is a demand signal: keep feeding that topic. If you see grounding queries on topics you cover but are not being cited for, that is the most actionable gap in modern content strategy. The model is asking the question. You have the answer. You are just not legible enough to be chosen.
How this fits the bigger picture
Grounding queries are one piece of the retrieval-augmented generation, or RAG, loop that now sits underneath most AI assistants: interpret the user, write a grounding query, retrieve, read, answer with citations. Bing exposing this loop to publishers is part of a larger shift we track closely — the machines that are quietly replacing search are starting to show their work.
Our take, and we will say it plainly because hedging helps no one: if your Bing AI Performance report shows grounding queries firing on your topics but few or zero citations, you do not have a content problem. You have a legibility problem. The answer is structured data and answer-shaped writing, not more volume. We pulled our own report and read it out loud — Copilot has cited our main site eleven times total — precisely so we could prove we use the same instrument we hand to customers. Every score we publish is capped at 95, because we guarantee five percent of any confident number is wrong, including this one.
How do AI models see YOUR brand?
AIPM has audited 250+ domains. 15 seconds. Free while still in beta.