Nick Fox, the Google senior vice president who runs Search, used an interview published on YouTube on September 25, 2026 to call the company's European compliance changes "the biggest degradation ever" to its search engine. Over 42 minutes recorded in Brussels, he also cast Google as the guardian of Europeans' most private questions, rejected the idea that AI answers are draining clicks from the web, and acknowledged that the search advertising system paying for all of it is being rebuilt.

In Short

Google's head of search said in a video interview that obeying EU competition law has made Google worse in Europe, for example by removing real-time airfares, the option to search for a trip by date and some spam filters. Critics say Google chose to remove those features rather than show rivals fairly, so part of the complaint is about Google's own design choices. For travel firms, shops, publishers and advertisers it matters because what Google shows or hides decides where your clicks, bookings and ad budgets go. The interview changes no rules, but it previews the case Google will make as Brussels keeps enforcing the law, including an order to share search data with rival engines from January 2027.

Who asked the questions, and where

The conversation was posted on September 25, 2026 to the YouTube channel of Pietro Valetto, a University of Antwerp PhD student in economics and social sciences who writes the Popular Economics Substack. It is not a big stage. When the page was captured for this article, the video had 97 views and the channel around 200 subscribers. It was recorded at a Google pop-up venue in Brussels on a date the video does not give. Fox said he was in the city that week partly to press the company's objections to European regulation, which may explain why one of Google's most senior executives spent three-quarters of an hour with a doctoral student.

Valetto's video description lists Google's clash with Brussels over the Digital Markets Act (DMA) among the topics and asks officials at DG COMP or DG CONNECT, the Commission's competition and digital directorates, to get in touch because he wants "to hear the other side of the DMA debate." No Commission voice appears in the video. Fox's claims about European regulation went largely unchallenged on camera, which is why this article tests them against the record.

This article relies on YouTube's automatic captions, which garble several terms - Antwerp appears as "Antworf", AI Overviews as "AIO reviews", Article 6(5) as "DMA 65" - and drop the amount of a Finnish data centre investment the interviewer cites. Quotations have been cleaned of repeated words and verbal fillers.

Fox has run Google's knowledge and information division, covering search, maps, commerce and advertising, since an October 2024 reorganisation. In April he described AI Mode splitting one request into tens, hundreds or even thousands of sub-queries, and in December 2025 he told publishers that optimising for AI search is the same as ordinary SEO while rejecting standardised licensing deals.

"The biggest degradation ever"

"Europe is an incredibly important market for us," Fox said, before setting out two objections to the DMA. The first concerns Article 6(5), which bars a gatekeeper from favouring its own services in ranking - the conduct known as self-preferencing. Google had set out its compliance approach two weeks before the conversation, he said, and it required removing features from Search.

"It's the biggest degradation ever," Fox said. "It literally makes me sad."

His examples were concrete. "You can't find real-time pricing for a flight," he said, and "if you're looking to book a trip, you can't put in the dates". Then came a rhetorical question: "Why does that make sense for anyone to not be able to search by date?" He added a third loss. "There's a lot of websites that try to sort of trick our algorithms," he said. "We've been forced to pull back on spam protections," with lower-quality results in Europe as a consequence.

He also insisted Google is not alone, saying competitors are degrading their products to comply, though he named neither the companies nor the features. "The goal of the DMA should not be for Europeans to have worse products. Europeans deserve the best."

Each claim maps onto a change already on the record. The Commission fined Google 890 million euros on July 23, 2026, 460 million of it for self-preferencing in Search, and the 153-page decision set a 60-day deadline that, counted from adoption, fell on September 21. On September 8 Google switched on a redesigned results page in the European Economic Area. Reuters reported that it places one vertical search service - an online travel agency, a metasearch engine or another vertical listings portal - at the top, two more beneath it with less detail, and a carousel of hotels, airlines and restaurants with no current prices. Fox told Reuters at the time that the redesign would make the experience worse for Europeans. Nine days later, over September 17 and 18, free product listings and popular product carousels vanished from EEA results, with trackers recording declines of roughly 90 to 100 percent in Germany, France, Belgium, Sweden and the Netherlands.

A degradation by design?

Here the argument gets harder for Google. The Commission's decision does not order the removal of any feature. It orders non-discriminatory treatment of rival services in ranking, as PPC Land noted when Kent Walker made the same product-removal case on July 23 and Google said it was weighing an appeal. Competition lawyer Thomas Höppner has argued for more than a year that stripping direct links was a strategic decision by Google rather than a regulatory mandate.

The shopping change shows the pattern. Faced with a rule against ranking its own listings above rivals, Google did not rank its own free listings lower. It removed them from general results and handed the space to comparison shopping services. A compliance strategy built on subtraction produces visible gaps, and the gaps then become evidence that the law makes products worse. Is that an unavoidable consequence of the regulation, or a choice that happens to serve the argument? Fox did not address the question, and Valetto did not put it to him.

Nor did Fox offer a measure. "Biggest degradation ever" arrived without a satisfaction score, a fall in queries or a change in bookings attached. Google's previous evidence came from its own side: a September 2025 consultation response cited a Nextrade survey of 5,000 European consumers, put the cost to European businesses at 114 billion euros and described more than 3,000 engineers and more than 40 compliance solutions devoted to Article 6(5). The complainants have their numbers too. European travel bodies say Google's share of hotel search usage rose from 37 percent in 2013 to 80 percent in 2023, and research by Mirai found hotels in DMA regions receiving 30 percent fewer clicks and 36 percent fewer direct bookings than those elsewhere. Every figure in this dispute comes from an interested party.

The spam argument has unexpected allies

On spam, Fox has more company. Since August 30, 2026, manual actions under Google's site reputation abuse policy no longer affect searches made inside the EEA; affected site sections are instead gradually separated from the host domain's ranking signals. The Commission opened DMA proceedings into the policy in November 2025 after complaints from publishers whose sponsored sections were demoted, and some search practitioners sided with Google at the time. Lars Lofgren, who documented Forbes Marketplace's affiliate operation, accused Brussels of defending large publishers' spam revenue while small sites burned.

"Forced", though, is Fox's word. The EEA carve-out was a change Google itself published on August 28 after discussions with the Commission, with no stated Commission finding and no stated closure of the proceedings. Google chose to settle the point by adjusting enforcement rather than defend the policy to a decision. The operational result is that the same URL can now hold two ranking positions, one for searchers inside the EEA and one for everyone else.

Privacy, argued by the company holding the data

Fox's second objection, to Article 6(11), was delivered with more force. The provision obliges a gatekeeper search engine to give rivals anonymised ranking, query, click and view data on fair, reasonable and non-discriminatory terms. Fox said it is about to cause "a really massive violation of European privacy".

His case rests on intimacy. People bring to search "questions that you don't ask your spouse, you don't ask your friends, you don't ask your doctor", he said, about health and money. "We invest an enormous amount of money in securing our data centers," and once queries leave "the sort of four walls of our data centers", he argued, they are exposed to hacking. The beneficiaries, he added, are "typically not European companies", so the rule "doesn't actually enable European competition in a meaningful way".

The record complicates each strand. Google has been handing anonymised query data to rivals since March 2024, under its own design, and selling it at between 1.50 and 9 euros per thousand queries. That offer found a single licensee. The Commission concluded on July 16, 2026 that Google's anonymisation stripped out between 90 and 100 percent of unique queries, leaving rivals with little of use. The dispute, in other words, is not whether European queries leave Google's walls. They already do. It is how much survives the trip.

The binding scheme is not a free-for-all. It removes records containing entities searched by fewer than 50 distinct users in a year, drops queries above the 95th percentile of length and generalises location until each record blends in with at least 1,000 people. Buyers face audits and may not use the data for advertising. Google has contested even that: a Google letter reported in May 2026 claimed users could be re-identified in under two hours under the Commission's preliminary method, a claim outsiders cannot yet check.

Two further facts sit awkwardly beside Fox's framing. First, the privacy objection is not uniquely European. In the United States, Judge Amit Mehta's September 2, 2025 remedies ruling also required Google to share search index and user-interaction data with qualified competitors, a remedy Google appealed in January 2026. Second, the timing. Four days before the interview appeared, Ireland's Data Protection Commission fined Google 403 million euros over its processing of location data through Web & App Activity, Location History and Location Accuracy between May 2018 and February 2020. Google said the case concerned historical policies it has since updated.

As for nationality, the eligibility rules require recipients to operate in the EEA with services aimed at EEA users and exclude firms controlled by non-EEA state actors or sanctioned entities, but set no European ownership test. The DMA was written to make markets contestable, not to favour European champions - and the argument that beneficiaries are not European sounds odd from a company headquartered in California. Sharing under the binding terms is due to start in January 2027.

"The best products win"

Valetto offered Fox an opening: does the DMA not also help Google, by forcing other gatekeepers to open their platforms? "In theory," Fox allowed, but "we have not seen that play out that way", citing iOS. His explanation for Google's position was simple. "The best products win," he said. "It's the best search engine in the world." People use Google on Android, iPhones, Macs and Windows machines alike, he argued, because "people find their way to Google because they want to use Google".

That is one reading of a market in which, by the Commission's account, Google Search has held more than 90 percent of the EU for decades. The Commission's reading is that scale itself is the barrier: the long-tail query and click data a rival would need exists almost only inside Google, which is precisely why Article 6(11) exists.

Fox then turned to cost. The regime is "more of a burden than the benefits right now", he said, and "DMA reduces innovation", with "the chilling effect that the DMA has on future innovations". His example was personalization, available in the US but not in Europe because doing it there is "much much harder". Google's recent product maps suggest the DMA is not the only filter. Gemini Spark, the company's always-on agent, excludes the EEA, Nigeria, Switzerland and the UK; three of those four jurisdictions sit outside the DMA.

His closing line on Europe paired the two objections. "Europeans have a substantially degraded version of search," he said. "Europeans should be able to search with the knowledge that their private queries will be kept private rather than handed over to someone else."

Keeping AI answers fast and cheap

The most technical passage came in answer to a cost question. Valetto put two figures to Fox: that Google Search now handles a record 15 billion questions a day, and that sending them all through a large language model would cost "hundreds of millions of dollars a day in token spend". The unit in question is tokens. Fox neither confirmed nor disputed either number. Google's last disclosed total, more than 5 trillion searches in 2024, works out at about 13.7 billion a day, so 15 billion would be roughly 9.5 percent higher. Alphabet has since said queries hit an all-time high in the first quarter of 2026, without giving a count.

Fox's answer began with latency. "Speed and cost are actually very correlated," he said. "If you wait too long to give a user a response, they've moved on." Early AI Overviews were slower, and users scrolled past them, which gave Google a reason to cut compute whatever the price. He then listed three techniques.

Model routing comes first. Google runs models from very small to very large and tries to "route queries to the right model" according to difficulty. "We can modulate the amount of thinking effort that goes into a query," he said. A hard mathematics or logic problem goes to a larger model; a question about the population of Brussels most likely goes to a much smaller one.

Caching is the second. Google stores some answers, Fox said, and the method is more complicated than it looks. "If you cache something that is out of date, that's a problem." A cached answer to part of a question can also misfire when the user is asking something else.

The third is vertical integration, "from silicon to data center to the interconnect between the data centers", including Google's long-running investment in tensor processing units (TPUs). "Our infrastructure is built for the speed of search," he said.

He gave no figures: no cost per query, no share of traffic served from cache, no split of queries by model size. The scale shows up elsewhere. On June 3, 2026, Alphabet told investors its models processed 3.2 quadrillion tokens a month, against 9.7 trillion two years earlier, and described eighth-generation TPUs designated 8i and 8t. In July it raised 2026 capital expenditure guidance to 195 billion to 205 billion dollars. Query fan-out multiplies the load, since one AI Mode request can spawn many background searches.

Routing deserves a second look, because it is a quality decision as much as a cost one. Google, not the user, estimates how hard a question is, and a question judged simple gets a cheaper model and less "thinking effort". Mistakes in AI answers carry real costs: in July a Missouri bakery, Anna Mae's, had to deny a closing date that a Google AI Overview had invented. Nothing in the interview, and nothing Google publishes, lets an outsider tell which model produced a given answer or whether it came from cache. Fox's own warning about stale caches describes a failure that users and businesses have no way to detect.

The click that did not happen

The sharpest moment on traffic came from the interviewer. Describing his own habits, Valetto said: "I see myself not even clicking on links anymore because the summary provides all the information I need."

Fox did not accept the implication. "Our queries are much much longer than they used to be," he said, describing users who once typed "car grinding noise" and now describe the make of car, the ignition, a warning light and the louder grinding when the wheel turns right. Users ask follow-up questions, going "five questions, 10 questions deep", and return to threads weeks later. On traffic, "people still click through to the web a lot", he said. Of the "zero click" narrative: "That's not what we're seeing." He conceded that very simple questions may end without a click, and added: "We send billions of clicks to the web through our AI experiences." Google, he said, takes "our commitment to the web ecosystem very very seriously".

The usage data is on his side. When AI Mode passed one billion monthly users in May 2026, Google said its queries had more than doubled each quarter, ran three times the length of traditional searches, and that follow-ups in the US were growing by more than 40 percent a month. The click data is less friendly. Fox's "billions" echoes a post he published on July 17, a volume figure without a rate attached. Ahrefs found in February that AI Overviews correlate with a 58 percent fall in click-through rate for top-ranked pages, and a randomised experiment with 1,065 desktop Chrome users found outbound organic clicks fell 39.8 percent when an overview appeared. Datos panel data points the other way, putting zero-click searches at their lowest level in its first-quarter 2026 series.

Google's own accounts settle part of the argument. In the first quarter of 2026, its Network segment - the advertising it sells on other people's websites - fell 4 percent to 6.97 billion dollars, and a study of more than 400 publishers in the same coverage found Google Web Search's share of traffic to news sites dropping from 51.10 percent in 2023 to 27.42 percent in the fourth quarter of 2025. In the second quarter, Network slipped another 1 percent to 7.3 billion dollars while Search advertising rose 17 percent to 63.3 billion dollars. Money made on Google's own pages is growing. Money made on the pages Google sends people to is not. A total of "billions of clicks" can rise while each publisher earns less, and Fox's figure speaks only to the total.

Rebuilding the ad machine

For advertisers, the passage that matters most was one of the shortest. "In the same way we've re-engineered search, we're re-engineering our advertising system and our search advertising system," Fox said. "So, we're in the process of doing that." The business, he explained, rests on matching people looking for something with merchants trying to sell it, and doing that "with higher signal, deeper signal of intent" is "a big opportunity to take that forward".

Set that beside his account of 2022. Valetto described Search and its ads as "probably the most profitable product in history" and asked about the innovator's dilemma. "We never saw it as a dilemma," Fox replied. Yet a company that faced no dilemma is now rebuilding the auction that funds it. The two statements are compatible only if the dilemma was simply postponed: AI answers could be adopted without hesitation because the monetisation question would be solved later. That is the work Fox described as in progress.

Fox gave no timeline or product names, but parts of it are public. At Google Marketing Live in May, Google showed new AI Mode ad formats alongside Direct Offers and Universal Cart. A limited experiment makes exact and phrase match keywords eligible to serve text ads in AI Mode on queries Google judges to carry explicit intent, while AI Max for Search and Performance Max keep exclusive access to newer formats such as Highlighted Answers. Outside AI Mode, a 383 million impression analysis by smec found AI Max expanding more than 25 percent of exact match impressions.

"Deeper signal of intent" has a practical meaning. The system, rather than the advertiser's keyword list, increasingly decides what a query means and which offer answers it. Productrise measured what that looks like in shopping: where the same product appeared in both AI Mode and conventional search on the same day, the lead price in AI Mode averaged 21.6 percent higher, the lead seller differed on 49.6 percent of matched products, and only 1.28 percent of conventional search products appeared in AI Mode at all.

How Google got here

Fox's history of late 2022 was candid in places. The chatbots of the time "hallucinated, they would get things wrong", he said, yet "there was still a lot of user appetite" for them. "It emboldened us." Google had already been using AI in ranking, through BERT and MUM, and Fox described each platform shift, from the web to mobile, as "the fuel for the next wave of Google's growth". Inside the company, he said, there was no resistance: "technology is like candy".

The rollout was staged. "Let's not change the experience for everyone on day one," Fox recalled telling colleagues. The first step was an opt-in Labs product, followed by AI Overviews on a very small share of queries before they became central to Search. The lesson Google drew from ChatGPT, in Fox's telling, was that users would tolerate imperfect answers if they arrived in the right form.

Agents, tutors, jobs and safety

"Step one is let me research something. Step two is let me do something," Fox said of agentic search, citing hotel bookings, refunds and driver's license or passport applications. Valetto credited Antigravity, the agent development platform Google rebuilt as Antigravity 2.0 at I/O in May, with changing how he does research; for Fox, "coding is so much more than coding".

On education, Fox predicted "everyone's going to have a personalized AI tutor over time" and called the ban-or-not debate "far too simplistic". "The answer isn't to ban it." On jobs, he said "AI will be a job creator rather than the opposite of that", dismissing much of the debate as "not rooted in facts and it's also not rooted in history". His one piece of evidence carried no numbers: "Google's hiring more software engineers this year than it hired last year."

On safety, asked about recent remarks by Sam Altman and Elon Musk on pacing frontier models, Fox said "safety has to be built in from the beginning" and that "we should assume that there will also be bad actors using AI", pointing to Google's cyber models for patching software. For coordination he backed "something sort of more like FINRA", the industry-run body that polices US securities firms. It is a telling choice. On search, Fox wants Brussels to regulate less; on frontier AI, the model he proposes is one in which the industry supervises itself.

The one number Fox watches

The last question asked for the metric that would prove Google's AI strategy worked. "The metric that I look at the most is the number of queries that people are doing," Fox said. "Maybe that's an uninteresting metric."

It is a revealing one. Queries rise when people ask Google more, including when they ask follow-up after follow-up without leaving. Google's own reporting treats each follow-up in AI Mode as a new query, so a conversation that once ended in a click to a website can now generate ten queries and no click. Clicks to the web, publisher traffic and advertiser returns do not appear in Fox's measure of success. He has promoted it before: in July 2025 he highlighted that AI Overviews were driving more than 10 percent more queries for the query types that show them. For advertisers, queries are also the base on which every search auction runs.

Why this matters for the marketing community

European travel and retail advertisers are the first audience. Fox's remarks indicate that Google intends to keep presenting feature removals as the price of compliance, and every removal so far - fares, trip dates, free product listings - sits on commercial queries where paid and unpaid placements compete. Whether comparison sites, suppliers or Google's own ads absorb the displaced clicks will be settled vertical by vertical, and the data each side publishes will be self-interested.

Measurement is the second problem. With site reputation penalties switched off inside the EEA, one URL can hold two ranking positions depending on where a query originates, and reports that blend European and non-European search data will average across two enforcement regimes.

The advertising rebuild matters most over time. Keyword structures built for two-word lookups were not designed for queries three times longer, and the evidence so far - AI Max absorbing exact match traffic, keyword ads being tested in AI Mode, different prices and sellers in AI answers - points to more of the matching decision moving from the advertiser to Google. Fox's success metric, queries rather than clicks, tells publishers and advertisers where Google's priorities sit.

Finally, the data sharing due in January 2027 covers paid as well as organic results. Ad URLs are removed and replaced with placeholders, but rival engines will still see where ads appeared on the page. Fox's privacy objection is the argument Google is carrying into that rollout.

Timeline

Summary

Who: Nick Fox, Google's senior vice president for knowledge and information, interviewed by Pietro Valetto, a PhD student at the University of Antwerp who runs the Popular Economics channel and Substack. The European Commission, publishers, travel and retail businesses, rival search engines and search advertisers are the parties most affected by the matters discussed.

What: A 42-minute interview in which Fox called Google's Article 6(5) compliance changes "the biggest degradation ever" to Search, citing the loss of real-time flight pricing, trip-date search and some spam protections in Europe; called Article 6(11) data sharing "a really massive violation of European privacy"; described three cost-control techniques for AI answers (model routing, caching and full-stack infrastructure); rejected the zero-click narrative; and said Google is re-engineering its search advertising system. Several claims sit uneasily with the record, including that the Commission's decision orders no feature removal and that Google has shared anonymised query data since March 2024.

When: Published on YouTube on September 25, 2026. The recording date is not stated; Fox placed it roughly two weeks after Google set out its Article 6(5) compliance approach, which took effect in European results on September 8. The 60-day compliance window had closed on September 21.

Where: Recorded at a Google pop-up venue in Brussels. The regulatory changes discussed apply to Google Search in the European Economic Area.

Why: Google is contesting how the Digital Markets Act is being enforced while complying with the July 2026 decisions, and Fox used the interview to argue that compliance has made Search worse and less private for Europeans. The technical and advertising remarks matter because they show Google rebuilding Search economics around longer, AI-answered queries, measured by query volume rather than clicks to the web.