AI Search Statistics 2026: The Complete State of AI Search

AI search shifted from novelty to infrastructure between 2024 and 2026. People stopped opening Google and started typing the same questions into ChatGPT, Perplexity, Gemini, Claude, and Copilot. Google itself answered back with AI Overviews and then AI Mode. Every brand that lives on organic traffic is now asking the same three questions. Are these engines actually being used? Do they send anyone? And what does it take to get cited?

This page is the full data answer. Every number traces to its primary source with the date of measurement so you can see whether each figure is current, trending, or starting to age.

The numbers at a glance

Twelve numbers that frame the rest of the story. Each one is unpacked in the section below with its primary source and date.

Part 1. The scale of AI search

The shorthand for "AI search is real now" has three numbers behind it. How many people actually use ChatGPT. How much traffic AI engines send to the open web. And how fast that traffic is growing.

ChatGPT alone is the size of a top-five search engine

On February 27, 2026, OpenAI disclosed that ChatGPT had reached 900 million weekly active users. The same disclosure broke down the paid base: roughly 50 million paying consumer subscribers and 9 million paying business users.

For scale, Google handles roughly 14 billion searches per day across all its surfaces. ChatGPT is not a one-for-one substitute. People use it for different question types, often as a complement rather than a replacement. But 900 million weekly actives puts ChatGPT firmly in the same conversation as the biggest search-style products, and that conversation no longer needs qualifiers.

AI engines send 1.13 billion referrals a month, growing 357% year over year

Similarweb's analysis of the top 1,000 sites by traffic, published June 2025, found that AI referral traffic reached 1.13 billion monthly visits across that group, up 357% year over year. ChatGPT accounted for roughly 80% of the AI referrals across those sites.

Two takeaways. First, AI search is no longer a footnote in referral reports. It is a category that grew nearly five-fold in twelve months. Second, ChatGPT's dominance of the referral mix means that "AI search visibility" in practice is heavily shaped by what ChatGPT cites. The other engines matter, but ChatGPT is the gravity well.

That figure is from mid-2025. Given the curve, the current number is materially larger. Treat 1.13B as the floor.

Part 2. The growth curve over time

One stat is a snapshot. Several stats with dates make a trend. Here is the trajectory of AI search from late 2024 through mid-2026.

Three patterns stand out. The growth in AI Overview coverage is not linear, it is geometric. The Princeton study is the only major academic intervention with a primary peer-reviewed citation, and most of the GEO advice you read online traces back to it. And GA4 finally getting an AI Assistant channel in May 2026 does not close the measurement gap. The channel covers three engines for user clicks. It does not cover Copilot, and it does not cover the crawlers that decide whether you get cited in the first place.

Part 3. AI Overviews coverage, the part Google does and does not tell you

AI Overviews are Google's own answer to the AI search question. They appear above the organic listings on selected queries, summarizing content from across the open web while linking to sources. The coverage question (how often AI Overviews actually appear) is one of the most cited and most miscited numbers in the entire category.

The number with a primary source is Onely's measurement, which tracked AI Overview presence across a large keyword sample and found coverage moved from 1.5% in September 2024 to 32% in September 2025.

The shape of that curve matters more than any single point on it. AI Overviews went from a rounding error to roughly a third of tracked queries in one year. The curve has not flattened. By the time you read this, the actual coverage figure is almost certainly higher than 32%.

Part 4. What AI engines actually cite

The biggest mental shift for SEO-trained operators is this. When an AI engine cites a brand, it usually cites someone else writing about that brand, not the brand's own site.

That split comes from AirOps's State of AI Search 2026 report. It is the most important number on this page for content strategy. What gets said about you across the open web matters more than what you say about yourself.

Freshness matters almost as much as source diversity. AirOps's same report found that more than 70% of all AI-cited pages were updated within the last 12 months, and 83% of commercial-intent citations come from pages refreshed in the past year. AI engines have a strong recency bias.

Where citations actually come from

The third-party share is not randomly distributed. AirOps and other practitioners consistently see the same source types over-represented in AI citations:

The bars above are directional. Exact mix varies by category and engine. The takeaway is the relative ordering, which is consistent across reports. Community and review sources outweigh owned media, often by a wide margin.

Our AI citation benchmark measured the engine-by-engine reality across 101 buyer questions on the four engines that support web search. ChatGPT cites 5.6 sources per answer on average. Perplexity 8.8. Claude 10.7. Gemini 14.3. On 70.3% of prompts, no two engines shared a top source at all. Each engine retrieves differently, which means the citation mix you target shifts by engine.

Part 5. Volatility, the part most dashboards hide

One of the most uncomfortable findings in AirOps's report is how unstable AI search outputs actually are. Run the same query twice and you may get materially different answers, with different sources cited.

The strategic implication: any rank tracking tool that gives you a single number for your AI visibility on a query is hiding variance. The right unit of measurement is share of voice across multiple runs, not a single snapshot.

That instability is also why brand-tracking dashboards need to run repeated queries instead of one-shot checks. A single query tells you very little. Five queries over a few days tells you whether you have real visibility or you got lucky once.

Part 6. The conversion fight

The question every CFO eventually asks: when AI engines send a visitor to your site, does that visitor do anything?

The answer depends almost entirely on who is asking. Two well-documented studies reach almost opposite conclusions because they measured almost opposite verticals.

The Seer data is from a single B2B research client and is documented in their case study on ChatGPT-referred traffic conversion. The retail counter-evidence comes from less centralized sources. Practitioner reports across e-commerce sites point in the same direction without a single canonical primary study to anchor the claim. Treat the Seer number as a strong floor for what AI traffic can do in research-heavy categories, not a ceiling that holds in retail.

Part 7. Brand visibility benchmarks

"How visible should we expect to be?" determines whether you are chasing the wrong target. Two data points anchor reasonable expectations.

Average mention rate is 16.3%, top performers hit 56.5%

AthenaHQ's 2026 State of AI Search analysis benchmarked brand mention rates across tracked AI engines for a large sample of B2B and consumer brands. The headline numbers:

Read that carefully. The median brand gets mentioned in roughly one out of six tracked queries. The top quartile in roughly one in three. Top performers (the brands genuinely winning at AI visibility) get mentioned in more than half of relevant queries.

If your tracked mention rate is in the single digits, you are not unusual. You are below median, but the median is also low. The goal is the top quartile and then the top decile, not 100% mention rate, which no brand achieves.

B2B share of voice typically sits below 30%

For B2B brands specifically, LLM Pulse's share-of-voice analysis finds that typical B2B brand SoV across AI engines sits below 30%. SoV in this context means the brand's share of total mentions across a tracked category set, against named competitors.

A category usually has 6 to 8 named competitors that AI engines rotate between, per Alex Birkett's analysis of AI share-of-voice patterns across multiple B2B verticals. Above 30% SoV against that competitor set means you are consistently in the top 2 to 3 mentioned brands in your category, which is the realistic ceiling for most categories.

Part 8. The measurement gap GA4 still cannot close

Even after Google added the AI Assistant channel to GA4 in May 2026, the gap between what AI engines actually do and what GA4 reports is enormous.

The ~99% missed figure comes from Atyla's analysis of AI bot traffic visibility in GA4 default configurations. The point is not that GA4 is broken. It is that GA4 measures users, not crawlers, and the crawler layer is where AI citation gets decided. Until you log server-side requests from bots like GPTBot, PerplexityBot, ClaudeBot, Google-Extended, OAI-SearchBot, and friends, you have no view into the half of AI search that determines whether you get cited at all. CiteTrack Crawl Academy is our reference on AI bot fingerprinting, reverse DNS verification, and the spoofing patterns that defeat naive user-agent logging.

Google's May 13, 2026 update added an "AI Assistant" channel grouping to GA4, defined in Google's Analytics Help documentation on channel groupings. It captures user-clicked referrals from ChatGPT, Gemini, and Claude. It does not include Copilot. It does not include AI crawlers. It does not include zero-click impressions.

Part 9. What actually moves citations

The strongest peer-reviewed evidence about what changes AI citation outcomes comes from Princeton's GEO study, published at ACM KDD 2024 and on arXiv as Aggarwal et al., 2311.09735.

The researchers tested nine content optimization techniques against a 10,000-query benchmark using multiple generative engines as testbeds. They measured a Position-Adjusted Word Count metric. How often the optimized content was cited and where in the answer it appeared. The top three techniques all clustered around the same 30 to 40% relative lift:

What is striking is what did not work. Classic SEO interventions like adding keywords or improving fluency had near-zero effect on citation. The things that move the needle look more like footnoting a research paper than optimizing for search. Cite your sources. Show your numbers. Quote your experts.

That study is from late 2024. It predates most of the current generation of AI search engines. The relative ordering of techniques is widely replicated in practitioner reports through 2026. Stats and citations consistently outperform keyword and fluency tweaks. But the specific 30 to 40% number should be read as "the magnitude is real" rather than "this is the exact lift you will see today." The Complete Guide to GEO walks through how to apply each Princeton-validated technique to live content work.

Part 10. The readiness gap

If the Princeton study shows what moves citations, the SEOScore.tools audit shows how few sites are actually doing any of it.

Between January and March 2026, SEOScore.tools audited more than 2,000 websites across three layers. Classic SEO. AEO (answer engine optimization, the broader category). And GEO (generative engine optimization, the content layer). The average scores:

Classic SEO is mature. The average site is roughly halfway competent. AEO readiness is about half of that. And GEO (the actual content-side work the Princeton study found matters) averages 13 out of 100 across two thousand sites.

That is the opportunity in one number. Most sites are not doing the work that AI engines reward. The bar to clear the bottom quartile is not high. The bar to reach the top quartile is mostly a question of being deliberate about it.

Part 11. Stats that need to be retired

A handful of widely repeated AI search statistics do not survive contact with their primary sources. These keep circulating because they sound right and are easy to cite. They are wrong, mis-scoped, or have no source at all.

Part 12. Putting it together

Twelve statistics give you the shape of AI search in 2026.

The category is big and growing fast. 900 million people use ChatGPT every week. AI engines send over a billion referrals a month to the top sites, growing 357% year over year. Google AI Overviews moved from 1.5% to 32% of tracked keywords in a year and the curve has not flattened.

Visibility is unevenly distributed. The median brand gets a 16% mention rate. Top performers hit 56%. B2B share of voice typically sits below 30% against a category set of 6 to 8 named competitors.

Where you get cited matters more than what you publish. 85% of brand mentions in AI search come from third-party pages, not your own domain. Reddit, Wikipedia, YouTube, and G2 are over-represented as citation sources. Freshness matters. 83% of commercial citations come from pages updated in the past year.

Stability is lower than you would think. Only 30% of AI answers stay consistent across two runs. Only 20% across five. Rank tracking with single snapshots hides the variance.

Conversion outcomes are vertical-specific. AI-referred traffic can convert dramatically better than Google organic in research-heavy B2B (15.9% to 1.76% in the Seer case). In transactional retail, conversion is often worse. Both are true.

The measurement layer is still broken. GA4 captures roughly 1% of relevant AI activity by default. The May 2026 AI Assistant channel covers three engines for user referrals but no crawlers and no Copilot. You need server-side bot logging to see the full picture.

The things that actually move citations look like academic writing. Statistics, citations, direct quotes. Classic SEO interventions like keyword stuffing or fluency rewrites have near-zero effect on AI citation.

And most sites are not doing any of it. The average GEO readiness score across 2,000 sites is 13 out of 100. That is the opportunity.

What to do next

If this page convinces you the category is real and your strategy needs to catch up, the next reads on this site are:

The Complete Guide to AEO covers the full discipline at the strategic level: on-site, off-site, and measurement.

The GEO playbook applies the Princeton findings to live content work, with the on-site interventions ranked by evidence weight.

Engine-by-engine specifics live in the dedicated guides: How to rank in ChatGPT, Google AI Overviews, Perplexity, Gemini, Claude, Copilot, and Google AI Mode.

The wider tool category sits in Best AEO Tools in 2026. 10 Best Tools to Track AI Citations is the citation-tracker subset.

Sources

Frequently asked questions