Google told the SEO industry something it's been suspecting for months: Search Console, the tool millions of site owners use as foundational truth for their organic search performance, doesn't work for AI search. Google confirmed the reporting gap directly. The positioning metrics that SEO teams have relied on for years don’t hold up once AI Overviews and AI Mode are a part of the results page.
That's not a minor caveat buried in a changelog. It's Google saying, in effect, that the dashboard you check every Monday morning is measuring a search experience that's already gone.
The Blind Spot Was Always There
None of this should be a surprise if you've been paying attention to what's showing up in search results. John Mueller said the quiet part out loud this week, too: The old "position 1–10" model is hard to map onto a results page where AI-generated answers sit above, inside, and around the traditional ten blue links. Almost 100% of People Also Ask boxes are now answered by AI Overviews as well. Fourteen months ago that figure was close to 12%. It’s a staggering increase.
Position tracking assumed a stable, rankable list, but AI search doesn't produce one. It produces a synthesized answer, assembled per query, that may or may not credit you, may or may not link out to you, and changes shape depending on the model, the user, and the moment. Search Console was built to report on the old model, and it was never going to survive contact with the new one gracefully. Google just said so first.
The Fine Print in Google's Admission
It's worth being precise about what broke, because "Search Console can't handle AI search" undersells it. The reporting problem Google confirmed isn't a missing feature bolted on later. It's a mismatch at the level of what the tool was designed to count in the first place. Average position, impressions, and click-through rate all assume a query returns a ranked list of links a user scans from top to bottom. AI Overviews and AI Mode don't produce that. They produce a generated paragraph that may cite you, may paraphrase you, or may fold your content into an answer alongside four other sources with no individual attribution visible to you as the publisher.
So when your Search Console dashboard shows a query with strong impressions and weak clicks, you have no way of knowing, from that report alone, whether you were ranked normally and simply outcompeted, cited inside an AI Overview and skipped over, or paraphrased without citation at all. Three completely different situations, one flat number. That's the gap Google confirmed exists, and it's a bigger admission than a single blog post headline suggests.
Why Citation Counts Never Told You Anything
We made this argument recently, before Google's admission gave it an assist: Tracking whether your brand got mentioned or not in an AI answer is a vanity metric. A citation isn't a click. It isn't a customer. It isn't even proof that a human being read your sentence rather than a paragraph an AI model paraphrased from three other sources at once.
Search Engine Journal made a version of the same case from a different angle, arguing that AI visibility scores collapse mentions, citations, rankings, and retrieval into one number that tells you almost nothing on its own. A Semrush study of manufacturing SEO reached the same conclusion from the data side: Getting cited by AI search doesn't reliably translate into traffic. The industry has built a KPI out of something that was only ever a symptom worth investigating, not a number worth chasing upward.
Now add Google's own Search Console admission on top of that, and the pattern is incredibly hard to wave away. The tool that was supposed to tell you how you're doing in AI search doesn't. The metric everyone reached for instead doesn't mean what it looks like it means. If you've been reporting AI citation counts to a boss or a client as a headline KPI, this is the week to stop.
What Deserves Measurement Instead
If citation volume is out, what replaces it? Three things, in order of how much they predict business outcomes:
- Structural health first. Can a crawler, human or AI, parse your content? Clean markup, valid structured data, and a crawl-and-render setup that doesn't choke are the prerequisites for showing up anywhere, in any format. We've written a full JSON-LD implementation walkthrough and a crawler governance guide if your last technical SEO audit predates the AI-crawler era, which for most sites it does.
- Retrieval quality second. Not "were we mentioned," but "was the thing the model said about us accurate, and did it point back to us," instead. That requires reading AI answers against your content, not counting mentions in aggregate. That's the distinction we laid out in detail here.
- Downstream behavior third. Referral traffic from AI platforms, branded search lift, and (the one nobody wants to hear) direct traffic growth, because a reader who remembered your name and typed it into a browser is worth more than a reader an AI paraphrased you to and never named you at all.
None of these live in Search Console today. Google just told us that. The SEO teams who build their own measurement layer on top of raw log data, crawl reports, and referral analytics, instead of waiting for Google to backfill a report that flatters its own product, are the ones who'll know what's happening to their traffic before their competitors do.
Where the Humans Come In
This is the kind of moment where "just let AI handle it" and "ignore AI entirely" are both wrong answers. AI tools can process crawl logs, flag structured data errors, and surface which pages are getting paraphrased versus ignored, at a scale no analyst could match by hand. That's the abundant part. The grunt work of assembling and cross-referencing data from ten different sources gets cheap and fast.
What doesn't get cheap is deciding what the data means for your specific business, your specific content, and your specific audience. A model can tell you that the citation rate dropped 12% this month, but it can't tell you whether that's because your content quality slipped, because a competitor shipped better schema, or because Google quietly changed which sources it favors for your category. That judgment call still needs a person who knows the business, not a dashboard.
Google admitting its own reporting is inadequate isn't entirely bad news though. It's also permission. Permission to stop treating a vendor's dashboard as the finish line and start building a measurement approach that answers the question that matters: Is this driving the outcome we're paid to drive, or just a number that looks good in a slide?
A Starting Checklist for the Next Sprint
None of this requires waiting for Google to fix its own tooling. Here are some concrete moves, roughly in the order they pay off:
- Pull raw server logs, not just Search Console exports. Log files show every crawler that hits your site, including AI crawlers Search Console doesn't break out separately. That's the first place to see whether AI systems are visiting pages Search Console says are ranking fine.
- Set up branded search tracking as its own signal. If AI citations are working, branded search volume should trend up over time as more people encounter your name without a link and go looking for you directly. If it's flat while citation counts climb, the citations aren't doing anything.
- Manually test retrieval on your ten highest-value pages. Ask the AI tools your audience actually uses the questions those pages answer. Read what comes back. Note whether it's accurate, whether it names you, and whether it links back. This doesn't scale to every page, and it shouldn't have to. It scales to the pages where getting it wrong costs money.
- Treat referral traffic from AI platforms as a real analytics segment, not a rounding error folded into "other." Even a small, clearly tagged number here is more honest than a citation count that can't tell you what it's counting.
None of that replaces Search Console. It simply fills in the parts that Google just admitted their tool can't see.
Where This Leaves You
There's a version of this story where the takeaway is despair: The platform your entire measurement stack was built on top of just told you it can't measure the thing that matters most right now. That's one way to read it.
The more useful read is that Google just handed the SEO industry permission to stop arguing about whether Search Console's AI reporting is good enough. It isn't. Google said so. That argument is over, and the only decision left is whether your team spends the next quarter or two (or more?) waiting for a fix, or builds the missing layer right now.
If Search Console can't tell you a specific insight yet, you can still find out. You just have to look somewhere else.
