Even if you hate ai you will use google ai search

Seventeen years have passed since I sat in on that fabled weekly search quality meeting in the Ouagadougou conference room at Google’s Mountain View campus. On that Thursday morning, roughly three dozen engineers, product managers, and executives gathered—some crowded around the main table, others sprawled across the floor—to dissect why certain search queries and content categories failed to deliver perfect results, and brainstorm how to fix those gaps. Back in 2010, these collaborative reviews led Google to roll out 550 updates to its core search algorithm, a number that felt strikingly ambitious at the time.

That old memory now feels like a faded tintype: an artifact of a long-gone era. At this week’s Google I/O developer conference, head of search Liz Reid used her keynote address to formally relegate traditional organic search to near-obscurity. This shift is just the latest step in a transition that launched two years ago, when Google debuted AI Overview—AI-generated summaries that pin to the very top of search results pages, hovering directly over Google’s famous “10 blue links.” Even before that launch, though, the iconic blue links had already been steadily eroded: more often than not, the most relevant results were buried beneath content aggregators, spam, and Google’s own in-house shopping and map outputs.

Now, in what Reid called the most transformative change to Google’s core search bar in the company’s history, users interact directly with the latest iteration of Google’s Gemini AI. Even the old term “search query” feels outdated here—human inputs are no longer just requests for indexed results, they are opening lines of a collaborative conversation with the AI. The new system can also draw on the vast trove of personal data Google holds about you to tailor outputs, and the end result can be a fully custom response, often augmented by specialized AI agents that dig for information across lesser-crawled corners of the internet. The overhaul is complete: Google stated it explicitly from the I/O stage: “Google Search is AI Search.”

Google’s search bar once served as an open gateway to the entire public web. The new “intelligent” search bar is instead a prompt to generate a Gemini-powered, custom-tailored response to your request—sometimes it even builds a one-of-a-kind mini-publication on the fly, complete with custom charts, bullet points, and even animated elements. Google once built its reputation on decoding vague, shorthand search terms to uncover exactly what a user intended. Now, it pushes searchers to hold extended conversational back-and-forths with Gemini. To drive this shift home, Google representatives at the conference wore branded T-shirts emblazoned with “Ask Me Anything,” mirroring the open-ended prompt Gemini offers users. Just like the AI experience, if you asked one of these smiling staffers for directions, you wouldn’t get sent off to click through to an external website for your answer.

Our digital lives are currently stuck in an awkward, uneasy transition. AI is now the driving force behind nearly every major tech business model, and giants like Google are weaving AI functionality into every one of their products and internal operations. At the same time, public pushback and even outright frustration with this powerful, often intimidating technology is growing as it becomes more embedded in daily life—just look at the boos that greet any commencement speaker that brings up AI these days. But from Google’s perspective, AI-powered search (if you still want to call it that) is an inevitable shift that even AI skeptics will end up embracing.

I was among the critics who recoiled at the launch of AI Overview back in 2024. But now, I have to admit: AI Overview, and the deeper “AI Mode” it encourages users to try, is simply a better tool for many common search tasks. That holds true whether you’re checking if Saturday Night Live aired a new episode recently, getting a clear explanation of what an agentic harness is, or even tracking down a specific old link. When I searched for the WIRED article I wrote years ago about that original Ouagadougou conference room meeting, the traditional blue links were nearly useless. But when I explained what I was looking for in plain, conversational language, I found it instantly.

So the new system works, at least for many common use cases. Google reports that more than one billion people per month now use AI Mode, a dedicated separate tab on Google’s search site where traditional external links are even more marginalized. The volume of AI Mode queries is doubling every quarter, the company says.

I spoke with Reid after her keynote and asked her a direct question: How does she define search in this new era? After a surprised pause, she leaned on Google’s long-stated core mission: “Can you truly make information not just organized, but actually useful and accessible to people?”

Of course, the original Google operated under the assumption that the key to fulfilling that mission was a thriving, open public web. Per Reid’s keynote, Google still scrapes billions of web pages every single day—but the goal now is no longer to index those pages for users to find themselves, it is to collect facts and insights to feed Google’s personalized AI responses.

In the I/O keynote, Google demonstrated how a single search can now deploy an entire fleet of AI agents to build a custom, one-off experience for the user on demand. “We’re talking dynamic layouts, interactive widgets, entire experiences created just for you,” said Robby Stein, Google’s vice president of search. To answer a query about black holes, for example, AI agents might pull together a fully interactive graphic explaining how the cosmic objects work. But all that information has to come from somewhere: the raw material for that output is the hard work of cosmologists, science writers, and visual artists, none of whom are easily credited or have their work prominently linked in the final AI response. These kinds of creators, and the websites that host their work, look to be the biggest losers in this industry transition.

Unsurprisingly, Reid rejects the idea that AI search pulls the rug out from under the traditional open web. “Some people will skip the AI response entirely and go straight to the links,” she argues. “Oftentimes people will open the AI view first, then click through to the links within the response.” I asked her for public data on how many users actually do that; Reid says Google does not share that metric.

She acknowledges that some websites will be hurt by the shift: low-quality “bottom feeder” sites that churn out generic content AI can easily replicate. But she claims that original voices, and work built on unique reporting or original research, will still find large audiences. (Really? Tell that to the news outlets that have already suffered devastating traffic drops from AI Overviews.) “We’re also learning in new ways how to identify which web content is relevant,” she says, noting that Google is working to direct more users to content that offers “firsthand perspectives.” I’ll believe that when I see that traffic hit my own work.

Of course, another core flaw of getting a single AI-generated answer is that, like all large language models, Gemini can be wrong, or even invent facts out of thin air. “The technology is certainly not perfect by any stretch,” Reid concedes, though she adds that errors and hallucinations are far less prevalent than they were in earlier versions of the model.

My final question for Reid was whether Google still holds the same kind of search quality meetings I attended back in 2009. Reid, who has worked at Google for more than two decades, says those discussions are now split across multiple smaller meetings, but the core goal remains the same. “At the end of the day, there’s still a human judgment,” she told me. And no matter how much you dislike AI, you’re probably going to end up using it. Overall search volume, Reid says, is at an all-time record high right now.

This is an installment of Steven Levy’s Backchannel newsletter. Read previous editions here.