TL;DR: Before an AI assistant answers you, it runs its own web searches and reads what comes back. Microsoft’s free analytics tool, Clarity, has been showing site owners which of those hidden searches ended with their page being quoted. On August 3 it added a split: searches that said your company’s name, and searches that didn’t. That distinction matters, because getting quoted when someone already typed your name is a much smaller achievement than getting quoted on a generic question. One catch runs through the whole thing — the citation numbers come from Microsoft’s own assistants, Copilot and Bing’s generative answers. ChatGPT, Gemini, Claude and Perplexity are not in there.
First, what is a “grounding query”?
You ask Copilot: which travel insurance covers off-piste skiing?
Copilot does not answer from memory. It first converts your question into a handful of ordinary search phrases — “off-piste ski insurance”, “winter sports policy exclusions”, things like that — runs them, reads the results, and writes an answer out of what it found. Those little machine-written searches are called grounding queries. Grounding, as in anchoring the answer to something real on the web.
You never see them. Neither does the person asking. They don’t show up in Google Analytics either, because the assistant did the browsing, not a human.
Clarity’s AI Citations dashboard is unusual because it exposes them. Instead of guessing why your page appeared in an answer, you get the actual phrase Copilot searched for before it quoted you. Search Engine Journal’s walkthrough of the feature describes one site with over 36,000 recorded citations traced back to 147 distinct grounding queries.
What shipped on August 3
Four changes, all inside the existing AI Visibility dashboard, per Microsoft’s own post:
- Individual queries now carry a branded label.
- Share of Authority, Clarity’s measure of how strongly you feature in the answers, is broken out branded versus non-branded.
- Dashboard filters apply the split across the rest of the reporting.
- Brand-led demand is separated from generic discovery.
It’s free, and it’s the third Clarity release in under a month. PPC Land counted them: Topic Insights on 9 July, Query Topics in beta on 22 July, this on 3 August. Citations itself only reached general availability on 13 May. Microsoft is clearly trying to make a firehose of citation data readable.
Why the split is the interesting part
Two citations can look identical in a report and mean opposite things.
If Copilot searched “Preferium pricing” and quoted our pricing page, that is not discovery. Somebody — or some model — already had the name. If it searched “white-label AI SEO platform” and quoted us, we won a question where we were one candidate among many. The first number tells you your brand is legible. The second tells you your content competes.
Lumped together, the second gets flattered by the first. In the sample PPC Land published, 43% of citations came from branded queries, and the two highest Share of Authority rows were both branded — 24.2% and 22.3%, against 20.3%, 18.6% and 15.9% for the rest. A single blended “authority” figure in that dataset would be carried by the brand’s own name.
This also lines up uncomfortably well with Monday’s research on model memory, which found assistants disproportionately search for brands they already recognise. If the model is biased toward looking you up by name, a branded-heavy citation profile is closer to a description of the model’s habits than a report card on your content.
The limit you have to write on the report
Clarity’s Citations data reflects Microsoft’s AI surfaces. OpenAI, Google, Anthropic and Perplexity don’t hand Microsoft their internal grounding logs, so nothing in that dashboard tells you how ChatGPT cited you.
Search Engine Journal’s case study makes the skew concrete: of those 147 grounding queries, Bing ranked the site for all but six. Google ranked it for none of them. The queries themselves are Bing-shaped.
There is a separate part of Clarity that does name ChatGPT, Copilot, Gemini, Claude and Perplexity — the AI referral traffic channel. That’s people clicking a link in an answer and landing on your site. Useful, and a different thing from being cited: one counts visits, the other counts appearances in an answer the reader may never click out of. Don’t let the two blur together in a client deck.
What to actually do with it this week
Switch it on and take a baseline. It costs nothing, and a dashboard you enable in September will not tell you what August looked like.
Then read the non-branded grounding queries as a content brief. They are the phrases a machine wrote when it went looking for your topic, which is a different vocabulary from the keywords people type — and it is the vocabulary that decides whether you get retrieved.
When you report it, label the surface. “Copilot and Bing citations” is honest. “AI visibility” is not, when four of the five engines your client’s customers use aren’t in the dataset.
Keeping that honest across a portfolio is the tedious part: every site needs its own baseline, its own branded filter, and a monthly re-read as Microsoft ships again. Preferium covers the half Clarity can’t see — citation tracking runs real prompts weekly against ChatGPT, Claude, Gemini and Perplexity with web search on, and records who gets cited and which URLs the engines actually pulled from. On the technical side, 47 automated checks crawl every page and score the site 0–1000, then the system fixes, deploys and re-verifies the live page in a real browser without a dev queue. More on how the system works.
Key takeaways
- Clarity’s AI Citations dashboard now labels each grounding query branded or non-branded, and splits Share of Authority the same way — free, live since 3 August.
- A grounding query is the search an assistant writes for itself before answering. Non-branded ones are where you actually competed.
- In PPC Land’s sample, 43% of citations were branded and branded rows held the top two authority scores — a blended number would overstate discovery.
- The citation data is Microsoft-only: Copilot and Bing generative search. Clarity’s referral-traffic report names other engines, but referrals are clicks, not citations.
- Say which surface a number came from before you put it in front of a client.
