Infinite Scroll

The Death Knell of Digital Search

数字搜索的丧钟
Companies such as Google used to be our portals to the rest of the web. Can the internet as we know it survive when A.I. answers all of our questions?
谷歌这类公司曾是我们通往整个互联网的门户。当人工智能包揽了我们所有问题的答案,我们所熟知的互联网还能存活下去吗?
Illustration of a search magnifying glass on a tombstone
Illustration by Ariel Davis

Lately, I’ve been having trouble finding stuff in my Gmail inbox. The Search function has become overzealous, jumping to summarize my conversations using artificial intelligence and skipping the actual addresses I’m attempting to search. It’s not just e-mail accounts, though; search is in crisis everywhere. Google Maps guesses where I want to go at a given time of day instead of simply listing places I’ve just searched for. When I try to use Google Search, the A.I.-overview function sometimes occupies my entire browser window, generating its own, not always correct, answers rather than directing me to relevant websites. Full-on A.I. mode skips the old blue links entirely.

最近,我在Gmail收件箱里找东西越来越费劲。搜索功能变得过分殷勤,动不动就用人工智能把我的往来邮件总结一番,却跳过了我真正想搜的那些具体地址。出问题的远不止电子邮箱——搜索正在全面陷入危机。谷歌地图会猜测我在一天中的某个时段想去哪里,而不是老老实实列出我刚搜过的地点。我试着用谷歌搜索时,人工智能概览功能有时会占满整个浏览器窗口,自顾自地生成答案——还不一定正确——而不是把我引向相关网站。完全开启人工智能模式后,那些老式的蓝色链接干脆消失不见了。

Search has gotten so annoying, in fact, that it has become a problem for Google’s very business model. Companies pay Google to splash ads into search results that are relevant to their businesses—advertising stoves, say, to someone searching for kitchen-renovation tips. More than half of Google’s revenue comes from those ads. But, if users aren’t clicking through to external sites, because they’re relying on A.I. to answer their queries, then the ads that Google hosts are suddenly much less valuable. In a recent earnings call, Google’s chief business officer, Philipp Schindler, admitted that the company has been seeing “lower monetization” on A.I. searches. After the call, the stock of Alphabet, Google’s parent company, fell by nearly a quarter, down five hundred billion dollars, the largest single-day drop in its history. Even Google may not be too big to fail, as pessimism mounts about the possibility of Search ever again working the way it used to.

搜索变得如此烦人,以至于它已经成了谷歌商业模式本身的一个麻烦。企业付钱给谷歌,让广告出现在与自身业务相关的搜索结果里——比如,向正在搜索厨房翻新技巧的人推送炉灶广告。谷歌一半以上的收入来自这些广告。但如果用户不再点击跳转到外部网站,而是依赖人工智能来回答他们的问题,那么谷歌承载的这些广告价值就会骤然暴跌。在最近一次财报电话会议上,谷歌首席商务官菲利普·申德勒(Philipp Schindler)承认,公司的人工智能搜索“变现率”一直在下降。电话会议结束后,谷歌母公司Alphabet的股价暴跌近四分之一,市值蒸发五千亿美元,创下其历史上最大单日跌幅。随着外界对搜索能否恢复往日运作方式的悲观情绪日益加重,即便是谷歌,恐怕也未必“大而不倒”。

Just as Google has faltered in Search, use of OpenAI’s ChatGPT has surged. OpenAI’s consumer-product lead, Adam Fry, recently told the tech journalist Alex Kantrowitz that ChatGPT is siphoning users away from the older company: over the past year and a half, it has grown to handle one-fifth the volume of “search-like queries” that Google does. This acceleration only makes sense: a growing number of people now rely on A.I. chatbots as their primary windows to the internet, and they’re less likely to click out of those apps to ask the kinds of simple questions that they might once have turned to Google to answer. It doesn’t help that bots have lately grown much better at browsing the internet in real time, using virtual browsers and plug-ins, rather than simply generating answers from preëxisting data. Google has its own suite of A.I. products, Gemini, but they’ve lagged in popularity behind Anthropic’s Claude and ChatGPT. The investment firm Menlo Ventures found that Gemini accounts for only nine per cent of A.I. spending for businesses in 2026, down from twenty-one per cent last year.

就在谷歌搜索步履蹒跚之际,OpenAI的ChatGPT使用量却在激增。OpenAI消费产品负责人亚当·弗莱(Adam Fry)最近告诉科技记者亚历克斯·坎特罗维茨(Alex Kantrowitz),ChatGPT正在从这家老牌公司手中吸走用户:在过去一年半里,它处理的“类搜索查询”量已增长到谷歌的五分之一。这种加速增长完全合乎逻辑:越来越多的人现在把人工智能聊天机器人当作他们上网的主要窗口,他们不太会再跳出这些应用,去问那些原本可能会求助于谷歌的简单问题。雪上加霜的是,这些机器人最近在实时浏览网页方面进步神速,它们使用虚拟浏览器和插件,而不是仅仅从既有数据中生成答案。谷歌也有自己的人工智能产品套件Gemini,但其受欢迎程度一直落后于Anthropic的Claude和ChatGPT。投资公司Menlo Ventures发现,2026年企业在人工智能支出中,Gemini仅占百分之九,低于去年的百分之二十一。

Can the internet as we know it survive without a reliable flood of search traffic? Chatbots treat websites as databases of information to draw from, but since they divert traffic from those sites they don’t provide much impetus for creators to keep posting new content. The end state we’re heading toward seems to be an internet both generated and traversed by bots—in which case, it will no longer matter how annoying it is for a mere human being to find information. Some websites are already moving defensively to keep out the trawling bots attempting to digest their content. Reddit, one of the internet’s major remaining repositories of human-generated knowledge and opinions, is moving toward “a primarily logged-in experience,” as its C.E.O., Steve Huffman, recently announced. In a third-quarter letter to shareholders, Huffman noted that Reddit’s search traffic “dropped by more than 1 million referrals on average per week” during October. The letter explained, “Web search is evolving from a gateway to the web to becoming an AI-powered destination itself.” In a recent earnings call, Huffman said that “people still want Reddit” but that “Google is not sending them to Reddit the way they used to.” Requiring users to log in is the equivalent of pulling up the drawbridge to keep away the A.I. models roaming outside. But it also means that the casual human searcher will be less likely to see the full conversations happening on Reddit.

如果没有可靠的搜索流量源源不断地涌入,我们所熟知的互联网还能存活下去吗?聊天机器人把网站当作汲取信息的数据库,但由于它们分流了这些网站的流量,也就无法为创作者持续发布新内容提供多少动力。我们正在走向的终局,似乎是一个由机器人生成、也由机器人穿行的互联网——到那时,一个普通人找信息有多烦人,就再也无关紧要了。一些网站已经开始采取防御措施,把那些试图消化其内容的爬取机器人挡在门外。Reddit作为互联网上仅存的主要人类知识与观点库之一,其首席执行官史蒂夫·霍夫曼(Steve Huffman)最近宣布,平台正在转向“以登录体验为主”。在致股东的第三季度信中,霍夫曼指出,Reddit的搜索流量在十月份“平均每周减少超过一百万次引荐”。信中解释道:“网络搜索正在从通往互联网的门户,演变为一个由人工智能驱动的目的地本身。”在最近一次财报电话会议上,霍夫曼说,“人们仍然想要Reddit”,但“谷歌不再像过去那样把用户送到Reddit来了”。要求用户登录,无异于拉起吊桥,把在外面游荡的人工智能模型挡在城外。但这也意味着,随便搜搜的普通用户将更难看到Reddit上正在发生的完整对话。

In 2019, before the A.I. boom, the entrepreneur Yancey Strickler presciently outlined a “dark-forest theory of the internet.” As the web experience got worse, Strickler argued—because of “the ads, the tracking, the trolling, the hype, and other predatory behaviors”—users would retreat to more private, hidden online spaces, or “dark forests,” in search of human interaction. We once used search to find those dark forests and find the humans on the internet who were doing whatever we were interested in. Now those old tricks of Boolean operators and half-remembered keywords don’t work, because we’re meant to trust a robot to do the navigating for us instead. I often feel that the internet isn’t built for us human users anymore.

2019年,在人工智能热潮到来之前,企业家扬西·斯特里克勒(Yancey Strickler)就颇有先见之明地提出了“互联网暗黑森林理论”。斯特里克勒认为,随着网络体验不断恶化——因为“广告、追踪、网络喷子、炒作以及其他掠夺性行为”——用户会退回到更私密、更隐蔽的网络空间,也就是“暗黑森林”,去寻找人与人之间的真实互动。我们曾经用搜索来找到那些暗黑森林,找到互联网上那些和我们兴趣相投的人。如今,布尔运算符和半记半忘的关键词这些老把戏都不管用了,因为人们期望我们信任一个机器人来替我们导航。我常常觉得,互联网已经不再为我们这些人类用户而建了。

One prominent tech-commentator account on X has described companies as locked in “a knife fight for the remaining and dwindling volume of human traffic on the internet.” Many seem to have bought into the narrative that the old-school open internet is done for, so they’re scrambling to lock users into their own, A.I.-automated walled gardens. (Not to be left behind, Reddit is introducing an algorithmic For You feed and A.I.-generated short-form videos based on its content.) Google recently inspired a public backlash when it announced that it would make A.I. functions optional, rather than permanently accessible, on its software suite for schools and students, but with a catch: if schools turned off the Gemini A.I. functions, they would lose access to Google Search, too. This policy amounts to a threat: use generative A.I. or use nothing. The largest tech companies are investing so much capital in A.I. infrastructure—chips, data centers, intensively trained models—that they must make the bet pay off, one way or another.

X平台上一个颇具影响力的科技评论账号形容,各家公司正陷入“一场争夺互联网上日益萎缩的人类流量的刀战”。许多人似乎已经接受了这样一种叙事:老派的开放互联网已经完蛋了,所以他们争先恐后地把用户锁进自己的人工智能自动化围墙花园里。(Reddit也不甘落后,正在推出算法驱动的“为你推荐”信息流,以及基于其内容生成的人工智能短视频。)谷歌最近宣布,将把人工智能功能设为可选而非永久启用,应用于面向学校和学生的软件套件,此举引发了公众的强烈反弹。但其中暗藏玄机:如果学校关闭Gemini人工智能功能,他们也将失去谷歌搜索的使用权。这项政策无异于一种威胁:要么用生成式人工智能,要么什么都别用。最大的几家科技公司正在人工智能基础设施上投入巨资——芯片、数据中心、高强度训练的模型——他们必须让这场豪赌以某种方式得到回报。

The upshot is that, more often than not, I find myself using A.I. to fight back against aggressive A.I. Instead of using the Gmail app on my phone, I’ve been texting my questions about my own inbox to an A.I. agent called Instinct, which I recently granted access to my account. (It’s a security risk, to be sure, but one I was willing to take to see if the product is any good for techno-normies.) In a matter of seconds, Instinct can dig up an event invitation or a note from someone whose name I can’t quite remember. Instead of Googling potential interview subjects, I have Claude scrape for their contact information. But I would really prefer if Search just worked the way it did in the first place. ♦

结果是,我发现自己越来越多地借助人工智能来对抗咄咄逼人的人工智能。我不再用手机上的Gmail应用,而是把自己收件箱里的问题通过短信发给一个叫Instinct的人工智能代理,我最近刚授权它访问我的账户。(这当然有安全风险,但为了看看这款产品对普通技术用户来说到底好不好用,我愿意冒这个险。)几秒钟之内,Instinct就能帮我翻出一份活动邀请,或者一个我记不太清名字的人发来的便条。我不再用谷歌搜索潜在的采访对象,而是让Claude去抓取他们的联系方式。但我真正希望的,还是搜索能像最初那样好好工作。 ♦

2 free previews left — register to get 30 translations a month