Semrush studied 126 million AI prompts — the takeaway is that almost no one can measure the thing that matters

What Semrush published

Per Semrush's announcement, dated June 26, 2026 and carried on BusinessWire, the index scaled from an initial 2,500-prompt pilot to 126 million US AI search prompts, spanning 22 industries. The findings worth acting on:

• Being mentioned ≠ being cited. Semrush found the overlap between the brands an engine names and the domains it cites as sources can be as low as 30% on Gemini — an AI can recommend you by name and never link or credit your site.
• Engines cite at wildly different depths: ChatGPT cites an average of 15 sources per response, Gemini an average of 3. The same question, asked on two engines, pulls from a completely different set of sources.
• Visibility is fragile and fragmented: only 36 brands held top-100 visibility across all four platforms in every month of the study — and they're giants like YouTube, Google, Reddit, Amazon, and Walmart.
• Almost no one can measure it: 45% of marketing leaders say they cannot accurately measure their brand's visibility inside AI answers, and only 9% have the tools to track all the relevant metrics across platforms.

Why 'mention ≠ citation' is the sentence to remember

Most AI-visibility advice quietly assumes the two move together — get cited, get mentioned, get traffic. Semrush's 30% Gemini overlap breaks that chain. An engine can tell a customer “try [your business]” while sourcing the answer entirely from a directory, a review site, or a competitor's page — so you'd see nothing in a backlink or referral tool, yet you're winning the recommendation. The reverse is just as real: you can be cited as a source and never named as the answer. This is the same split we keep measuring from our side — Cloudflare's data shows AI crawlers take up to 50,000 pages for every click back, and Previsible's report shows referral traffic is a tiny, skewed sliver of AI activity. Citations, referrals, and recommendations are three different channels. The one that books the customer is the recommendation — whether the model names you when someone asks.

The 45% measurement gap is the actual opportunity

That nearly half of marketing leaders can't measure their AI visibility, and only 9% have the tooling, is not a footnote — it's the state of the market. And the depth-of-citation gap (ChatGPT 15 sources, Gemini 3) means measuring one engine tells you almost nothing about the others; a business can look present on ChatGPT and be absent on Gemini for the identical query. For local businesses — med spas, law firms, home-services companies — this is decisive, because you don't have a brand the size of Walmart forcing your way into every engine's answer. In our own 1,120-answer comparison, the four engines disagreed sharply about who to recommend, and in our run-to-run study only 31% of recommended brands survived every repeat of the same question. You cannot manage a number you can't see across every engine your buyers use.

The takeaway

Semrush's index is a useful map of the terrain — big brands, big datasets, real fragmentation. The practical lesson for a business that isn't a household name is smaller and sharper: don't measure citations or referral clicks, and don't measure a single engine. Measure whether AI recommends you, across all of them. A free 60-second check samples ChatGPT and Perplexity (the $79 audit adds Gemini and Claude, 25 questions sampled three times each) and reports how often each names your business right now — the exact number Semrush says 91% of marketers still can't see. If you're comparing tools to track it over time, here's how the category stacks up.