AI Overviews Hit 43% of Searches. Citations Still Lag.

AI Overviews now show up on 43% of Google searches, up from 15% a year ago, according to Similarweb data reported by TechCrunch on July 27, 2026. Search behavior has flipped: AI-generated answers are no longer a layer bolted onto results, they're becoming the default surface. But the citation supply feeding those answers hasn't kept pace with the growth in usage. Only 6.8% of U.S. ChatGPT desktop queries carried a citation as of May 2026, even after fivefold growth over the year. Most GEO programs are optimizing for a shrinking share of an exploding channel, and winning a citation may not mean what most teams think it means.

AI Overviews Now Touch 43% of All Google Searches

A year ago, AI Overviews appeared on 15% of Google searches. As of July 2026, that number is 43%, per Similarweb's 2026 Generative AI usage analysis, reported by TechCrunch. That's not incremental growth, that's nearly triple the coverage in twelve months, on the search engine that still handles the majority of the world's queries. Average query length is rising too, as searchers shift toward longer, conversational phrasing built for AI systems rather than ten-blue-links retrieval. If your content is still written for keyword-match snippets, you're writing for a shrinking share of the page.

Google AI Mode is compounding the shift. Visits rose from 126 million in June 2025 to 279 million by May 2026, more than doubling in under a year. And the mechanics changed too: after a May 7, 2026 Google search update, the share of desktop AI-related visits that referred users onward to actual webpages more than doubled, from 25% in March 2026 to nearly 60% by May 30, 2026. Google tuned AI Mode to send more traffic out, not less. That's a meaningful shift in AI search visibility economics, and it means the pages Google chooses to link from AI Mode answers now matter more than they did three months ago.

None of this means abandon keyword-driven pages. It means stop treating them as the whole strategy. A page built to rank for a ten-word query still has to work for a forty-word conversational one, because that's the query shape AI Overviews is increasingly built to answer. Structure matters more than density now: clear headers, direct answers near the top, specific numbers instead of vague claims. That's the format both Google's summarizer and a human skimming an AI answer reward, and it's the baseline every page needs before any citation strategy on top of it matters.

ChatGPT Citations Grew 5x — But Still Sit at 6.8%

Citation inclusion in AI answers rose more than fivefold over the past year. That sounds like the headline. It isn't. Only 6.8% of U.S. ChatGPT desktop queries actually included a citation as of May 2026, up from about 1.3% in June 2025, per Similarweb's report, covered by ppc.land. Growth rate and absolute opportunity are two different stories. A 5x jump from 1.3% still lands under 7 in 100 queries, or roughly 1 in 15 ChatGPT queries surfacing a citation at all. For everything else, the model answers without naming a source, meaning no link, no ChatGPT citations, no attribution, no matter how well-optimized your content is.

This matters more because ChatGPT itself is losing ground. Its share of GenAI web traffic fell from above 80% in late 2024 to just above 50% by May 2026, as Gemini, Claude, Perplexity, and DeepSeek absorbed the difference. Average monthly visits across GenAI platforms hit 9.5 billion between June 2025 and May 2026, up 70% from 5.6 billion the year before. Monthly unique visitors climbed to 655 million, up 57% from 418 million. The audience is real and growing fast. The citation mechanism that would actually route that audience to your site is not growing at the same rate. It's growing off a near-zero base, and it's fragmenting across five platforms with five different citation behaviors.

The category split is a hint, not a rulebook. Travel, retail, and sports lead ChatGPT citations because those queries have a clean answer shape: a specific place, product, or stat that a single source can settle. B2B research queries rarely look like that. They're comparative, multi-step, and often answered by stitching together several sources instead of citing one. That's likely part of why B2B categories don't crack the top of ChatGPT's citation list yet, not because the content is weaker, but because the query pattern doesn't reward a single clean citation the way a 'best time to visit Lisbon' query does.

AI Mode and the Multi-Platform Land Grab

Treat AI search visibility as one channel and you'll misallocate budget. It's five channels wearing one label. Google AI Overviews and AI Mode sit on top of the search engine most B2B buyers still start with. ChatGPT still commands roughly half of GenAI traffic on its own, even after losing 30 points of share in eighteen months. Gemini, Claude, Perplexity, and DeepSeek are splitting the rest, and each surfaces sources differently: some cite inline, some list sources in a sidebar, some barely cite at all. A GEO program built around one platform's citation format will underperform the moment that platform loses another five points of share.

The economics behind this shift are changing too, not just the traffic numbers. Crawl activity from AI systems doesn't convert to referral traffic at anything close to a 1:1 rate, we've broken down that gap in crawl-to-referral economics, and the pattern holds here as well. High crawl volume from an AI system is not evidence of citation share, and citation share is not evidence of referral traffic. Each is a separate funnel stage, and most reporting collapses all three into a single AI visibility number that tells you almost nothing about revenue impact. That collapse is especially costly for B2B SaaS teams, where a single enterprise citation can be worth more than a thousand retail ones, but only if it's the citation type that actually drives a click. Budget accordingly: a platform sending you 30% of a shrinking pie can still outperform one sending you 5% of a pie that tripled, and most GEO budgets in 2026 are still allocated by last year's platform mix rather than this year's traffic reality.

Why an AI Overviews Citation Doesn't Guarantee Traffic

Here's the part most GEO reporting skips: getting cited and getting traffic are not the same win. Aleyda Solis, SEO consultant and founder of Orainti, flagged the gap directly in Similarweb's report.

“65% of cited URLs sit two or three folders deep... while 58.8% of AI referral traffic lands on homepages.”

Read that twice. The pages AI systems cite most often are deep pages, specific guides, product docs, comparison content, three folders into your site architecture. But the pages actually receiving AI referral traffic are overwhelmingly homepages. Citation and traffic are running on different tracks. A page can be well-cited and never see a visitor. A homepage can absorb the traffic from citations it never earned. If your dashboard tracks a single blended mention rate, you're averaging two different phenomena into one number that guides neither content strategy nor traffic forecasting correctly.

Fix Your GEO Program: Track Citations and Referrals Separately

Stop reporting citation rate as if it were traffic. Split the measurement, then act on each half independently. Three moves, in order.

Do this next: export your last 90 days of AI-platform referral traffic by landing page, then pull citation mentions for the same pages from your monitoring tool. If the two lists barely overlap, you don't have a citation problem or a traffic problem, you have a measurement problem, and it's been hiding which pages actually need work. That's the audit we run at the start of every one of our GEO engagements, and it usually reshuffles the top five priorities on a client's content roadmap in the first week. Don't wait for citation rates to climb before you fix this. The 6.8% figure will keep rising, and if your reporting still treats it as one number, a bigger citation rate just means a bigger blind spot next quarter, not a better one.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.