Sundar says they love it. The install charts say otherwise.

When an incumbent insists users love a feature and a rival's install graph spikes 30% in the same week, the revealed preference is the story. Google's AI Mode is not failing on engagement metrics, it is failing on consent, and DuckDuckGo has just made that failure numerically legible.

The week the graphs disagreed with the CEO. On 25 May, DuckDuckGo's US app installs peaked 30.5% above baseline, with iOS briefly hitting 69.9%. Visits to noai.duckduckgo.com, the company's AI-stripped results page, ran 22.7% above baseline for the week and topped out near 27.7% on 24 May. The week before, Sundar Pichai had publicly said users "love" AI Mode. Gabriel Weinberg, DuckDuckGo's CEO, called it force-feeding. The install data sided with Weinberg.

This is not the anti-AI story it looks like. DuckDuckGo's own assistant, Duck.ai, routes users to Claude Haiku 4.5, GPT-5 mini and Llama 4 Scout, with private routing in front. The company is not selling abstinence. It is selling consent. The product decision that matters is not the absence of AI but the existence of a dedicated noai.duckduckgo.com toggle, which DuckDuckGo built before this week's surge. The toggle existed because the demand existed. The surge is the demand becoming visible.

That distinction matters because it changes what is being priced. If users wanted no AI, Duck.ai would be irrelevant and DuckDuckGo's growth would be capped by the size of the anti-AI niche. If users want opt-in AI, DuckDuckGo is competing for the much larger segment of people who are fine with the technology and furious about the defaults. The second market is bigger and more durable than the first.

The product is not AI-free search. The product is the off switch.

The strongest counter is Google's revenue, and it is genuinely strong. Google's Q1 2026 search revenue grew 19% year-over-year on exactly the AI-integrated product users are reportedly fleeing. Advertiser spend is itself a revealed preference, and a more expensive one than an app install. If AI Mode were destroying the search experience, CPMs (the price advertisers pay per thousand impressions) and query monetisation should be the first signals to crack. They have not cracked. They have accelerated.

There are three ways to read this. One: advertisers are pricing intent capture, and AI Mode captures intent better than ten blue links, regardless of whether the user enjoys the interaction. Two: the install surge is a vocal-minority signal amplified by Hacker News (the story hit #4 on 27 May) and tech press, not a representative consumer shift. Three: both data points are real, and they describe different populations. The users who hate AI Mode are switching. The users who tolerate it are converting better than before. Google is monetising the second group hard enough that losing the first does not show up in the quarterly print.

I think the third reading is the honest one. A 30% install spike on DuckDuckGo's base is a real signal about a real segment, and it is not yet a structural threat to Google's query volume. Both things are true. The interesting question is which segment grows.

The segment that has no exit is the one to watch. Consumer switching is the easy case. The harder case is the employee whose company mandates Google Workspace, or the student whose school does. AI Mode arrives as the default on a search box they did not choose and cannot replace. The week-over-week install graph captures none of these people, because they cannot install their way out. The consent failure is sharper here than in the consumer story, and there is no DuckDuckGo equivalent for institutional defaults. Whatever Google decides about AI in Workspace search is, for that population, what happens.

The threat neither side is pricing is the conversational interface. If users increasingly start their queries inside ChatGPT or Gemini's assistant mode rather than a search box, the entire AI-Mode-versus-private-search argument is a fight over the silver medal. Google's AI Mode is, among other things, a defensive bet that the search box remains the front door. DuckDuckGo's private routing is the same bet with a different value proposition. Both lose if the front door moves.

The DuckDuckGo install surge is, in this light, slightly reassuring for Google, and slightly damning. Reassuring because users protesting AI Mode by switching to a different search engine are at least still using search engines. Damning because the protest reveals that the search box is now a contested interface, which is the first condition for users abandoning it for something else.

What I would watch from here. Three signals.

First, whether DuckDuckGo's install surge holds past the news cycle. A spike that decays back to baseline by mid-June is a protest. A spike that compounds is a shift. The noai.duckduckgo.com retention curve is the cleanest read.

Second, whether Google ships a real opt-out. The current product gives users no native way to turn AI Mode off. If a toggle appears within a quarter, that is Google conceding the consent argument while keeping the revenue. If no toggle appears, Pichai's "users love it" stops being a claim and becomes a strategy: monetise the captured intent and accept the churn at the margin.

Third, whether ChatGPT's referral traffic to publishers grows faster than Google's search traffic shrinks for the same queries. That is the upstream signal. If it does, the AI Mode debate is already a rearguard action.

The honest summary is that Pichai is probably right about aggregate engagement and clearly wrong about consent, and the gap between those two facts is what DuckDuckGo just monetised. Whether that gap closes through a Google opt-out, widens into a structural shift, or gets overtaken by the conversational interface entirely, is the question the next two quarters answer. For now, the install graph is doing the talking the survey data was not.

Footnotes

¶