June 25, 2026

/ AEO

8 min read

AI search statistics 2026: the numbers that define GEO

Most AI search statistics circulating in 2026 trace back to nothing. Here are the ones with a primary source behind them: adoption, citations, conversions, and what the data does not support.

AI search statistics 2026: the numbers that define GEO

The 2026 AI search numbers that hold up to a source check: Google AI Overviews reach more than 2.5 billion monthly users, generative AI platforms average 9.5 billion monthly visits, only 38 percent of AI Overview citations come from pages ranking in the organic top 10, and ChatGPT referrals convert to transactional sites at about 7 percent versus 5 percent from Google. Every figure below links to the organisation that produced it, not to a roundup of a roundup. Several widely repeated statistics did not survive that check, and we have listed those too, because knowing which numbers are fake is worth as much as knowing which are real.

How big is AI search in 2026?

AI search is a primary discovery surface. Google’s AI Overviews reach more than 2.5 billion monthly users, a figure Sundar Pichai gave in May 2026. OpenAI announced more than 1 billion active users on July 31, 2026, though that number spans consumer and enterprise products and is not a like-for-like ChatGPT figure. Sam Altman put ChatGPT at 2.5 billion prompts per day, but that was July 2025, so treat it as a floor rather than a current reading.

Similarweb measured generative AI platforms averaging 9.5 billion monthly web visits between June 2025 and May 2026, up roughly 70 percent year over year. On the demand side, Similarweb’s US panel found about 35 percent of consumers now use AI tools at the product-discovery stage, against 13.6 percent for search. That is the number that should move your planning: your buyers are forming a shortlist inside an AI answer before they ever reach a search results page.

Is ChatGPT losing share to other engines?

This is where the widely circulated numbers are worst, and where we were wrong in an earlier version of this post. You will see it claimed that ChatGPT’s share of AI-referred traffic fell from 87.2 percent to 68 percent. We repeated that figure. It traces to no primary source, and the metric is mislabelled: those percentages describe share of visits to the chatbots, not traffic they refer out.

On the metric as actually named, the evidence runs the other way. Previsible’s July 2026 analysis of 6.77 million sessions across 166 GA4 properties put ChatGPT at 92.4 percent of standalone LLM referral traffic, and gaining rather than shrinking. So the honest 2026 picture is not a four-engine plurality. It is one dominant referrer and a long tail.

That does not make per-engine work pointless, because the engines cite different sources. Profound found only 11 percent domain overlap between ChatGPT and Perplexity citations across 100,000 prompts, and Writesonic’s July 2026 study of 161,286 prompts put it at 13 percent. Being cited in one engine tells you almost nothing about the others. But if you have to pick one to win first, the referral data says pick ChatGPT.

How often do AI engines actually cite sources?

Citation rates are climbing, which makes this more winnable every quarter. Similarweb’s 2026 Generative AI Landscape found ChatGPT’s citation rate rose from roughly 1.3 to 1.6 percent in June 2025 to 6.8 percent by May 2026. It climbed non-linearly, so do not model it as a straight line.

Concentration matters more than frequency. Similarweb’s most-cited domains analysis from April 2026 puts Wikipedia at 13.15 percent of ChatGPT citations, Reddit at 11.97 percent, OpenAI’s own domains at 6.21 percent, Walmart at 2.90 percent, YouTube at 2.67 percent, and NIH at 2.22 percent. Engines lean hard on a short list of trusted sources. Presence on the platforms an engine already trusts beats chasing raw backlink volume, and Google AI Overviews draw on a different set again, which is why a blended strategy underperforms a per-engine one.

Do AI search visitors actually convert?

Better than organic search, though not as dramatically as the popular numbers claim. Similarweb reports ChatGPT referrals converting on transactional sites at about 7 percent versus about 5 percent from Google. You will see 7.1 percent quoted against 7.8 percent for paid search. Neither of those appears in any Similarweb publication.

One contrary finding is worth knowing before you build a forecast on this. A study of 973 ecommerce sites covering roughly $20 billion in revenue found ChatGPT referrals converting worse than organic search. Seer Interactive’s much-quoted figures (15.9 percent for ChatGPT against 1.76 percent organic) come from a single B2B client’s GA4, not an industry benchmark. The defensible claim is that AI referrals convert at least as well as organic and often better, because the visitor arrives with the question already framed. Anything more precise than that is currently unsupported.

Similarweb did observe referral traffic rising around 150 percent week over week when ChatGPT began surfacing clickable brand links in answers around May 7, 2026. SE Ranking’s independent study of 101,574 sites measured a smaller 36.7 percent month-over-month move and cautions the causal link is unproven.

What do the citation stats say about how to win?

Structure beats rank, and this part is well evidenced. Ahrefs analysed 863,000 keywords and found only 38 percent of AI Overview citations come from pages in the organic top 10. The remaining 62.2 percent split between positions 11 to 100 (31.2 percent) and beyond position 100 (31.0 percent). You do not have to outrank everyone to get quoted.

Covering the full question cluster is the strongest lever with real data behind it. Surfer’s study of 173,902 URLs found pages ranking for both a main query and its fan-out sub-queries were 161 percent more likely to be cited, 51.2 percent against 19.6 percent. Ahrefs separately found branded mentions correlate with AI Overview presence roughly three times more strongly than backlinks across 75,000 brands, though it flags every factor as moderate to weak.

Freshness is weaker than commonly claimed. Across 17 million citations Ahrefs found AI-cited content is 25.7 percent fresher on average, but the typical cited page is still 2.9 years old, and Google AI Overviews actually cite pages about 16 days older than organic results. There is no 30-day cliff.

The statistics we removed, and why

We previously published several numbers that do not survive a source check. Naming them is more useful than quietly deleting them, because you will meet all of them in other people’s decks.

“Structured data gives a 73 percent higher selection rate in AI Overviews.” This traces to a domain that is now a parked spam page. The only controlled test we can find points the other way: Ahrefs compared 1,885 pages that added JSON-LD against matched controls and found no significant uplift, and a significantly negative effect for AI Overviews. We still implement schema for clients, because it governs rich-result eligibility and machine readability, but we are not going to claim a citation lift the evidence does not support.

“Multi-modal pages get 156 percent higher selection.” Same dead source. No vendor with a real dataset has published this comparison.

“59.6 percent of citations come from URLs not ranking in the top 20.” Invented. Ahrefs does not use a top-20 bucket at all.

“Citation frequency rose from 0.6 percent in January 2025 to 2.8 percent in August 2025.” Neither figure nor either date appears in any dataset we can locate.

“83 percent of searches end without a click when an AI Overview appears.” No primary source publishes 83 percent. Pew found users clicked a result on 8 percent of visits with an AI summary against 15 percent without. SparkToro’s 2026 figure for zero-click generally is 68 percent.

“ChatGPT Search handles 250 to 500 million queries per week.” Attributed to Similarweb, which measures visits and cannot measure queries. OpenAI itself reported over 1 billion searches in a single week back in April 2025.

“44 percent of AI Overview citations come from content published in 2025.” Dead-ends at an unattributed vendor blog.

What this means for a small or mid-size business

The opening is real, and it is narrower than the hype suggests. Citation favours structure over domain size, so a well-built page from a small firm genuinely can be quoted alongside a national brand when only 38 percent of citations come from top-10 pages. AI referrals convert at least as well as organic. And most local and professional niches still have nobody optimising for this.

What the data does not support is a precise ROI forecast. Anyone showing you a conversion multiple to one decimal place is quoting a number that has been through three aggregators. Build on the findings that have a named study and a real sample behind them: answer-first structure, full question-cluster coverage, branded mentions, and presence on the sources each engine already trusts.

Frequently asked questions

How many people use AI search in 2026? Google AI Overviews reach more than 2.5 billion monthly users per Sundar Pichai in May 2026, and OpenAI reported more than 1 billion active users across its products in July 2026. Generative AI platforms average about 9.5 billion monthly web visits, up roughly 70 percent year over year per Similarweb.

Do AI search visitors convert better than regular search visitors? Similarweb puts ChatGPT referrals at about 7 percent conversion on transactional sites against about 5 percent from Google. Treat more precise multiples with suspicion, and note that one study of 973 ecommerce sites found the opposite.

Is ChatGPT losing market share in AI search? Not on referral traffic. Previsible measured ChatGPT at 92.4 percent of standalone LLM referral traffic in July 2026 and gaining. The widely quoted drop from 87.2 percent to 68 percent has no primary source and describes a different metric.

How often do AI answers include citations? ChatGPT’s citation rate rose from roughly 1.3 to 1.6 percent in June 2025 to about 6.8 percent by May 2026 per Similarweb. It is still low, which is exactly why the opportunity is open.

Does schema markup increase AI citations? The only controlled test available found no significant uplift, and a negative effect specifically for AI Overviews. Schema still matters for rich-result eligibility and machine readability, but claims of a large citation lift are not supported.

Do you need to rank on page one to get cited by AI? No. Ahrefs found only 38 percent of AI Overview citations come from organic top-10 pages, with 62.2 percent coming from positions 11 and beyond.

Turn the numbers into a plan

The honest 2026 picture: AI search is very large, ChatGPT dominates referrals rather than sharing them, citation rates are rising fast from a low base, and structure beats rank. It is also a field where most published statistics are laundered through aggregators until they stop meaning anything. Build on sourced findings, and check anyone who cannot tell you who ran the study. For the full method, read AI search optimization: the complete 2026 guide and GEO vs SEO. To see where your visibility stands, run our free GSC analysis or book a call.

Sources, primary only: Alphabet Q1 earnings, Sundar Pichai, OpenAI: Scaling AI for everyone, Similarweb: generative AI stats, Similarweb: most-cited domains in LLMs, Ahrefs: AI Overview citations and the top 10, Ahrefs: schema and AI citations, Ahrefs: do AI assistants prefer fresh content, Ahrefs: branded mentions correlation, Surfer: query fan-out impact, Profound: citation overlap, Previsible: AI traffic report July 2026, SparkToro: zero-click 2026, Seer Interactive: how ChatGPT traffic converts

Tagged

ai search statistics geo aeo chatgpt data