There is a fatal contradiction at the heart of Google’s AI strategy that has been niggling away at me for some time. Its executives have not adequately addressed it, and the market seems oblivious to it. Yet as a news publisher in my day job, it’s one I’d love someone with a bigger brain than me to explain.
Google’s parent Alphabet plans to splash between $175 billion and $185 billion on capex in 2026, roughly double the $91 billion it spent in 2025 and more than triple the $52.5 billion it committed as recently as 2024. CEO Sundar Pichai has framed this almost as an existential necessity. The risk of underinvesting, he told analysts last year, is greater than the risk of overinvesting.
And yet the product those billions are funding is destroying the ecosystem on which Google’s entire model depends: ad revenue.
The numbers publishers don’t want to publish
Google search traffic to publishers declined globally by a third in the year to November 2025, according to Chartbeat data published in the Reuters Institute’s Journalism and Technology Trends and Predictions 2026 report. Referrals from Google Discover fell 21% over the same period.
Zero-click searches now account for 60% of all queries. AI Overviews, which appear in 13% of searches, cut click-through rates nearly in half. The Pew Research Center tracked nearly 69,000 actual Google searches conducted by 900 US adults in March 2025 and found that only 8% of users who encountered an AI Overview clicked on a traditional search result, against 15% when no summary appeared. Less than 1% clicked the links cited inside the AI Overview itself.
The damage is not evenly distributed, but it is widespread. Business Insider saw its organic search traffic fall 55% between April 2022 and April 2025, leading the company to cut 21% of its staff. HuffPost’s desktop and mobile sites lost half of their search referrals over the same period. DMG Media, which publishes the Daily Mail, reported click-through rate drops of up to 89% attributable to AI Overviews. Publishers now forecast search engine traffic down 43% within three years. A fifth expects losses above 75%.
The paradox Google won’t name
Here is the problem no one at Alphabet is fessing up to: AI Overviews are trained on, and summarise, content produced by the very publishers whose businesses they are eliminating.
This is a supply-chain problem. If publishers cannot generate revenue from content, they produce less of it, or less of the expensive kind: original reporting, on-the-ground coverage, specialist analysis. Google’s AI is then left summarising an increasingly thin and degraded web. The training data gets worse. The summaries get worse. Users leave.
Google is spending $185 billion to hollow out its own inputs.
What the financial results conceal
The counterargument from Google and its investors is that the numbers show no sign of distress. Revenues from services rose 14% to $95.9 billion in Q4 2025, driven in part by 17% growth in Google search. The ad business, supposedly at risk of cannibalisation, keeps growing.
This is accurate. But it may also be a lagging indicator rather than a clean bill of health.
Google’s ad revenue does not require every search to generate a click to a publisher. It requires users to stay within the Google ecosystem long enough to see a commercial result. Alphabet acknowledged the potential impact of AI on its core advertising business in its 2025 annual report, raising the possibility of excess capacity from its infrastructure commitments. It held a $25 billion bond sale in November. Its long-term debt quadrupled in 2025 to $46.5 billion.
In essence, the company is borrowing against a future it cannot fully price.
The bear case for Alphabet centres on cannibalisation: Gemini-powered answers reducing clicks on traditional search ads and pressuring the company’s highest-margin revenue stream. That has not happened yet. Whether it does depends on how far Google expands AI Overviews into commercial and transactional queries, which is where its ad business lives.
The antitrust shadow
There is a further complication that Google seems intent on minimising.
In 2025, Alphabet was found guilty of operating both an illegal search monopoly and an illegal publisher ad tech monopoly. The search case ended without a breakup order. The judge softened the remedies, noting that AI’s advance had changed the course of the case. The proliferation of AI competitors provided Google with its most effective legal defence.
The DOJ appealed on 4 February 2026, arguing that behavioural remedies amounted to a slap on the wrist for a repeat offender. A separate ruling on Google’s ad tech practices, which could require it to sell its ad exchange, is expected imminently.
The News Media Alliance’s response to the search remedies ruling was precise: publishers are forced to give away content to remain in search, a no-win scenario. If you opt your content out of Google’s indexing to prevent it being scraped into AI Overviews, you disappear from search. If you stay in, you subsidise the feature, cannibalising your business.
What publishers can actually do
There are options, but none of them restores what is being lost.
The first is to become hard to summarise. Content that AI cannot compress into a tidy paragraph holds more of its traffic value. Live coverage, original investigation, first-person reporting, local knowledge, specialist data, and continuously updated information all sit in this category. Publishers in the Reuters Institute survey pointed to original investigations and on-the-ground reporting as their most important focus going forward, followed by contextual analysis.
The second is to get cited rather than bypassed. Research shows that brands cited inside AI Overviews earn 35% more organic clicks and 91% more paid clicks than those that are not. Structuring content to appear in AI responses, now called Generative Engine Optimisation, is a legitimate and growing strategic priority.
The third is to reduce Google dependency. Future plc, whose titles include TechRadar and Marie Claire, reports that only 27% of its sessions originate from Google search. The company calls its strategy Google Zero, built around direct audience engagement through apps, email, subscriptions, and social channels. More publishers will arrive at a version of this model, by choice or by necessity.
The fourth is legal. Chegg’s antitrust lawsuit against Google, which alleges the company used publisher content to train AI systems that now compete with those publishers, is the first time this argument has been litigated directly. It will not be the last.
The question Google has not answered
Google’s position is that AI Overviews serve users, that search revenue is growing, and that the billions being deployed will generate returns through Cloud, subscriptions, and a more defensible ad franchise. That argument is coherent. It may even prove correct in the medium term.
But it skips a structural step. The web Google is summarising was built by publishers operating on an economic model Google is now dismantling. The question is whether the content ecosystem it depends on can survive that. And if it cannot, what exactly is Google’s AI left to say?