AI Citations Run on a Separate Clock: What Google's Generative AI Data Revealed

💡 Quick Summary (TL;DR):

• The setup: In my Bing Citation Share piece I floated a mechanism: AI search fetches pages live at query time, so a page you remove (now a 404) drops out of AI answers almost instantly.
• The new data: Days later, Google added a Generative AI report to Search Console. For the first time I had first-party AI-visibility data from the second major engine, not just Bing.
• What both engines agree on: Remove a page and its AI citations fall away roughly two weeks before it leaves the search index. The direction held in both.
• What broke my theory: A page I removed back in March kept showing up in Google's AI answers until mid-May. A real, at-query-time fetch could never produce that. AI isn't reading live. It runs on its own cached index that refreshes ahead of search but is not real-time.
• What actually differs: Speed. Bing turned the whole pipeline in a couple of weeks; Google took one to two months to react at all.
• The honest caveat: Two different products, a single site on the Google side, and a feature that isn't widely rolled out yet. This is a report, not a law.

A week ago I published what I had found in Bing's new Citation Share data, and I ended it with a small mechanical theory I was fairly confident about. Then Google shipped its own version of the same feature, I went to check my theory against it, and the theory did not survive contact with the second dataset.

This is the follow-up I'd want to read if I were the one who had taken my earlier explanation at face value.

A Second First-Party Source Shows Up

In the Bing piece, my one real complaint about the whole "AI Visibility" industry was that it runs on synthetic data: vendor-chosen queries, not real users. Bing's Citation Share was the first honest, first-party number against that. The obvious limitation was that it was one engine. Bing Copilot is not Google, and reasoning from a single source is reasoning with one eye closed.

Then, only a couple of days after Bing's launch, Google added a Generative AI section to Search Console: first-party data on how a site performs inside Google's AI surfaces, sitting right next to the familiar organic search report. Suddenly I had two independent first-party windows into the same question instead of one.

So I did the obvious thing and lined Google's data up against the theory I had just published.

What Both Engines Agree On

Start with the part that held up, because it's the more important finding.

In the Bing piece I described removing a large amount of older content (a deliberate pruning, not a traffic decision) and noticing that the removed pages fell out of AI citations roughly two weeks before they fell out of ordinary search. Google's data showed the same shape. The AI-surface drop led the organic-search drop by about the same two-week margin.

So the directional finding is not a Bing quirk. On both engines, when a page goes away, the AI layer notices before the classic search index does. That part I'd now state with more confidence than I had a week ago, precisely because two unrelated systems agree on it.

Where Google Broke My Explanation

Here's the part I got wrong.

In the Bing piece I explained that two-week lead with a specific mechanism: AI search still uses the index to find candidate pages, but then, unlike classic search, it fetches those pages live, at the moment of the query, to read what's actually on them. A removed page is still sitting in the index, so the AI finds it, tries to open it, gets a 404, and quietly leaves it out. Instant drop. Clean story.

Google's timeline doesn't fit that story at all.

I removed those pages back in March. If Google's AI were genuinely fetching them live at query time, they would have started failing in March, the moment they began returning 404, and dropped out of AI answers right then. They didn't. They kept appearing in Google's AI answers until mid-May, then dropped, with organic following about two weeks later.

A page that is gone for two months but still cited cannot be getting fetched live. There is no version of "read it fresh at query time" that produces that behavior. Whatever Google's AI is reading, it is reading a stored picture of the page (its own cached index), and that cache refreshes on its own schedule, ahead of the main search index but plainly not in real time.

Which means my Bing observation was right and my Bing explanation was wrong. The faster-than-search drop is real. "Live fetch" was the wrong reason for it.

And to be fair to the evidence: Bing's fast drop never actually proved the live-fetch theory in the first place. A drop "within a few days" is equally consistent with a cache that simply refreshes quickly. It was Google, by being slow, that exposed the real mechanism. Sometimes the system that lags tells you more than the system that keeps up.

What Actually Differs: Speed, Not Logic

Once you drop the live-fetch idea, the contrast between the two engines gets sharper and more interesting.

Both engines run AI and search on separate refresh cycles, with AI ahead by about two weeks. What differs is the absolute speed of the whole machine. Bing turned the entire pipeline (AI citations gone, then organic impressions gone) in a matter of weeks after the removal. Google sat on it: the pages were gone since March, yet nothing visibly moved until mid-May, and the dust didn't settle on organic until the end of the month. One engine reacted in weeks; the other took one to two months to react at all.

I would resist the urge to turn "about two weeks" into a constant. It happens to be the gap I measured on both engines, but these are two different products with two different crawl cadences, and the honest reading is that each runs its own clock: same direction, very different tempo.

A Quieter Signal: Who Even Gets the Data

There's one more thing worth flagging. It's about availability rather than timing, and it's the kind of thing that quietly confuses people.

Google's Generative AI report only appeared on one of my properties: the single highest-traffic site I manage. Across roughly ten smaller personal and company sites, the section isn't empty. It's absent entirely. There's no Generative AI report to open. So if you've read that the feature has rolled out, gone looking for it in your own Search Console, and found nothing, this is why. It is genuinely not everywhere yet.

That asymmetry is tempting to over-read, so let me be careful. The boring explanation, and the one I lean toward precisely because the report is missing rather than empty, is a phased rollout: Google is opening the feature to large or eligible properties first, which tells you nothing about your AI citations one way or the other. The more flattering explanation is that the report only exists where there's enough AI-surface activity to fill it, which would make its absence on small sites a quiet confirmation of the whole authority argument. I can't rule that out, but a blank-yet-present report would support it far better than no report at all, so I'm not going to lean on it.

Either way, there's a real limitation worth stating plainly: my entire Google-side analysis rests on a single property. The smaller sites can't corroborate any of it, because the report simply isn't there for them to check.

The Caveat, Restated

Everything here comes with the same warning the Bing piece ended on, only larger.

This is two engines, not the universe of AI search. It's a single high-traffic property on the Google side. The feature is new enough that it isn't broadly available, and citation behavior will keep evolving as these systems mature. I'm describing what two first-party datasets did over one removal event: a report, not a decree. The value isn't in the precise numbers; it's that two independent systems pointed the same direction and one of them ruled out an explanation I'd published.

What It Means In Practice

The practical takeaway is asymmetric, and it lines up with the core argument from the Bing piece: authority is authority.

When you remove or break a page, expect to lose AI citations first, faster than you lose ordinary search visibility on both engines. AI's separate, faster-refreshing cache catches the loss early. But the reverse does not hold. You cannot publish your way into AI citations any quicker than the index learns the page exists and the underlying authority accrues. Removal propagates fast; earning propagates at the old, slow pace.

So if you're pruning content to sharpen topical authority, which is exactly what I was doing, know that the AI surfaces will reflect the cut before your search reports do, and don't panic when they dip first. And if you're trying to gain AI visibility, there is still no shortcut. The work is the same work it always was: publish authoritative content, earn credible links, be consistent, and wait for the index and the model to catch up. A separate clock for AI doesn't change which clock you have to earn your way onto.

Google's Generative AI report is now the second number I'll watch alongside Bing's Citation Share. Two honest dials are better than one, even when one of them tells me I was wrong.