SEO and AI Visibility Are Drifting Apart. Measure Them Separately.
A market correlation makes SEO and AI-answer visibility look like they are converging, but individual sites diverge, so measure the two separately.
If you do SEO for a living, you have probably heard the reassuring version of the AI-search story: keep the rankings strong and you will show up in AI answers too, because the two move together. This month a widely shared dataset looked like proof. Kevin Indig's Search Signals Index, which scores the same sites on both traditional search visibility and presence in Google's AI Overviews, the AI-generated answer box that now sits above the classic blue links, found that the link between the two flipped from slightly negative to slightly positive month over month: from -0.21 in June to +0.22 in July across the 143 largest companies it tracks (Growth Memo).
A positive correlation reads like a free lunch: do the SEO work, collect the AI presence on the side. But +0.22 is a faint signal on a scale that runs to 1.0, and treating a market-wide average that weak as a promise about your specific site is exactly where this goes wrong.
The same dataset shows sites moving in opposite directions
The tell is buried in the very report that produced the friendly correlation. Over the same 30-day window, Reddit's organic search visibility fell 14 percent while its mentions inside AI Overviews rose 5 percent (Growth Memo). One site, two surfaces, opposite directions. Stock-photo libraries showed the same split from the other side: Shutterfly, Getty, Shutterstock and six peers lost organic visibility together, Shutterfly by nearly 45 percent, while those same libraries kept turning up as citations in AI answers.
Then there is what AI Mode, Google's chat-style search experience, chose to cite. In the same window it roughly halved how often it pointed at publishers and instead tripled its mentions of Home Depot and quadrupled Wayfair, a swing from informational sources toward commercial ones (Growth Memo). If you publish content for a living, your AI trajectory and your ranking trajectory are not the same line, and a +0.22 average across 143 giants tells you almost nothing about which way yours bends.
Key Insight
A correlation measured across the whole market is not a forecast for your account. In this dataset the July number went positive and the individual-site divergences widened in the same breath.
The platforms have made the AI surface hard to see
Deciding to measure the AI surface on its own is easy to say and, right now, deliberately hard to do. Google folds AI Overviews and AI Mode data into the general Search Console performance report rather than breaking it out, as John Mueller confirmed (Digital Applied). That is why strange query rows like "yes" and "yes go on" have started appearing in people's reports: a follow-up inside an AI Mode conversation counts as a brand-new query, so its impressions and clicks land in the same table as your normal searches.
The dedicated generative-AI report that was supposed to solve this carries no query dimension and shows no clicks or click-through rate (Digital Applied). A separate practitioner teardown reached the same verdict, that queries and clicks are the two things the AI report leaves out and there is no API to pull them either (Suganthan Mohanadasan). So the surface everyone is telling you to win is the one your own analytics is currently worst at showing.
What to actually measure when the clicks do not come
If the dashboards will not hand you AI visibility cleanly, you fall back on proxies, and the useful ones survive the fact that AI answers often resolve a question without ever sending a click.
Branded search volume is the first. Semrush found that pages which begin appearing in AI Overviews see roughly a 9 percent lift in branded search within 90 days (Kieran Flanagan). A rise in people searching your name, right after you start surfacing in answers, is the click the AI box swallowed reappearing one step later.
It also pays to separate being cited from being recommended, because they are not the same event. In a Semrush sample of hundreds of thousands of ChatGPT responses, brands were cited as a source far more often than they were named as a recommendation: in topically distant categories about 50 percent of appearances were citations and only 25 percent were named mentions (Growth Memo). A single blended "AI visibility" score hides which one you are getting, and the two call for different work. That same study found expanding into adjacent topics helped visibility in finance and real estate but hurt it in legal and healthcare, and its author is careful to call these associations, not proven cause. Your result is category-specific, which is the whole argument for measuring your own.
None of this is worth doing if the traffic is junk, so watch whether it converts. Conductor's 2026 report, relayed by Flanagan, put visitors arriving from LLMs at about twice the conversion rate of traditional search while making up roughly a third of sessions where anyone bothers to measure it (Kieran Flanagan). Thinner traffic, better traffic. That asymmetry is the reason to instrument this surface now instead of waiting for Google to build you a clean report.
Tip
Pick 10 to 15 questions a real buyer would actually type, run them across the engines you care about on a set schedule, and log whether you were cited, named, or absent each time. It is a spreadsheet, not a platform, and it beats a blended score.
The uncomfortable version
Here is the case against everything above. The correlation did move positive, and it might keep climbing. Maybe by next quarter the two surfaces genuinely converge and separate measurement looks like wasted motion. That is possible. But the risk is lopsided. If you assume convergence and you are wrong, you optimize one surface while flying blind on the other, and you find out only when a competitor has become the default recommendation inside the answer your buyers actually read.
This is the part we keep returning to when we teach measurement at Promptafire: a market statistic is not a personal guarantee. The durable skill in AI search is unglamorous, and it is the same one that mattered in the SEO era: check your own data before you act on someone else's average. The correlation makes the headline. What is happening to your own site is a separate question, and for now it is one you have to go and measure yourself.
Sources
- Kevin Indig, Growth Memo, Growth Intelligence Brief #22: https://www.growth-memo.com/p/growth-intelligence-brief-22
- Kevin Indig, Growth Memo, on topical focus and AI brand visibility (Semrush AI Visibility Toolkit): https://www.growth-memo.com/p/does-topical-focus-make-your-brand
- Kieran Flanagan, The State of B2B Marketing: https://www.kieranflanagan.io/p/the-state-of-b2b-marketing-whats
- Digital Applied, AI Mode follow-up queries in Search Console: https://www.digitalapplied.com/blog/search-console-ai-mode-followup-queries-data-quality
- Suganthan Mohanadasan, AI Mode queries in Search Console: https://suganthan.com/blog/ai-mode-queries-search-console/