A market briefing on where AI search actually stands right now: what it costs, who's winning, how adoption differs by country, and what's already changed in how content earns visibility.
What This Briefing Is (and Isn't)
This isn't a study SumukhAI ran ourselves. It's a compiled briefing: we pulled together the market research, government data and academic papers that are actually shaping how AI search works right now, checked each figure against its original source, and cut or clearly flagged anything we couldn't verify or that turned out to be misquoted or conflated elsewhere online. Where a number is disputed between research firms, we've said so instead of picking whichever one sounds most impressive. That's the same "properly, or not at all" standard the rest of this Insights section is held to.
The Market, By the Numbers
The global AI software market reached roughly USD 174 billion in 2025 and is projected to grow to USD 467 billion by 2030, a compound annual growth rate of about 25%, according to ABI Research. The generative AI segment specifically is growing faster still: from USD 63.7 billion in 2025 to a projected USD 220 billion by 2030, a 29% CAGR, which would take generative AI from 37% to 47% of the total AI software market by the end of the decade.
The AI search engine market itself was valued between USD 16.3 billion and USD 18.84 billion in 2025, depending on which research firm you ask (Precedence Research and SNS Insider, respectively). Longer-range forecasts diverge even more: Grand View Research projects the market reaching roughly USD 50.9 billion by 2033, while Precedence Research's own longer forecast puts it at USD 182.17 billion by 2035. Different firms, different modeling assumptions, both worth knowing rather than treating either as the single "real" number.
Investment has followed. Total global AI venture funding reached USD 225.8 billion in 2025, roughly double the previous highs set in 2021 and 2024 (both around USD 114-115 billion), according to Crunchbase and CB Insights data. AI accounted for about 48% of all equity funding raised globally despite representing only 23% of total deals, which tells you the money is concentrating into fewer, larger bets rather than spreading wider. Those bets include OpenAI's USD 40 billion raise in March 2025 at a roughly USD 300 billion valuation, Anthropic's two 2025 raises totaling around USD 16.5 billion (a USD 3.5 billion round in March, a USD 13 billion round in September), and Meta's USD 14.3 billion investment in Scale AI in June 2025 for a 49% non-voting stake, valuing Scale at roughly USD 29 billion.
What an AI Answer Actually Costs
A standard Google search consumes roughly 0.0003 kilowatt-hours of energy, producing up to about 0.2 grams of CO2 equivalent, per Google's own long-standing estimate. A single Gemini text prompt, by Google's own August 2025 environmental report, uses roughly 0.24 watt-hours and emits about 0.03 grams of CO2e. Independent estimates from Goldman Sachs and the International Energy Agency put a typical AI-generated answer at somewhere around 10 times the energy of a standard search, with a meaningfully higher carbon footprint to match. This is a big part of why "compute nationalism" has become a real phrase: the US, EU and China are all racing to lock down chip manufacturing, data center capacity and electrical grid supply, because the infrastructure underneath generative search is genuinely more expensive to run than what it's replacing.
The Platforms, One by One
Google's AI Mode passed 1 billion monthly active users by May 2026, announced at that year's Google I/O, with query volume roughly doubling every quarter since its 2025 launch. AI Overviews show up in anywhere from about a quarter to over half of Google searches depending on which tracker measures it (Conductor's Q1 2026 benchmark put it at 25.11%; other trackers report figures over 40%), reaching an estimated 1.7 billion monthly users. Google Gemini has become the clear number two in the standalone AI chatbot market, holding somewhere around 18-21.5% of chatbot referral traffic and roughly 2.6 billion monthly visits as of August 2026, up about 388% year over year, per Similarweb.
OpenAI's ChatGPT remains the dominant conversational agent by a wide margin. It crossed 1 billion weekly active users by August 2026, up from 900 million in February and 800 million the previous autumn, and now processes upward of 2.5 billion queries a day. Exactly how much of the broader AI referral traffic market ChatGPT holds depends heavily on the methodology used, with published estimates ranging from roughly half to over 80%. What isn't in dispute is that it's the largest single destination for AI-driven research by a comfortable margin. Anthropic's Claude, by comparison, held around 20 million users and roughly 2% of the consumer chatbot market as of January 2026, though its share in specific categories like AI-assisted coding has been climbing since.
Perplexity has positioned itself deliberately differently: not a general chatbot, but an "answer engine." Based on the one figure the company has actually disclosed (780 million monthly queries in May 2025), industry estimates put it at roughly 1.2 to 1.5 billion monthly queries by mid-2026. It reached USD 450 million in annualized revenue in March 2026 per the Financial Times, and is valued at approximately USD 20 billion. Worth noting: Perplexity tested an advertising product in late 2025 ("Sponsored Follow-Up Questions") and walked away from it within months, after advertisers pushed back on the lack of standard conversion tracking and unproven return on ad spend. It has since leaned entirely into subscriptions and enterprise API revenue instead.
How Adoption Differs by Country
United States
The US remains the dominant capital market for AI, capturing somewhere between three-quarters and the vast majority of global AI venture funding depending on the period measured. Enterprise adoption is commonly cited in the high-70s to high-80s percent range across recent surveys. AI websites drew more than 26.9 billion visits from US users in 2025 alone, more than India, Brazil and Germany's totals combined.
China
China's AI ecosystem is largely self-contained, operating through domestic platforms rather than the Western tools discussed above. Public trust runs unusually high: 83% of the Chinese population believes AI brings more benefit than harm, compared with 40% in Canada, per Ipsos data cited in Stanford's AI Index. China also accounts for close to 70% of generative-AI-specific patent families filed worldwide between 2014 and 2023, according to the World Intellectual Property Organization, a genuinely dominant position specifically in GenAI patenting (its share of all AI patents generally is lower).
The domestic platforms themselves are large and, in some cases, volatile: DeepSeek holds roughly 130-140 million monthly active users as of mid-2026, ByteDance's Doubao has swung anywhere from around 150 million to over 340 million MAU across 2026 depending on the month measured (a paywall change triggered a real user backlash at one point), and Baidu's Ernie Bot has passed 300 million users. A structural difference worth understanding: Chinese AI platforms lean much more heavily on platform-native content, long-form WeChat articles, Douyin video transcripts, Xiaohongshu reviews, Baidu Baike entries, than on the open web, which changes what "being findable" actually means for a business operating there.
India
India's adoption numbers get cited inconsistently because they mix two different things: IBM's 2023 Global AI Adoption Index found 59% of large Indian organizations were actively deploying AI, an enterprise figure, not an individual one, and more recent 2025 research suggests the share of individual professionals using AI tools weekly has climbed well past that since. What's genuinely distinctive about India's ecosystem is the voice-first, vernacular-language push: with the large majority of Indian internet users more comfortable in a regional language than in typed English, three homegrown efforts are building around that reality rather than around text search. Sarvam AI, valued at USD 1.5 billion following a 2026 round backed by HCLTech and Peak XV, builds voice models (including Sarvam-105B and Shuka v1) alongside Saaras (speech-to-text) and Bulbul (text-to-speech) APIs tuned for sub-500-millisecond voice latency and fluent "Hinglish" handling. Krutrim AI, built by Ola, became India's first AI unicorn in January 2024. And Bhashini, run by India's Ministry of Electronics and Information Technology, provides free, open-source language APIs across all 22 official scheduled languages.
Europe, UAE and Singapore
Eurostat's 2025 data puts AI adoption among EU enterprises at 20%, up roughly 6.5 percentage points from the year before, but that average hides a large gap: 55.03% of large enterprises have adopted AI compared with just 17.00% of small businesses, a 38-point split. Most non-adopting EU businesses point to a lack of in-house technical expertise as the main barrier, not lack of interest. In the Gulf, roughly 63-64% of the UAE's working-age population report actively using AI tools, per Microsoft's AI Economy Institute (a population figure, not a corporate-adoption rate). Singapore's company-level AI usage sits at just under half as of 2025, up from around 40% the year before, with the government's National AI Strategy 2.0 continuing to drive investment through high-profile 2026 partnerships with Google and OpenAI.
Where the Disruption Is Sharpest
44% of enterprises reported scaling AI across their operations in 2026, up from 38% the year before, according to McKinsey's global State of AI survey. In B2B software specifically, 94% of buyers used a generative AI tool somewhere in their 2025 purchasing process, per Forrester, a big enough number that it's no longer a question of whether AI shapes B2B buying, only how much of the early, unseen research phase now happens inside a chatbot before a salesperson is ever contacted. Advertising economics have shifted alongside it: 2026 benchmarks for B2B SaaS put Meta CPMs between USD 8 and USD 22, CPCs between USD 1.50 and USD 4.20, and cost-per-sales-qualified-lead anywhere from USD 400 to USD 2,400, with typical 180-day return on ad spend in the 1.8x to 4.5x range.
Healthcare is the sector where accuracy requirements are least forgiving. Published studies on AI symptom-checker accuracy vary enormously, some report strong agreement with clinical guidelines on common conditions, others far lower, which is exactly why generic, unattributed health content is being squeezed out of AI-generated health answers in favor of clearly credentialed clinical sources. There's no shortcut around real medical authority in this category.
Retail tells a more encouraging story for AI-referred traffic. 56% of US consumers used generative AI for product research during the 2025 holiday season, up from just 11% the year before, per Synchrony. AI-referred traffic is still a small slice of total web traffic, commonly measured around 1%, but it converts dramatically better: 31-42% higher conversion rates than non-AI traffic, a 254% year-over-year jump in revenue per visit, roughly 48% more time on-site, and a meaningfully lower bounce rate. By the time someone clicks through from an AI answer to an actual storefront, their intent is usually already crystallized.
The Legal Reckoning
The most consequential fight over how these models get built is playing out in The New York Times Co. v. Microsoft Corp. and OpenAI, filed in late 2023. Court filings unsealed in September 2026 revealed that OpenAI's mid-training datasets contained more than 91,692 copies of New York Times, Daily News and Center for Investigative Reporting content, and that a Common Crawl-derived training set included over 2 million documents from nytimes.com alone. The same filings showed Microsoft's own internal data: Copilot's click-through rate to nytimes.com dropped by as much as 93% compared with traditional Bing search.
Those words are Brent Hecht, Microsoft's Director of Applied Science, writing in a January 2023 internal memo about AI training practices, not a plaintiff's lawyer. Separately, Nick Turley, who leads ChatGPT at OpenAI, described the products built this way as posing an "existential threat" to publishers. Both statements came from the companies' own people, which is what makes them significant: this isn't outside criticism, it's an internal acknowledgment of exactly the substitution effect the lawsuit is arguing about.
In response, publishers have split into two camps: blocking AI crawlers outright via robots.txt, or licensing. OpenAI has signed a five-year, USD 250 million agreement with News Corp, a reported USD 60-70 million-a-year deal with Reddit, and confirmed multi-year licensing partnerships with Axel Springer and Condé Nast, though neither of those last two has disclosed financial terms publicly.
Why the Click Is Disappearing
By 2025, 58.5% of US Google searches and 59.7% of European Google searches ended without the user clicking any external result, rising to somewhere around 75-77% on mobile. When an AI Overview actually appears on the page, the zero-click rate reaches 83%, and organic click-through to standard blue links drops by 61%, according to a Seer Interactive study from September 2025. Gartner has forecast that traditional organic search traffic volume could fall by as much as 50% by 2028 as AI agents intercept more queries before they ever reach a brand's own website.
Underneath that shift is a psychological one. Researchers have begun validating dedicated trust scales for human-AI interaction (sometimes referred to in the literature as HAITS), built on the idea that people don't evaluate an AI's answer purely on accuracy: they also respond to tone, consistency and what looks like understanding. The healthiest outcome isn't maximum trust in every AI answer, it's calibrated trust, where people rely on a system in proportion to how reliable it actually is. One of the clearest, most consistent findings in this research is trust transfer: when an AI response cites an institution someone already trusts, a known hospital, university or news brand, that existing trust carries over to the AI's answer itself. Which is a fairly direct explanation for why third-party citations now matter as much as they do.
Generative Engine Optimization: What Actually Moves the Needle
The most rigorous evidence on what makes AI systems actually cite a source comes from a 2024 paper out of Princeton and IIT Delhi, presented at the ACM SIGKDD conference, that built a 10,000-query benchmark specifically to test this. Its headline finding: GEO techniques can lift a page's visibility inside generative engine answers by up to 40%.
Three specific tactics did the heavy lifting, each landing in roughly the same 30-40% lift range on the paper's visibility metric: adding direct, attributed quotes from named experts; adding specific, sourced statistics in place of vague claims; and adding rigorous inline citations to primary sources. Keyword stuffing, the old SEO habit, showed little to no improvement under the same test. The mechanism makes sense once you see it: LLMs are pattern-matching against what a trustworthy, well-sourced passage looks like, closer to Wikipedia's citation style than to a keyword-optimized landing page.
The category built around this finding is growing fast in its own right: the GEO market was valued at roughly USD 848 million in 2025 and is projected to reach USD 33.7 billion by 2034, a 50.5% compound annual growth rate, per Dimension Market Research.
What This Means If You Run a Business
Put the pieces together and the pattern is consistent across every section above: AI search is no longer emerging, it's the default starting point for a large and growing share of how people research anything, and it rewards specific, sourced, well-attributed content over keyword density or accumulated domain authority. That's genuinely good news for a business willing to do the work properly. It also means the businesses that keep treating this as "SEO plus a chatbot footnote" are going to keep losing visibility to competitors who don't.
This is the kind of landscape SumukhAI was built to track and act on, across search visibility, AI visibility and reputation, not as three separate line items, but as one connected picture of whether your business is actually findable, understood and trusted wherever people are looking now.
Want to know how your own business shows up across this landscape right now?
Get Your Visibility Audit →Related reading: The Answer Box Has Fewer Seats Than You Think