Last week a privacy-focused search engine that has spent its entire existence stuck at around two percent of the United States market suddenly mattered. DuckDuckGo reported that its U.S. app installs climbed an average of 18.1 percent week-over-week between May 20 and May 25, that the growth held for six straight days, and that it peaked at 30.5 percent on May 25. On iOS the numbers were even louder: a 33 percent average week-over-week jump, peaking at 69.9 percent. Traffic to its dedicated AI-free page, noai.duckduckgo.com, grew an average of 22.7 percent week-over-week and peaked at 27.7 percent on May 24. The company noted that the trend was strongest in the United States and that it kept climbing through Memorial Day weekend, a stretch when DuckDuckGo normally sees traffic dip. People did not go looking for a privacy crusade. They went looking for a search engine that still works like a search engine, and for the first time in a long time they found one somewhere other than Google.
To understand why a two-percent player suddenly spiked, you have to look at what Google did on May 19. At its I/O developer conference, Google announced what it openly called the biggest change to its search box in more than twenty-five years. The familiar list of ten blue links is being demoted and, in many cases, replaced outright by an AI-powered interface built on Gemini 3.5 Flash. The company is calling it an intelligent search box. Instead of typing a few keywords and getting a ranked list of pages you can evaluate yourself, you now drop into AI-generated interactive experiences. Liz Reid, Google’s head of Search, described a future where you dispatch information agents that monitor the web around the clock and ping you when conditions are met, where generative interfaces build dashboards on the fly so you never have to visit the website that produced the underlying information at all. AI Overviews now reach 2.5 billion users a month, and AI Mode, launched barely a year ago, has crossed a billion monthly users, with query volume reportedly more than doubling every quarter. Sundar Pichai’s company is not nudging users toward AI. It is closing every other door.
Gabriel Weinberg, DuckDuckGo’s chief executive, put it more bluntly than I could. He said Google is force-feeding AI with no way to opt out, and that as a result its results are getting worse, not better. He is not wrong, and you do not need a study to confirm it. You can confirm it by typing a single word. After the I/O update, if you Google the word disregard, Google serves you a giant block of AI verbiage that has no conceivable value to anyone who wanted to know what the word means, while the actual Merriam-Webster definition sits buried beneath a wall of empty space most users will never scroll past. Russell Brandom, a tech journalist of nearly fifteen years, ran the same search on Bing and found Bing more useful. He said, accurately, that he could not remember a single time in his career when a Bing result beat the Google equivalent. There is, apparently, a first time for everything. When your product breaks on the word disregard, you have not built an assistant. You have built a tool that no longer does the one job it was hired to do.
This is the part Google’s leadership seems unable to internalize, because it has spent two decades measuring engagement instead of satisfaction. The blue link was never the limitation. The blue link was the product. It was a contract: you ask a question, we hand you a ranked set of sources, and you decide which one deserves your trust and your click. That contract respected the user as a competent adult capable of evaluating evidence. The AI overview tears the contract up. It reads ten sources you will never see, blends them into a confident paragraph with no accountability, and keeps you inside Google’s walls. When it is right, it saves you fifteen seconds. When it is wrong, and it is wrong constantly, it launders a hallucination into something that looks authoritative because it is sitting where the answer used to be.
The cost of that trade is not theoretical, and it is not only borne by users. It is being paid by every website that ever made Google useful in the first place. Pew Research Center tracked roughly 68,000 real search queries and found that users clicked through to a result only 8 percent of the time when an AI summary appeared, compared with 15 percent when it did not. That is a relative drop of nearly half. Worse, only about 1 percent of users clicked any of the sources cited inside the AI summary. Similarweb data showed zero-click searches rising from 56 percent to 69 percent between May 2024 and May 2025. Authoritas found that traffic to top-ranked news sites fell by as much as 79 percent when their link appeared below an AI Overview. By one Semrush measure, 93 percent of searches in AI Mode now end with no click to any external site at all. Google built the most valuable company in history by indexing the open web and sending it traffic. Now it is harvesting that same web, strip-mining the publishers, the forums, the small specialist sites, the independent reviewers, and the bloggers who supplied the raw material, and returning almost nothing. It is eating the orchard and calling it dinner.
So when you stack Google’s AI search on top of the ads and the sponsored shopping carousels that already pushed organic results halfway down the page, you arrive at a search experience that is loud, cluttered, frequently wrong, and structurally designed to keep you from leaving. The query goes in. A monetized, AI-generated approximation comes out. The actual web recedes further with every release. For a meaningful and growing slice of users, the value Google delivers per query has been falling for years, and the I/O overhaul finally made the decline impossible to ignore. The woman a TechCrunch reporter overheard on the phone last week, the one who said she was switching to DuckDuckGo because you can opt out of using AI and that Google just isn’t Google anymore, was not making a privacy argument. She was making a quality argument. The privacy was a bonus.
Here is the larger pattern, and it is the one every company building an AI roadmap should be staring at right now. The public is not neutral on this technology. It is hostile, and the hostility is hardening. Pew found that 50 percent of U.S. adults are now more concerned than excited about the increased use of AI in daily life, up from 37 percent in 2021, while only 10 percent are more excited than concerned. Fifty-seven percent rate AI’s societal risks as high; just 25 percent say the same about its benefits. Forty-three percent believe AI will harm them personally, against 24 percent who expect to benefit. Sixty-four percent expect AI to mean fewer jobs over the next two decades. And 61 percent say they want more control over how AI is used in their lives, which is precisely the control Google just took away. There is a yawning gap between the people building this technology and the people living with it: 47 percent of AI experts say they are more excited than concerned, a sentiment shared by only 11 percent of the public. Silicon Valley is throwing itself a party that the rest of the country was not invited to, and increasingly does not want to attend.
That resentment did not appear from nowhere. It was earned, company by company, layoff by layoff, by executives who decided the smart move was to brag about replacing people. Sebastian Siemiatkowski, the chief executive of Klarna, became the poster child for this posture. He boasted that Klarna’s AI chatbot was doing the work of 700 customer service agents, later revised upward to 800, and that AI had let him shrink his workforce by 40 percent. Klarna went from more than 7,000 employees to fewer than 3,000, and Siemiatkowski floated dropping below 2,000 by 2030. He publicly mused that AI was capable of doing all jobs, including his own. Then reality arrived. Customer satisfaction scores on complex interactions deteriorated, and through 2025 and into 2026 Klarna quietly rebuilt its human support capacity, shifting to a hybrid model where the AI handles routine volume and humans handle anything that actually matters. Klarna is now the canonical cautionary tale that boards cite when they want to know how not to do this. Its stock has fallen roughly 59 percent since its IPO, and it posted a 99 million dollar quarterly net loss. The man who could not stop talking about firing people had to start quietly hiring them back.
Luis von Ahn, the chief executive of Duolingo, learned the same lesson faster and more painfully. In April he posted an internal memo announcing that Duolingo would become an AI-first company. The backlash was immediate and ferocious, the kind that follows you into the comments of every post for months. Within a week he was walking it back, clarifying that he did not actually see AI replacing what his employees do. He later admitted he had not expected the amount of blowback, and offered the most revealing defense imaginable: every tech company is doing similar things, he said, but we were open about it. That is the whole story in one sentence. The crime, in the eyes of the public, was not using AI. The crime was saying so out loud, treating human livelihoods as a line item to be optimized away, and expecting applause. Meanwhile Dario Amodei, the chief executive of Anthropic, was on record predicting that AI could eliminate roughly half of all entry-level jobs within five years. It sounds crazy, he said, and people just don’t believe it. People believe it. That is exactly the problem. They believe it, they resent it, and they are starting to vote with the only ballot they have, which is their attention and their installs.
This is the backlash mechanism that Google walked straight into, and that every public-facing brand should now treat as a live business risk. When a company loudly adopts AI in a way that degrades the customer’s experience, or fires people and frames it as innovation, it is not just absorbing a one-time public relations hit. It is creating a standing reservoir of customers who are actively looking for a reason and a route to leave. Most of the time they cannot, because there is no alternative. The brilliance of what just happened with search is that, for once, the alternative was sitting right there, free, one app-store download away, and Google handed those customers the motivation on a silver platter. The 30 percent spike in DuckDuckGo installs is what defection looks like when the exit door is finally unlocked. Every CEO who has spent the last two years bragging about headcount reductions should understand that they are accumulating the same kind of latent ill will, and that the moment a credible competitor offers an off-ramp, it converts.
Which brings me to the actual point of this piece, the part where I stop diagnosing and start recommending. You should try DuckDuckGo. Not as a political statement, not as a protest, but because for an enormous range of everyday tasks it is now simply the better tool, and the only way you will believe that is to use it for a week.
Start with the thing that triggered all of this. DuckDuckGo lets you decide how much AI you want, instead of deciding for you. If you want the clean experience, you go to noai.duckduckgo.com and every AI feature, the assisted answers and the AI-generated images, is off by default. If you want a summary, there is an optional Search Assist that behaves like Google’s AI Overviews, and an AI Image Filter that strips AI-generated junk out of your image results, which is rapidly becoming one of the most useful features on the internet as the web fills up with synthetic slop. Kamyl Bazbaz, the company’s policy chief, said the AI features and the no-AI page are both among its most popular, which tells you everything: people just want a choice. That is not a hard product principle to grasp. Google could not bring itself to offer it.
Then notice what DuckDuckGo does that Google structurally cannot, because Google’s business model forbids it. DuckDuckGo does not build a profile of you. It processes well over 100 million searches a day without storing your search history. It makes money from contextual search ads tied to the query you just typed, not from a dossier assembled across every site you have ever visited. Its browser and extensions block third-party trackers from Google, Meta, and the rest of the ad-tech complex before those scripts ever load, which has the pleasant side effect of making pages load faster and feel lighter. On the desktop extension it will even block Google’s newer in-browser tracking mechanisms, the Topics and Protected Audience interfaces, that exist specifically to keep auctioning your behavior after third-party cookies die. The contrast is almost comedic. Google’s tools are built to watch you. DuckDuckGo’s tools are built to stop the watching.
The everyday quality-of-life features are where it stops feeling like a sacrifice and starts feeling like an upgrade. Email Protection gives you a free at-duck-dot-com forwarding address that strips trackers out of incoming mail before it reaches your real inbox, so the marketing emails you open can no longer phone home to tell a vendor exactly when and where you read them. On Android, App Tracking Protection blocks third-party trackers across all the apps on your phone, not just inside the browser, working system-wide in a way that even Apple’s much-publicized tracking transparency does not fully match. The Fire Button wipes your tabs and browsing data in one tap. And if you ever want a chatbot, Duck.ai gives you Claude, Llama, Mistral, and GPT models for free, with no account required, while stripping your IP address before requests reach the model providers, deleting conversations within thirty days, and guaranteeing nothing you type is used for training. You get the AI when you want it, on your terms, without paying for it with your identity.
Run a real comparison for a few days and the difference stops being abstract. Look up a medication and Google may hand you an AI summary you are explicitly warned can make mistakes, sitting where a pharmacist-grade source used to be; DuckDuckGo hands you the sources and lets you read them. Search a recipe and Google increasingly tries to assemble one for you from scraped fragments while burying the food blogger who actually developed and tested it; DuckDuckGo sends you to the blogger. Search a local business, a product review, a piece of code documentation, a definition, and you will find the same pattern over and over: Google interposes a confident machine between you and the answer, and DuckDuckGo simply gives you the answer. For the queries where you genuinely want a synthesized response, you can turn Search Assist on. For everything else, you get the web back. That optionality, the thing Google decided you were not adult enough to handle, is the entire pitch.
None of this means DuckDuckGo is perfect or that the spike is permanent. It still leans on Bing for parts of its index, so it inherits some of Bing’s blind spots. It cannot hide which sites you visit from your internet provider without a VPN. Two percent is a long, long way from dethroning a company that still owns roughly nine in ten searches and pays billions to remain the default on every device it can reach. Habit is the strongest moat in consumer technology, and most people will drift back to Google the moment switching requires more than one tap. Google will also, almost certainly, ship an opt-out eventually, once the install numbers scare it enough, and absorb the lesson the way incumbents always do, a few quarters too late and only under duress.
But something real happened here, and it is bigger than one search engine’s good week. For twenty-five years the entire industry operated on the assumption that consumers would accept any amount of degradation as long as the product stayed free and convenient, and for twenty-five years that assumption held. The DuckDuckGo spike is the first hard, quantified evidence that the assumption has a ceiling. When you force-feed people an AI they did not ask for, break the basic function they relied on, and tell them it is for their own good, a measurable number of them will get up and walk out the moment they discover the door is unlocked. They were not asking for a revolution. They were asking for a search engine. It turns out they wanted one all along.
Go download the app. Set noai.duckduckgo.com as your homepage. Use it as your only search engine for one week and pay attention to how often you reach for the answer and how rarely you feel managed. If it does not stick, you have lost nothing and Google will welcome you back with open arms and an open profile. But I suspect that for more of you than expected, the quiet will feel less like a downgrade and more like something you forgot you missed. The web is still out there. Somebody just has to be willing to point you at it instead of standing in the way.
About the author
Gal Ratner is a software architect with thirty years on the Microsoft and .NET stack and a long-standing skepticism of whatever the industry is currently calling inevitable. He was employee number six at Break.com, founded the consultancy Inverted Software in 2001, and has delivered enterprise work for Microsoft, Sony, Yahoo, Best Buy, Rockstar Games, and 2K Games. Today he is Chief Architect at Prana Entertainment, where he builds across a portfolio of web properties, and he spends most of his time on agentic AI, MCP servers, and the unglamorous question of whether any of it actually makes products better for the people using them. He was a Los Angeles Business Journal CTO of the Year finalist, is the author of the novel The Archive of Lost Suns, and co-hosts the motorcycle-focused Edge Grip Podcast. He writes about technology, incentives, and the gap between the two at galratner.com.