Internet Is Getting Remade For AI. What Does It Mean For You?

Internet Is Getting Remade For AI. What Does It Mean For You?

 Chandrima.Banerjee@timesofindia.com 16.02.2026

Less than half the people on the internet are “people” — only about 44% of online traffic came from humans in 2025 — but even within the traffic driven by relentless bots “using” the internet, a small but significant share of 4% belongs to AI bots. If that share keeps growing (and it’s really likely that it will because of how much AI companies are pouring into agentic AI), most websites will eventually be built for AI and not us. Not in the conspiracy-heavy “dead internet theory” way but in the codeand-structures-tech-and-science way. 

THE WEB IS NOT BUILT TO MAKE THINGS EASY FOR AI … YET 

When an AI browser was launched a while ago, I was testing the agentic mode (in which AI takes over your browser to “do” all the work). I wanted it to find available slots for driving licence renewal. But when I checked back after a few minutes, I found that the agent was stuck. The page had a huge popup covering nearly the entire window, and the AI didn’t know what it was supposed to do. The buttons and menus it needed to access were behind the popup — but how would it get to them? To us, it seems easy enough. Shut the popup, and move on. But behind the scenes, a click is a series of tiny tasks — hover, pointer move, mouse down, mouse up, the click itself. Websites can react to any of these steps, or only if these steps happen in the right order.

An AI agent has to do all that, in the correct sequence, and with the right timing. If the page happens to shift mid-click — like when a popup appears — the click can miss or just do nothing. Also, for AI, the decision to close the popup or interact with it has to be based on some kind of logic. Does it know what’s behind the popup? What if engaging with the popup is an important step? What if the popup is the next step? This kind of logic is easier for AI to navigate if the website has an API that AI agents can use. (An API, Application Programming Interface, is a set of rules and definitions that software components use to talk to each other.)

 When an AI agent  uses an API to get your work done, it doesn’t have to bypass all the garrulous persuasion that populates most websites today. Instead of navigating pages built with visual and contextual cues meant for human eyes, it can ask the site directly what it needs — like “show me the available slots” — and get back a clean, structured answer on which it can act for you. A survey of developers in 2025 found that 24% are already designing APIs for AI agents. But every API is different, with its own little quirks. And an AI agent can’t possibly learn every one of them. So, Anthropic came up with Model Context Protocol, an open protocol for AI agents to coordinate their conversations with services and sites and apps. It’s now the frontrunner for becoming the “USB-C port for AI applications” .

WHAT HAPPENS WHEN AI IS THE AUDIENCE? Deloitte estimates that AI platforms drive 6.5% of organic traffic already, and it’s expected to go up to 14.5% within a year. As this happens, AI will be “prioritizing semantic richness over keywords, author expertise over backlinks, and being cited in AI responses over page views.” In plain words, there’ll be less room for froth. More and more “research” already happens inside AI summaries and chats, and they don’t lead to clicks. Also, as Parag Agrawal, former Twitter CEO and now the founder of AI startup Parallel Web Systems, told The Economist , the web was built for humans to read at human speed — “agents face no such limits”. Which means that, over time, we will need more useful information online, certainly not less. But the way things stand now, there is a mismatch between what AI takes and what it gives back to those putting out that information. Over the past year, for every visit OpenAI sent to a website, its bots crawled about 1,100 pages. For Anthropic, the ratio was one visit for about 53,500 webpages crawled. If users don’t click on pages, the goal for anyone with a website becomes being cited, summarised, or used as a canonical source. And money will be made from each crawl instead of each view. Cloudflare has already begun a pay-per-crawl marketplace that lets site owners allow, block, or charge AI crawlers per request. So, more information-dense sites survive. Which, in a roundabout way, might just restore the internet to what it was supposed to be — a place with actual answers. The ‘click’ is fading away 

About 60% of searches end without the person ever reaching a destination site — they simply get their answers on the search page without a click, research by the consulting firm Bain & Company found. But searches at least provide a list of pages that might have the answer. AI would whittle it down even more. Bain’s survey also found that about 80% of search users rely on AI summaries at least 40% of the time. And a Pew Research Center analysis found that only 1% of users who came across AI summaries clicked on the links inside AI summaries.