Somewhere in the last eighteen months, a growth engine that most CEOs never had to think about quietly stopped working, and almost nobody in the boardroom noticed until the pipeline numbers did. For twenty-five years, the deal was simple: build something good, describe it clearly on the internet, rank reasonably well, and a share of the people searching for a solution to their problem would click through and become a lead. That deal is now expiring, not because Google broke it on purpose, but because Google, along with OpenAI, Perplexity, Microsoft and Anthropic, have built something that answers the question before anyone reaches your website at all.
The numbers are no longer ambiguous. As of early 2026, AI Overviews appear on 48 percent of Google searches, up 58 percent year over year, and when they show up, organic click-through rates fall by roughly 47 to 61 percent depending on the study. Position one, the spot every SEO team spent a decade fighting for, now loses more than half its clicks when an AI summary sits above it. Across the web, 58 to 60 percent of searches end without a single click to any external site. In B2B, where purchase decisions are supposedly more deliberate and research-heavy, the damage is worse: average organic traffic is down roughly a third year over year, and organic lead volume has fallen by close to half. None of this is a temporary dip while users adjust to a new interface. Bain’s research suggests 80 percent of consumers now rely on AI-generated answers for a meaningful share of their searches, and a quarter of users stop looking entirely once they’ve seen one.
Executives have seen search algorithms change before and lived to tell the story, so the instinct is to treat this as another Panda or Penguin update: annoying, disruptive to whoever ranked incumbent, but ultimately a problem you out-optimise. That instinct is wrong, and understanding why it’s wrong is the actual strategic issue here, not the traffic dashboard.
The game has changed, not just the rules
Every previous search update rearranged who won inside the same basic structure: a public, crawlable index; a roughly auditable set of ranking signals; a click as the unit of value exchanged between a business and a customer. A determined challenger with a smaller budget but sharper content could still climb past an incumbent, and that possibility was what made SEO a genuine competitive equaliser for two decades of digital business, India’s included, where an entire economy of content agencies, affiliate publishers and D2C brands grew precisely because organic discovery was cheap and comparatively fair.
What is emerging now is structurally different. There is no page two to rank on inside a conversational answer. The model does not show a list of ten links for a user to compare; it selects, synthesises and asserts, and the underlying logic for how it decides what to cite or recommend is neither fully documented nor independently auditable the way PageRank eventually became. Being “cited” inside an AI Overview or a ChatGPT answer carries a real reward, a 35 percent lift in organic click-through and a startling 91 percent lift in paid click-through according to Seer Interactive’s research, and AI-referred visitors convert at more than four times the rate of ordinary organic traffic. But the number of businesses that get to compete for that citation, in any given answer, has shrunk from ten blue links to often three or four names, chosen by a system whose commercial incentives nobody outside the AI labs can fully see.
The natural corporate response, pouring the old SEO budget into what the industry now calls generative engine optimisation, treats the symptom without addressing the deeper shift. GEO remains largely unbundled and immature as a discipline (barely a quarter of agencies offer it as a distinct paid service rather than folding it into existing SEO retainers), and more importantly, it is trying to reverse-engineer trust signals for systems that were never built to be gamed the way search indexes were. You cannot buy a backlink to a language model’s judgment.
When the answer becomes the transaction
The more consequential shift, and the one that should worry a CEO far more than a dip in blog traffic, is that the same AI systems are moving from answering questions to completing transactions. Agentic commerce, where an AI assistant browses, compares and buys on a customer’s behalf through protocols like MCP, is still early in India, but the direction is unmistakable. Swiggy has already built its own MCP integration so customers can order food and groceries directly through ChatGPT, Claude or Gemini without opening the Swiggy app at all. Bigbasket’s leadership talks openly about agentic ordering as the natural evolution for recurring grocery purchases, where a customer no longer even needs to browse a category page, they simply let the agent reorder what they bought last time. This is not a UX tweak. It is the checkout counter moving from a company’s own storefront into infrastructure the company does not own and cannot fully influence.
And here the equity problem sharpens further. As Cashfree’s Nitin Pulyani has observed, agentic commerce integration currently favours large, well-resourced companies with the engineering capacity to build these protocol connections; smaller merchants and challenger brands, the same players who once used SEO to punch above their weight, are largely locked out for now. Bain’s research on B2B buying found that 85 percent of purchases already come from a “day one” consideration list formed before any search even begins. An AI shopping agent, trained on existing brand signals and pre-formed associations, is structurally more likely to reinforce that shortlist than to discover a worthy newcomer. The open web rewarded merit that was legible to a crawler. The agentic answer economy, at least in its current form, rewards merit that was already legible to the model’s training data, which quietly means it rewards incumbency.
What this means at the top of the house
The strategic error available to any CEO right now is to delegate this entirely to the CMO as a channel-mix problem. It isn’t. Whether your company’s products, pricing, reviews and technical documentation are structured in a way a model can parse, trust and cite is now closer to core infrastructure than to marketing content, which means CTOs and CIOs need a seat at that table alongside marketing, not as an afterthought. Boards that already track platform-dependency risk from app stores or ad-network algorithm changes should add AI-answer dependency to that same register, because the concentration risk is arguably worse: a handful of firms now mediate discovery and, increasingly, purchase, for categories that took decades to build genuinely competitive, multi-vendor markets.
There is a narrower, more useful way to think about the opportunity inside this disruption. The funnel is not simply shrinking, it is being replaced by something with sharper intent at a smaller volume: fewer visits, but visits worth roughly four times as much. Companies that win here will stop measuring success in clicks and start measuring “share of model”, their frequency and favourability of citation across the two or three AI systems their customers actually use, treating that metric with the same seriousness that a previous generation of executives gave to share of shelf in a supermarket aisle. The tollbooth on the road to the customer has already gone up. The only real choice left for a CEO is whether the company understands the toll well enough to be waved through, or finds out the hard way, in a quarter’s missed pipeline number, that it didn’t.
