Notes on Google's Attempt at Self-Disruption
Can Google sustain itself this way, as it makes a very high stakes bet on transforming itself from gateway to gatekeeper?…
Can Google sustain itself this way, as it makes a very high stakes bet on transforming itself from gateway to gatekeeper?…
This is, at some level, gonzo:
Rich Holmes: Google is destroying its own business - and still winning <https://departmentofproduct.substack.com/p/google-is-destroying-its-own-business>: ‘Google delivered an impressive earnings call that told a curious tale of a company that is both destroying itself and winning at the same time…. Google users who encounter an AI summary are less likely to click on links than those who don’t. Just 8% of users who saw a page with an AI summary clicked on a link vs 15% who didn’t. 26% of users ended their browser session after seeing the AI summary vs just 16% who saw pages without the summary.… [And yet] search revenue hit a record $54.2 billion in Q2, up 12% year-over-year…
Ben Thompson offers a vey positive analysis for the future of Google-as-a-money-tree, concluding that perhaps its opportunities to turn MAMLMs into a sustaining innovation are braked primarily by its own caution:
Ben Thompson: Rumors of Google’s Demise… <https://stratechery.com/2025/rumors-of-googles-demise/>: ‘I’ve repeatedly laid out the theoretical case for why AI is potentially disruptive to Google, starting with 2023’s AI and the Big Five. It remains something to monitor. Once again, however, I have to come to Google’s defense: all available metrics suggest that the company is doing quite well. Indeed, I would go further: you can make the case that the company’s biggest mistake is not going harder!… Pichai and Chief Business Officer Philipp Schindler worked extremely hard to not answer questions about search volume and especially search click-through rates, which was telling in its own right…. Pay clicks were up 4% and search revenue was up 12% makes it clear that Search growth is primarily being driven by higher prices…
And Alistair Barr spotlights exactly how big is Google’s unexpectedly aggressive investment in AI infrastructure:
Alistair Barr: Tech Memo July 25, 2025 <https://l.businessinsider.com/s/vb/bS9v95B>: ‘Big Tech is in an AI arms race, each company trying to outspend the others on data centers, GPUs, networking gear, and talent. Engineers can be let go. But the infrastructure? That’s permanent. If the AGI dream fades, you’re stuck with massive, costly assets. So when Google announced it would hike capex by $10 billion to $85 billion in 2025 eyebrows went up.… It’s no shock when Elon, Zuck, and Sam flex on capex. But Google? That’s surprising…. Will these swelling bets pay off?… Since May, Google’s monthly token processing (the currency of generative AI) has doubled from 480 trillion to nearly a quadrillion. Search grew 12% in Q2, beating forecasts. Cloud sales surged 32%. CEO Sundar Pichai said Google is ramping up capex to support all this growth. But it’s still a huge gamble…
Huh…
What do I make of this?
Here is an attempt at a take:
Google’s decision to incorporate AI-generated summaries directly into its search results represents a profound shift. Google had functioned as a gateway, directing users outward to a vast constellation of third-party sites, via its blue links. Now this is upended, as outward traffic is halved, with implications for publishers, e-commerce platforms, and information providers who have built their business models around the expectation of Google-driven traffic. They are unsustainable on their old patterns should the AI-summary become the destination and not just the gatekeeper.
In the pre-AI era, the company’s dominance was already formidable; most online journeys began with a query in the search box. But by answering more questions directly—whether through featured snippets, knowledge panels, or now AI summaries—Google transitions from being a mere index to an oracle. This is history rhyming with how the social networks, by keeping users engaged within their platforms, siphoned attention away from the broader web, to our great detriment. AI summaries perform a similar enclosure, but with the added imprimatur of algorithmic authority.
And yet Google is now claiming that, so far, even as the total volume of outbound clicks declines, the value of each remaining click rises. Advertisers, desperate for access to the dwindling pool of users who actually leave Google’s domain, find themselves bidding up the price. For Google, this would be a masterstroke—extracting greater rents from advertisers while providing users with what appears to be a more efficient search and knowledge acquisition experience.
The missing piece: When users encounter an AI-generated summary on Google and subsequently click through to an advertiser’s website, what do they then do? Their post-click behavior becomes a crucial determinant of value for advertisers. Are these users are simply verifying the AI summary’s accuracy before bouncing? Or does the AI summary act as a primer, sending only the most motivated, high-quality traffic onward—users primed to engage, convert, or purchase? This dynamic has analogues in the past history of digital advertising: consider the shift from indiscriminate banner ads to targeted, intent-driven search ads in the early 2000s, which dramatically increased both the value-per-click and the efficacy of online marketing.
Plus the AI summary may change user expectations and behaviors in ways that are difficult to anticipate or measure.
Back up: Google’s fear is being disrupted by a firm with a MAMLM that stands between individuals and the web as a whole—that serves as an intermediary that crawls the web once, and then does RLHF and RAG to be a better summarization and brainstorming agent than an individual checking on Google search-result blue links can.
This MAMLM would not simply index the web and provide links, but would synthesize, summarize, and even brainstorm on demand, offering users answers and insights far richer than what can be gleaned from clicking through a list of blue links. In this world, the user’s primary and monetizable relationship is not with the open web, nor even with Google, but with the AI intermediary itself—a shift that would render Google’s core business model, and the web’s traditional architecture of discovery and monetization, increasingly obsolete.
Google’s response: if anyone is going to disrupt our business by doing this, it is going to be us. We will spend whatever it takes to pre-empt such disintermediation by directly building and deploying our own AI summarization and answer-generation capabilities directly into our flagship search product. We will thus cannibalize its own legacy business and simultaneously define the next paradigm—one in which the search page itself becomes the user’s destination. One thinks of IBM’s embrace of the PC, or Apple’s willingness to let the iPhone kill the iPod. But the stakes here are higher, and the feedback loops more treacherous. For Google, the logic is as brutal as it is clear: better to risk undermining the golden goose than to have it stolen outright by a new breed of algorithmic upstart.
The vague outlines of the developing situation are, so far, confused:
Perhaps the new regime promises greater value for users, as the MAMLM-as-summarization-engine does its work: instead of wading through a swamp of SEO-chaff and clickbait, the user is presented with a concise, context-aware answer—sometimes even a synthetic argument or brainstorm—without ever needing to click away from the search results page. For the harried, time-constrained knowledge worker, this is a godsend. Imagine a world in which one confronts not a list of blue links, but a cogent synthesis, drawing on JSTOR, Wikipedia, and the latest Substack polemic, all in one go. Frictionless knowledge, delivered at the speed of thought, with the AI acting as both librarian and tutor.
At the same time, Google claims it is seeing “equal value” from monetizing AI-summaries as from monetizing search, on its side of the ledger, at least so far. Even as the volume of outbound clicks falls, advertisers are bidding up the price of each click that remains. Is the theory that these are now higher-intent, more motivated users, pre-filtered by the AI? Are they just so starved for traffic that their willingness to pay is great? Trading quantity for quality, or exerting monopoly power? So far, their auction model is holding up—though one wonders how long that equilibrium can last, and what happens if advertisers start to see diminishing returns.
But the losers in this brave new world are the websites.
Less traffic flows to the underlying sources, as users are increasingly satisfied by the AI’s answer and see little need to click through to the original. For the vast ecosystem of publishers, bloggers, and even e-commerce sites that have built their business models on the back of Google-driven traffic, this is an existential threat. The web risks becoming a substrate for AI extraction, rather than a living network of destinations. And then we get a new form of SEO, dedicated to maximizing AI-slop.
Plus: While the promise of greater value for users is seductive, but. The MAMLM-as-summarization-engine promises to collapse the search-and-scan process into a single, authoritative-seeming paragraph. This promises, for the user, a productivity revolution: research that once took hours now happens in seconds. But there is a price: obscured nuance, flattened debate, and consensus prioritized over dissent. Moreover, there is the loss of “productive friction”—the serendipity and discovery that comes from wandering through primary sources. Roaming through the stacks of Widener Library, the book you wanted was not the one you had the classification code for but the one three to its left—something you could only learn with a stack pass. The value proposition for users is a Faustian bargain, and we have seen such bargains before.
There are two roads:
First, if “AI” intermediaries are not the future—if it turns out that the grand experiment with AI intermediaries—MAMLMs, summarizers, chatbots, whatever you want to call them—amounts to little more than a high-tech cul-de-sac, Google will have burned through eye-watering sums of capital for what, in retrospect, will look like an overreaction to a phantom threat. The company will have spent nine figures on data centers, GPUs, and armies of engineers, all to defend against the possibility that its search business would be leapfrogged by a new digital oracle. Google’s expenditures will have functioned as a kind of “strategic insurance”—a hedge against a disruption that never materialized. But the sunk costs are not entirely wasted. Google, at the very least, will have acquired technical capabilities, organizational muscle, and a deeper understanding of the AI landscape—assets that can perhaps be redeployed as the technological winds shift.
Second, if “AI” intermediaries are the future of digital knowledge navigation, Google is not only betting the farm, but buying up the neighboring counties as well. The company is investing in this transition at a scale that outmatches all rivals. The bet is that in a winner-take-most environment, scale and speed are everything. By moving first and moving big, Google positions itself to capture whatever monetization opportunities arise. This is so as long as “sufficient quantity has a quality all its own”. History is full of examples: Microsoft’s dominance in operating systems, Amazon’s in logistics, or even Google’s own early days in search, when scale begat quality, and quality begat scale, in a virtuous (for the incumbent) cycle.
It is interesting to watch: As AI-powered summarizers threaten to interpose themselves between users and the web, Google is racing to become its own self-disruptor. If you can trade volume for value, and open discovery for algorithmic enclosure, it will. If it turns out to be a dead end, it will have built enormous infrastructure-layer data-processing tools that can be turned to other purposes. Will it get a healthy ROI on those? No. But the enthusiasm of AI-addled investors is keeping this commitment from causing any financial blowback.
Of course, this strategy risks eroding the web’s ecosystem of content creators, fragmenting user journeys, and triggering regulatory backlash. Advertisers pay more for each precious click, while publishers see their traffic decimated. For users, the promise of instant answers is seductive, but the long-term consequences for information quality, diversity, and innovation remain uncertain.
But Google has decided that these things are not its problem now, but somebody else’s.
With lots of further questions. Here are five:
Will advertisers eventually balk at escalating costs, or will the law of diminishing returns force a reckoning with the current model?
Is there a potential for a counter-movement—perhaps a new “open web” alliance—seeking to reassert the value of outbound linking and decentralized discovery?
Could publishers respond by designing landing pages specifically for “AI-primed” visitors—pages that acknowledge the summary, offer deeper context, or present unique value not captured in the initial answer?
Might we see a bifurcation of the web: one layer optimized for human navigation and engagement, another for AI ingestion and synthesis?
How can users maintain agency, critical thinking, and serendipitous discovery in a world where the MAMLM’s output is tailored, persuasive, and frictionless?
How will Google’s internal culture—famously risk-averse and bureaucratic in its later years—adapt to the demands of continuous, existential self-cannibalization?
References:
Barr, Alistair. 2025. “Tech Memo July 25, 2025.” Business Insider. https://l.businessinsider.com/s/vb/bS9v95B
Heath, Alex. 2025. “Google gets its swag back: Reports of Google’s death at the hands of ChatGPT have been greatly exaggerated.” The Verge, July 25. https://www.theverge.com/command-line-newsletter/713603/google-search-chatgpt-openai-earnings-ai-race-swag
Holmes, Rich. 2025. “Google is destroying its own business - and still winning.” Department of Product (Substack). https://departmentofproduct.substack.com/p/google-is-destroying-is-own-business
Pew Research Center. 2025. “AI and Search: How AI Summaries Impact User Behavior.” Pew Research Center. https://www.pewresearch.org/internet/2025/07/25/ai-and-search-how-ai-summaries-impact-user-behavior/
Shmulik, Mark. 2025. “Google is ‘jumping aboard the AI crazy train’.” Bernstein Research Report. (Accessed via Business Insider). https://l.businessinsider.com/s/vb/bS9v95B
Thompson, Ben. 2023. “AI and the Big Five.” Stratechery. https://stratechery.com/2023/ai-and-the-big-five/
Thompson, Ben. 2025. “Google Earnings, Google Flips the Switch on Cloud, Search Notes.” Stratechery. https://stratechery.com/2025/google-earnings-google-flips-the-switch-on-cloud-search-notes/
Thompson, Ben. 2025. “Rumors of Google’s Demise…” Stratechery. https://stratechery.com/2025/rumors-of-googles-demise/
You know, one can already put a tag in one's website that forbids search engines to crawl it.
One of the staples of the web is teaching/information content. It is a big winner on YouTube, for instance. But the described practice might undermine the attention economy reasons for creating such content.
Perhaps we need a tag that says, "go ahead and index, but you are forbidden to feed this to a LLM or similar AI".
And then I wonder, would this make the web better or worse? Would it make such sites less travelled or more travelled.
I use the web for shopping, yes, but also for generic information about the world. Information to help me program better. Information to help me understand world events better. Sometimes to understand mechanics/strategy in a computer game. And so on.
Sometimes the AI response is helpful, but I can't ever trust it. Not after seeing all the times it makes things up.
This post seems to be missing the known enshittification of Google Search to boost ad revenue. The search was deliberately made worse so that the "best results are to be found on page 1" has been downgraded, so that one needs to dig through more pages. To get more revenue, this implies that just having an ad or paying website on the search page generates revenue for Google. Yes, increasing ad rates boost that further.
But as any business knows, increasing ad costs without a comparable gain just results in advertising elsewhere or through other means. Squeezing businesses and worsening the ad experience is eventually self-destructive.
Despite the increasingly poor quality, I still use Google Search with the AI summaries turned off, but if MAMLMs could deliver what I want, I might end my use of Google Search. However, I don't expect that to happen any time soon.