Two new AI visibility reports landed. Neither one gives you a click.

For two years the honest answer to how much AI search is worth to you has been that nobody could measure it from a first-party source. That changed twice in ten days, and both changes are smaller than the announcements suggest. Search Console generative AI impressions and Microsoft Clarity's new citation split are real, free, and worth wiring up. They also both count how often, and neither counts how much.

Barry Schwartz reported on August 11 that Search Console's generative AI performance report had expanded to far more properties. John Mueller confirmed on Reddit that coverage is still not universal, noting that like the Discover and News reports, it only appears when there is sufficient data to display. Carolyn Shelby had laid out how to read the report at Search Engine Journal a week earlier. Separately, Microsoft shipped an update to Clarity on August 3 that separates branded from non-branded AI citations, covered by Slobodan Manic on August 10.

What Search Console generative AI impressions actually count

Google defines an impression here as an instance in which a link to a website is shown to a user within a generative AI feature. That is a narrow, specific thing, and the narrowness is what makes the report useful and limited at the same time. It is a count of link exposure, not of your brand being named, not of being the basis of an answer, and not of anyone doing anything about it.

Read that omitted list as a set. No queries means no intent segmentation. No clicks means no efficiency ratio. No placement means a mention buried in an expandable drawer counts exactly the same as being the first source named. Those three absences together mean the report cannot answer whether your AI visibility is improving in any sense a CFO would recognize. It can tell you whether it is happening more often.

The aggregation trap that will inflate your dashboard

This is the part that will quietly corrupt reporting across the industry over the next quarter, and it is worth understanding precisely. Property-level and page-level impressions are counted differently. At property level, when a single generative response surfaces several URLs from your site, that may register as one impression. At page level, each of those URLs may register its own.

The consequence: page-level impressions do not sum to the property total, and any dashboard that adds up per-page numbers to produce a site figure will overstate exposure, potentially by a lot on content-heavy sites where multiple pages get cited together. Nobody is doing this maliciously. It is the default behavior of every reporting tool that aggregates a page table upward.

The arithmetic in that illustration is deliberately simple: if every generative response that surfaces you shows two of your URLs, the page-level rows will total roughly double the property-level number, and half of what your dashboard reports is an artifact of the join. Pick one level of aggregation, state it on the chart, and never mix them in the same number. If you report both, label them as different metrics with different names, because that is what they are.

What Clarity adds that Search Console does not

Microsoft's update is the more conceptually interesting of the two, because it splits the metric on the axis that actually decides who owns the work. Clarity now labels grounding queries as branded or non-branded — branded meaning the query that triggered retrieval mentioned your brand, non-branded meaning it covered a broader category or topic — and reports citation counts for each.

That distinction is the one most third-party AI visibility tools blur, and blurring it is how attribution arguments start. Citations against branded queries largely reflect demand you already created; someone typed your name. Citations against non-branded category queries are the ones that represent new reach into a buying conversation you were not already part of. Reporting them as one number lets brand-strength gains masquerade as category-capture gains, which flatters everyone and informs nobody.

Clarity also introduces Share of Authority: the percentage of citations attributed to your domain compared with other cited domains, calculated daily, counting only query-days on which your domain received citations at all. That last clause is the one to read twice. Conditioning on days you were cited means the metric describes your standing among sources when you appear, and says nothing about how often you fail to appear. It is a quality-of-appearance measure wearing the costume of a share measure.

“A share metric that only counts the days you showed up is not measuring share. It is measuring how you did on your good days.”

Microsoft is upfront about the boundaries: the data is described as a representative view across supported AI experiences rather than complete coverage of all platforms, and citation counts indicate frequency rather than ranking position or prominence. Both caveats are honest and both should appear as footnotes on any slide that uses the number.

Working without the query dimension

The missing queries hurt more than the missing clicks, because intent is what turns a count into a decision. There are three workable substitutes, and each one recovers a different part of what the report withholds.

Frequency is not visibility, and neither is value

Zoom out and the two releases have the same shape. Both give you counts. Neither gives you position, passage, intent or outcome. Meanwhile the external research says the gap between exposure and value is enormous: Similarweb's data, reported by Greg Jarboe in late July, found only 6.8% of US ChatGPT answers included a link to an external source as of May 2026, that 65% of the URLs ChatGPT does cite sit two or three folders deep, and that 58.8% of the referral traffic AI sends back lands on the homepage.

Put those together and the measurement problem sharpens. Most answers cite nobody. When they do cite, the destination is usually a deep page. When a human follows through, they frequently arrive somewhere else entirely — the homepage — which severs the link between the cited URL and the session in your analytics. That is the mechanism behind the reporting gap we have written about before, and neither new report closes it.

There is a live example of why annotation matters as much as measurement. On August 13, Glenn Gabe observed that ChatGPT had moved visible source links into an overflow menu, so a user now has to open a three-dot menu to see sources. If that holds, referral sessions from ChatGPT will step down for reasons that have nothing to do with your content or your citations. A team without an annotation on that date will spend a month investigating a content problem that does not exist.

A reporting model that survives the gaps

Do this next

Check today whether the generative AI report has reached your properties, and if it has, export a baseline before you change anything. Set your aggregation level, write the definition into the dashboard itself, and add an annotation for August 13 covering the ChatGPT source-link change so the coming step down is already explained.

Then wire Clarity's branded and non-branded split into the same view, and start reporting category capture separately from brand reinforcement. That single change does more for the credibility of AI visibility reporting than any vendor score. Our reporting and analytics practice builds this structure first, and the background arguments sit in our pieces on the AI search visibility reporting gap, what the AI traffic conversion multiples actually show, and the opt-out decision that this same missing click data blocks. Search Engine Journal's walkthrough of the Clarity update has the field definitions in full.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.