62% of Google's AI Citations Now Skip the Pages That Rank. Marketing's Dashboards Can't See It.

On July 10, Kevin Indig published the 21st edition of his Growth Intelligence Brief with a chart that should stop any SEO lead cold. In the four weeks to June 29, LinkedIn’s Google search visibility climbed 43.3%. Its mentions inside Google’s AI Overviews fell 2%.

It wasn’t alone. Reddit gained 18.1% in visibility and lost 12.8% of its AI mentions. Facebook: up 20.1%, down 12.1%. Instagram: up 21.4%, down 9.7%. The five biggest platforms on the web all won the ranking game and lost the AI game in the same month, measured by Indig’s Search Signals Index, which tracks roughly 2,600 companies across 26 verticals.

Indig, a growth advisor who writes the Growth Memo newsletter, put it plainly:

“SEO visibility and AI citation used to drift together for the mega-platforms. In June, they came apart.”

Google is running two search engines now, and most marketing teams measure only one of them.

For a year the story was that AI answers were eating the click. This is the sequel, and it’s harder on anyone who has to prove the channel works. The demand is moving to a surface the rank report can’t see, and the rank report is what most SEO budgets are still graded on.

The split shows up in the citation data too. Ahrefs studied 863,000 keyword results and 4 million AI Overview links in a March analysis and found that only 38% of the pages cited in AI answers still ranked in Google’s top ten for that query. In July 2025 the figure was 76%. In eight months, page one went from a near-guarantee of getting quoted to a coin flip. The other 62% of citations now come almost evenly from pages ranking 11 to 100 and pages ranking past 100.

Louise Linehan, the content marketer at Ahrefs who ran the study, drew the line for anyone still using rank as a stand-in for AI visibility. Citation analysis, she wrote, “moves you from ‘ranking well’ to ‘being selected as a source,’ which as we know, may not always be the same thing.”

Even a citation you do win pays less than it used to. Search Engine Land’s analysis found that ranking first inside an AI Overview earns about the clicks of a sixth-place organic result. The top spot in the new box is worth a middle spot in the old one.

Here’s why the two numbers came apart. When a search triggers an AI Overview, Google no longer just reads the page that ranks. It runs a query fan-out, quietly breaking your question into a spray of related sub-questions, finding a source for each, then stitching the answer from all of them. A page sitting 40th for your head term can get cited because it answered one narrow branch better than the page in first. Rankings measure the head term. Citations measure the branches.

The measurement tools never caught up. Google Search Console, the free dashboard most teams treat as ground truth, counts an AI Overview citation as a plain impression, the tally of times your link was shown. It files that citation in the same bucket as every ordinary result. Illia Zub, writing in Search Engine Journal, was blunt about the blind spot: Search Console gives you “No filters. No segmentation. No way to know if your content appears in AI Overviews.” The rank trackers marketers pay for read position, and position is the signal that stopped predicting AI visibility.

For a marketing team, that breaks the ROI story from both ends. The dashboard showing rankings holding steady now hides a channel quietly rerouting demand, so the numbers read survivable right up until pipeline slips. And the work that wins AI citations, structuring a page so a model can lift a clean answer, doesn’t move the rank a CFO asked you to defend. You can do the new job well and show nothing for it on the old scorecard. It’s the latest turn in what AI is doing to content and discovery, and it lands hardest on the teams that grade themselves by sessions.

The gap only widened in the quarter after Indig’s chart. AI agents made 17.7 billion requests to DataDome’s network from April to June, up 45%, while the crawler behind ChatGPT sent 6% fewer visitors to the pages it read. The reading is up and the referrals are down, which is why referral traffic has stopped measuring AI visibility at all. Grade the channel by clicks and you understate it twice over.

This is the same gap a new telemetry standard is trying to close: teams have been building pages to get cited in AI answers with no instrument that confirms it worked. The advantage now sits with whoever can own the answer, not just the ranking, and prove it.

So stop grading SEO on the number that came unmoored from the result. Rank still matters as an input, since page-one presence raises your odds of getting pulled into the fan. But it quit being the scoreboard the day citations started coming from page ten. Put citation share on the report next to rankings, track how often the AI answer names you for the questions your buyers actually ask, and defend that line the way you once defended position one. Google built a second search engine inside the first. Measure the one your customers are already using.