You’ve noticed it, but maybe you’ve not thought too hard about it. Over the past few months, your Google search habits have changed. What was once mostly call and response – query, results, click, repeat – is now more of a conversation. If you’re anything like me, you’ve started typing longer and sometimes less precise questions into Google, because you know they’ll get picked up in “AI Mode.” Once you’re in that chatbot interface, you’ve probably found Google’s gotten pretty good at answering your questions and helping you accomplish your goals.*
You’re not alone. Google integrated AI Mode into search less than four months ago. Almost overnight, more than a billion people started using the new service. Put another way, with one update to its core search service, Google lapped OpenAI and instantly became the largest consumer AI application in the world. According to it own estimates, Google’s AI Mode queries are now doubling each quarter.
As Claude’s AI might write: “That is not linear growth. That is exponential.” (I put that in quotes because the construction is so dreadfully AI-ish, I didn’t want it flagged as slop. Also, it was a joke. But having to explain it over the course of three sentences inside a parenthetical perhaps is too long a walk down too short a pier…)
In any case, today I want to think out loud about Google’s pivot to an AI-driven interface. When it happened back in May, my first take was critical: “Google has enclosed the entire world wide web and turned it into a walled garden which it alone can monetize,” I wrote. “The web is no longer a destination, it is merely ‘raw material.'” I then asked the obvious next question: “Is this sustainable? Is it good for a free and open society that demands quality information to thrive? It certainly doesn’t feel that way to me.”
But four months into the transition, I have to give Google credit. Yes, I still think Google has enclosed the web and built an Instagram-like container around search. But what choice did it have? Facing the existential threat of an upstart competitor who, at the time, was unconstrained by silly things like profitability, Google found a way to integrate AI not only its core product, but its core business model as well. Just two months after releasing AI Mode into the wild, Google announced a more than 200 percent beat on earnings. I might not like it, but Google’s strategy is working, at least on Wall Street.
But isn’t AI a ridiculously expensive product to deliver? This is where Google’s heritage as a super-scaler comes into focus. 18 months ago I worried that integrating AI directly into Google’s search business would prove impossible due to cost issues. I couldn’t imagine Google offering a free version of Claude or ChatGPT alongside its main search product – those companies were losing billions of dollars a month, and they had just a fraction of Google’s usage.
But I underestimated Google’s infrastructure and engineering chops. AI Mode is driven by Google’s Flash models, which are optimized for rapid web-browsing tasks and cost orders of magnitude less to run than frontier models like Claude or ChatGPT. They are, in the words of an AI Mode summary, “designed to look at a query, scan the web, and give you an answer in seconds.” Through Flash, Google has found a way to build a better mousetrap, once that not only scales to Google’s user base, it also continues to make gobs of money for the mother ship. AI Mode uses exactly the same business model as traditional search, and clearly, it’s working.
If I were in charge of OpenAI’s advertising business, I’d probably quit, because I’m not sure how I’d parry Google’s obvious success in navigating the AI transition. And given all the other pressures impairing OpenAI’s death march toward the public markets, its advertising business has to work. It’s off to a decent start, but can it lap Google? That’s a tall order.
*I’ve heard that this is less true with certain kinds of commercial and navigational queries, and would love to hear your thoughts on the topic…
—