Both ends of the information chain have flipped to AI: more than half of newly published articles are machine-written, and 57.5% of web requests now come from agents and automation. Here is the raw data behind that claim, with sources and caveats.
2026-08-21
Before deciding whether something is worth doing, look at the numbers. Below is the raw data behind our thinking on A2A Marketing, all of it from public third-party research. Every figure carries its source and its methodology. Where a number is contested, we say so rather than reading it one way.
1. The entry point is moving
Portals sent you to websites. Search sent you to keywords. Feeds sent information to you. The fourth shift is different: AI gives the answer, and agents make the call.
AI users (billions)
| Year | Outside China | China | Total |
|---|---|---|---|
| 2023 | 0.18 | / | 0.18 |
| 2024 | 0.30 | 0.25 | 0.55 |
| 2025 | 0.80 | 0.60 | 1.40 |
| 2026 (through May) | 1.00 | 0.60 | 1.60+ |
The non-China series uses ChatGPT as a proxy, since it is the only platform with a clean year-over-year record (OpenAI / DemandSage / TechCrunch / Sensor Tower). The China series uses CNNIC’s generative-AI user count: 249M in its 55th report, 602M in its 57th, a 42.8% penetration rate and 141.7% year-over-year growth.
Three notes on methodology:
- We do not sum platform MAUs. Gemini at 750M, Copilot at 420M, Claude at 56M, and Perplexity at 34M would double-count heavily, since most people use several. The deduplicated figure for standalone AI platforms worldwide is more than 1 billion (DataReportal / Similarweb, October 2025).
- The non-China series mixes metrics: 2023 is MAU, 2024–25 are WAU. Since WAU is lower than MAU, the resulting curve is conservative, not inflated.
- China’s 2026 figure repeats 2025. CNNIC’s mid-year report is not out, and we do not extrapolate.
Global AI penetration (2026)
| Segment | People | Share |
|---|---|---|
| Never used AI | ~6.6B | ~80% |
| Have used / currently use AI | 1.6B+ | ~20% |
| Pay for AI | ~47M | / |
Penetration is still in the 20% band. That is the anchor for anyone asking whether this is early.
2. Production: machines are writing the internet
50%+ of newly published articles are now AI-written. The crossover happened in November 2024, twenty-four months after ChatGPT launched. Source: Graphite, sampling Common Crawl.
The honest framing is “zero to half in two years,” not “multiplying every year.” Graphite’s data shows the share flattening around 50% after mid-2024 rather than continuing to climb.
74.2% of new web pages contain AI-generated content. Purely human-written pages are down to 25.8%. Source: Ahrefs, April 2025, sampling 900,000 new pages.
30 billion AI-generated images have been produced. Photography took 149 years to reach 15 billion. Generative AI took 18 months to reach twice that. Source: Everypixel Journal.
201.7 ZB of data was created and moved worldwide in 2025, roughly 100× the 2010 figure. Source: IDC DataSphere. The metric covers data created, captured, copied, and consumed; genuinely new data is only about 10% of it, and the rest is duplication and consumption.
Global data volume over time (ZB):
| 2010 | 2012 | 2020 | 2023 | 2025 | 2028E | 2030E |
|---|---|---|---|---|---|---|
| 2 | 6.5 | 64 | 132 | 202 | 394 | 700+ |
IDC gives only a threshold for 2030 (“surpass 700 ZB”), not a precise figure, so we write 700+. The 2010 and 2012 points are secondhand citations we could not trace to a specific IDC edition, and we flag them as such.
3. Consumption: agents are doing the asking
57.5% of global HTML page requests now come from agents and automation. In June 2026 they overtook human requests (42.5%) for the first time. Source: Cloudflare Radar, which covers roughly one fifth of the web.
The methodology matters here: this is total automated traffic. AI crawlers are the main driver of growth, with training-oriented crawling rising from 22% to 52% of it, but the number is not “pure AI agent traffic.” The crossover arrived about eighteen months earlier than Cloudflare’s own forecast.
88.6% of Chinese consumers consult AI before making a purchase. Asking AI first is becoming the default step. Source: iResearch, February 2026.
41% of Chinese users have shifted their search behavior almost entirely to AI. Source: iResearch, February 2026.
Visitors referred by AI convert at roughly 4× the rate of search-referred visitors. Source: Similarweb / BrightEdge. AI referrals are about 1.08% of all web traffic today (Similarweb, March 2026): a small base, but a high-quality one.
+778% year-over-year growth in AI referral traffic (2024→2025). Source: BrightEdge.
4. The market is following the entry point
AI marketing market size (USD billions, at ¥6.8/$1)
| Year | China | Rest of world | Total |
|---|---|---|---|
| 2025 | 0.37 | 1.01 | 1.37 |
| 2026E | 1.48 | 4.40 | 5.88 |
| 2027E | 2.15 | 13.20 | 15.35 |
China: Analysys, 2026 China GEO Industry Market Report, on a narrow basis counting service fees only. Rest of world: Intel Market Research, which publishes a 45.5% CAGR; our 2027 figure extrapolates one year forward at that rate and is not a published number.
$2.1 trillion: total global marketing spend, 2026E. That covers advertising ($832B) plus non-advertising marketing ($1.088T). The base is PQ Media’s measured $1.920T for 2025, grown at their forecast rate of 9.8%; the absolute 2026 figure is our own calculation.
AI marketing is 0.3% of that total today. Which is the point: this is not a small new market, it is a multi-trillion-dollar existing one starting to move.
Three caveats belong up front:
- The China and rest-of-world lines come from two firms using two different rulers. Analysys measures the Chinese domestic services market on the ground; Intel Market Research estimates remotely, weighted toward Western tool vendors. The two columns are not directly comparable.
- We checked Valuates, Dimension Market Research, Navistrat, and Marketintelo. None offers a clean China / rest-of-world split. We also found aggregated figures with unclear methodology and implausibly high values (one source puts China’s first half of 2025 at $3.65B), and did not use them.
- Forecasts indicate direction and order of magnitude, nothing more. We bet on the direction, not the slope.
5. What the numbers add up to
Both ends flipped at once. More than half of new content is written by machines, and more than half of web requests are made by them. The doors in and out of information are closing around AI, and more decisions now start with a single AI answer.
AI only remembers and cites the sources it trusts. The more information there is, the scarcer accuracy and consistency become. That is the entire premise behind AI answer visibility (GEO) and A2A Marketing.
For a brand, one action follows: find out how AI describes you right now. See how we work · Book a diagnostic
Concept images on this page are AI-generated.