Who controls the timeline

The AI capabilities that affect businesses have arrived through two distinct mechanisms. The first was consumer adoption. The second has been deployment into products consumers already use. Forecasting depends on which mechanism you watch.

April 2026

The adoption phase

ChatGPT launched on November 30, 2022. It reached 1 million users in five days and 100 million monthly active users in roughly two months, faster than any consumer application in history (TikTok took nine months; Instagram took two and a half years). By February 2026, ChatGPT had 900 million weekly active users and 50 million paying subscribers (TechCrunch, Feb 2026).

Perplexity, which launched in August 2022, grew to 45 million monthly active users by late 2025 (Backlinko). These products were adopted by users who chose to visit a new website or download a new app. The growth was demand-driven.

The deployment phase

Once hundreds of millions of people were already using these products, the companies that controlled them started adding new capabilities that changed how those products interact with the web.

Google AI Overviews shipped as a default in Google Search on May 14, 2024 (Google blog). No opt-in. No user-facing opt-out. ChatGPT web search was added as automatic behavior on October 31, 2024 for paid users (OpenAI) and February 5, 2025 for all users (gHacks); ChatGPT now searches the web when a prompt "might benefit from information on the web." Apple Intelligence arrived on October 28, 2024 as a software update to existing iPhones (Apple Newsroom), adding AI features to Mail, Messages, Photos, and Siri. Google-Agent was added to Google's fetcher documentation on March 20, 2026 (developers.google.com), enabling agents to browse websites on behalf of users, bypassing robots.txt.

None of these required users to adopt a new product. Google AI Overviews appeared at the top of search results for everyone; NPR reported there is no official user-level toggle to disable them. ChatGPT began automatically searching the web without users toggling a setting. Apple Intelligence arrived as a software update. Google-Agent started browsing websites without websites opting in.

Google AI Overviews went from nonexistent to reaching 1 billion monthly users in five months (Q3 2024 earnings) and 2 billion by July 2025 (Q2 2025 earnings). That is not a growth curve. It is the reach of a product that already had the users.

The counter-example

Microsoft tried the same approach and it mostly did not work. Microsoft integrated Copilot directly into Windows, Edge, and Office, products used by roughly 450 million commercial users. After two years, only 15 million had paid Copilot seats, about 3.3% of the installed base. Copilot's share among paid AI subscribers dropped from 18.8% to 11.5% between July 2025 and January 2026 (Stackmatix). Microsoft's CEO acknowledged that the integrations "don't really work" (PPC Land).

The difference: Google AI Overviews succeeded as a deployment because users cannot avoid it. It appears in search results whether you want it or not. Copilot was integrated into Office but users could ignore it, and most did. Deployment into an existing product controls when a capability reaches users, but it does not guarantee that users will engage with it. The capability also has to be useful enough, or unavoidable enough, that it changes behavior.

The effects are already measurable

The web traffic consequences of these deployment decisions are visible now. Seer Interactive tracked 3,119 queries across 42 organizations and found organic CTR dropped 61% for queries where AI Overviews appeared (Sep 2025). Ahrefs found AI Overviews reduced position-one clicks by 58% across 300,000 keywords (Dec 2025). These effects followed directly from deployment decisions by Google. No user opted in. No user switched products. The results page changed, and clicks declined.

What we observe

The capabilities that affect businesses, specifically AI answering questions that used to drive website traffic, AI agents interacting with websites directly, and AI intermediating purchase research, have so far arrived through deployment decisions at Google, OpenAI, Apple, and Anthropic. Those companies collectively control products already installed on billions of devices.

Google also controls Chrome, which holds roughly two-thirds of global browser market share. A new browser API, a default-on feature, or a protocol shipped into Chrome reaches billions of devices without anyone installing anything. That is a deployment surface larger than any single AI product.

The question "when will consumers adopt AI agents?" may be less predictive than "when will the next capability ship as a default feature in a product that already has the users?" The track record so far is that each deployment compressed the timeline by months or years compared to what an adoption curve would have predicted. But Microsoft Copilot shows that deployment alone is not sufficient. The capability has to work well enough, or be unavoidable enough, to change behavior at scale.

Agent-mediated commerce has yet to mature. The agents produce errors. The interaction layer between agents and websites is fragile. But the mechanism by which it will reach users, when it does, is more likely to be a product update than an adoption curve.

If you want to understand how AI agents interact with your site today, see how AI systems actually browse the web. If you want help detecting what's already visiting, talk to us.