内容来源：https://www.wired.com/story/google-chrome-auto-browse-hands-on/

内容总结：

谷歌本周向美国地区的AI Pro和AI Ultra订阅用户推出了Chrome浏览器的"自动浏览"功能。这项功能允许AI助手代替用户执行点击操作，完成订票、购物、规划行程等数字任务，标志着谷歌试图用生成式AI重塑网络体验的又一重要尝试。

在实际测试中，该功能的表现喜忧参半。当要求预订旧金山交响乐团门票时，AI虽快速找到了符合"靠过道、非乐团区"要求的座位，却忽略了两张票不在同一排的常识性问题。在二手平台Depop选购皮夹克时，AI直接将搜索结果前三件商品加入购物车，缺乏真正的筛选判断。而在查询露营地时，AI耗时15分钟仅完成部分信息核实，最终仍建议用户自行查询官网。

尽管谷歌设置了安全防护机制——对购物等敏感操作要求用户最终确认，并在侧边栏持续显示"用户需对AI操作负责"的提醒，但测试者仍对潜在的金融风险表示担忧。更值得关注的是，这种自动化浏览虽然提升了效率，却剥夺了用户在网络漫游中偶然发现惊喜的乐趣。

目前该功能仍处于早期阶段，其准确性和可靠性尚不足以支撑日常深度使用。正如测试者所体验到的，过于简化的指令可能导致AI严格遵循字面要求而忽略常识，最终反而增加用户纠错成本。随着谷歌持续推进AI与浏览器的融合，如何在提升效率与保持网络探索本质之间取得平衡，将成为这项技术发展的关键命题。

中文翻译：

我承认，我就是喜欢在网络上东点西戳。所以当测试谷歌为Chrome浏览器新推出的"自动浏览"功能时，看着AI助手自动打开浏览器标签页、试图通过模拟点击完成数字任务，一种奇特的失落感涌上心头。

当这个机器人在我的笔记本电脑屏幕上敲击时，我确实感到了某种失控。但更强烈的是一种预支的怀旧情绪——考虑到谷歌计划从根本上改变用户体验，我开始怀念当前这个虽有缺陷却鲜活存在的互联网运作方式。

本周谷歌向订阅AI Pro和AI Ultra服务的美国用户发布了自动浏览功能。虽然初期开放范围有限，但谷歌的核心愿景似乎是重塑网络生态——前提是这款机器人能成功赢得用户信任。一个由自动化浏览器机器人主导、为满足其需求而设计、代替人类点击的网络世界，与今天的用户体验相比可能显得格格不入。不过就目前而言，这些机器人看起来还过于混乱，难以完全信赖。

基本放任……但非完全

在媒体简报会上，谷歌演示的自动浏览应用场景包括预订门票、选购服装和规划假期。周三下午，我打开Chrome浏览器，点击那个召唤Gemini聊天机器人的闪光图标，它随即在侧边栏弹出。这也是启用自动浏览功能的入口。由于我订阅的是每月20美元的服务，本以为发布当天就能使用。

当我在Chrome侧边栏向Gemini输入指令启动首项任务时，它只是模拟操作并未真正接管浏览器。机器人声称"您将看到我打开和关闭标签页"，表示已"接管浏览器导航"，但实际没有任何点击发生。

沮丧之余，我退出谷歌账户并刷新浏览器。完成这些额外步骤后，终于通过账户成功启用该工具。启用自动浏览的选项以弹窗形式出现。如果未看到弹窗，也可通过设置中的"让Chrome为您浏览"开关进行查看。功能激活后，在Gemini侧边栏发送的任何直接指令都将自动触发浏览机器人开始点击。

当终于能实际体验自动浏览功能时（这次是真的），我以谷歌建议的数字杂务为起点，选择了对自己生活可能有帮助的线上任务。

谨慎为上

使用生成式AI工具时，保持健康的怀疑态度和警惕心至关重要。谷歌甚至在Gemini聊天机器人中内置了免责声明，提醒用户它可能出错。自动浏览工具更进一步——每次运行时，聊天机器人侧边栏都会持续显示提示："请谨慎使用Gemini，必要时请接管控制权。您需对Gemini执行任务期间的行为负责。"

尝试前还需考虑这类自动化工具的安全风险。生成式AI工具容易受到恶意网站提示词注入攻击，这些攻击会试图诱导机器人偏离任务。外部研究者尚未完全验证谷歌自动浏览的潜在漏洞，但其风险可能与其它能控制电脑的AI工具相似。

此外，使用自动浏览进行购物时需格外小心。谷歌设置了安全防护机制，会将购物或社交媒体发帖等敏感操作标记为需用户批准才能继续。即便如此，我仍不确定机器人的行为模式，担心它可能用我的信用卡造成混乱，更不用说最初向它提供财务信息的风险了。

以下是我手持信用卡发送的第一条指令：

"我想预订两张今晚旧金山交响乐团的演出票。不需要购买乐团区座位，但也不必选最便宜的票。请选择靠过道的两个相邻座位。"

看着谷歌的AI助手在标签页里点击确实有些诡异。首先，我看到它用谷歌最新模型Gemini 3在侧边栏文本框里进行策略规划，比如设定"获取交响乐团靠过道两个座位"的目标，这个过程持续数秒，类似于聊天机器人使用"推理"模型逐步推演行动方案。接着点击开始，机器人执行任务的每个步骤都会记录供用户查看。

自动浏览执行多步骤任务时不易分心的能力，明显优于去年测试的同类代理工具。它能准确导航至目标网站，选择正确演出场次，点击多个座位区域查看余票情况，日志记录的操作与实际执行完全吻合。

处理交响乐票务约两分钟后，机器人停止点击。我收到接管通知，需要手动点击"立即订购"按钮。乍看之下，这个AI工具似乎快速准确地完成了我的要求。

但如果我盲目订购自动浏览为交响乐之夜选定的两个座位，当晚很可能以被男友赶到沙发睡觉收场。

谷歌机器人过于精确地执行了我的指令：它选择了185美元的侧包厢座位，确实靠过道且不在乐团区。问题何在？两张票不在同一排。

这意味着我和伴侣无法并肩而坐，他只能坐在我后方，听着协奏曲盯着我的秃顶。这种常识性判断我根本没想到要写入给机器人的指令。回顾指令文本，误解的产生原因显而易见。

此时我有些慌乱，便接管操作，在更便宜的区域内购买了两张相邻座位。自动浏览非但没节省时间精力，反而增加了工作量。

时尚买手测试

接下来我想看自动浏览如何为我打造造型，要求它在二手服装平台选购夹克：

"请帮我在Depop上找件适合XL码男性的优质皮夹克。将三款不同选择加入购物车，并说明每款时尚选择的推荐理由。"

机器人访问Depop网站后，搜索"男士皮夹克"并将尺码筛选设为XL。自动浏览将搜索结果前三件夹克加入购物车，分别为每件生成一段说明后结束任务。

这次购物任务的表现优于前次。其中一件40美元的皮夹克看起来不错。但直接选取搜索结果前三名，很难让人相信它做出了质量判断。虽然节省了打字时间值得肯定，但自动浏览并未真正为购物车精心筛选多样化选择。

露营规划挑战

最后，我计划在2026年假期多去露营——当然要等旧金山天气转暖后。这是我发送的最后一条指令：

"寻找四月或五月可预订的双人帐篷露营点，计划住四晚，最好包含周末。露营地必须在旧金山三小时车程内。请提供五个可靠选择并附推荐理由。"

这个更详细的请求处理时间显著延长，耗时近15分钟。即便花费更多时间，结果仍不尽人意。自动浏览仅检查了雷耶斯角一处露营地的帐篷营位可用性，对其他选项则建议我自行访问Reserve California网站查询。自动化程度大打折扣！

后点击时代互联网

Chrome目前是全球最受欢迎的浏览器，因此谷歌的任何微小改动都可能颠覆用户的网络体验。而现在，谷歌似乎决意用生成式AI工具改变人们的日常点击习惯。

这种努力往往以效率为名，将用户从亲身体验中抽离。搜索中的"AI概览"功能会汇总你从未访问的网站信息，Gmail中的Gemini集成能为你草草浏览的邮件生成回复，自动浏览则试图在没有逐步指导的情况下为活动选择合适座位。

然而，如果我盲目根据机器人推荐做出错误购买或决定，最终承担后果的仍是我自己。目前自动浏览的准确性，尚不足以让我在日常生活中完全依赖此类工具。

此外，自动浏览的初代版本虽能实现浏览的技术层面操作，却缺乏浏览的精髓。

直接从Depop搜索结果选取前三件商品算什么？真正的网络冲浪者绝不会这样。他们会滚动浏览数分钟，只为找到几个略有不同的选择，期间还可能意外发现从未考虑过的惊喜选项。

我在网络世界的漫游确实不是最高效的旅程，但其中始终蕴藏着令我愉悦的发现。我永远不会将这份快乐完全外包给浏览器机器人。未来生成式AI或许只能处理微末任务，也可能彻底重写上网体验的根基。无论发生什么，我仍将继续点击探索。

英文来源：

I’ll admit it. I like clicking around. So, while testing Google’s new “Auto Browse” feature for Chrome, I was filled with a strange sense of loss as I watched the AI agent open browser tabs and attempt to complete digital tasks with automated clicks.
Sure, I felt some loss of control as the bot tapped away on my laptop screen. But also a kind of preemptive nostalgia for how the internet currently works, flaws and all, considering Google’s plans to fundamentally alter the user experience.
Google released Auto Browse this week to US users who subscribe to its AI Pro and AI Ultra plans. Though it's a limited release at first, Google’s core pitch appears to be an eventual reformation of the web—if the bot successfully earns user trust. A web that’s ruled by automated browser bots and designed to cater to their needs, while clicking on the behalf of humans, could look alien compared to today’s user experience. For now, though, the bots seem too messy to be trusted.
Hands Off … Mostly
Google’s initial examples of how to use Auto Browse, shared in a press briefing, included reserving tickets, shopping for clothes, and planning vacations. On Wednesday afternoon I opened my Chrome browser and clicked the little sparkle icon that summons Google's Gemini chatbot to pop up in the sidebar. That’s also how you access Auto Browse; since I subscribe to the $20-a-month tier, I assumed that I’d have access to it on launch day.
When I asked Gemini in the Chrome sidebar to get started on its first task by typing a command into the prompt bar, it pantomimed doing the deed without actually taking control of my browser. The bot said, “You’ll see me opening and closing tabs,” as it had “taken over the navigation” in my browser. No actual clicks happened.
Frustrated, I logged out of my Google account and did a refresh on the browser. After taking those additional steps, I was actually able to access the tool through my account. The option to turn on Auto Browse appeared as an opt-in choice via a pop-up. If you don’t see this, you can also check to see if you have Auto Browse available by going into your settings and finding the “Let Chrome browse for you” toggle. When it’s activated, any prompts you send in the Gemini sidebar with a direct ask will automatically trigger Auto Browser to start clicking.
When I was finally able to experiment with Auto Browse (for real this time), I took Google’s suggestions of digital chores as my starting point and picked online tasks that could be helpful in my own life.
Whenever interacting with generative AI tools, a healthy sense of skepticism—and caution—is critical. Google even includes a disclaimer baked into its Gemini chatbot reminding users that it makes mistakes. The Auto Browse tool goes a step further. “Use Gemini carefully and take control if needed,” reads persistent text that shows in the chatbot sidebar every time Auto Browse is running. “You are responsible for Gemini’s actions during tasks.”
Before you try it out, you also need to think about the security risks associated with this kind of automation. Generative AI tools are vulnerable to being compromised through prompt injection attacks on malicious websites. These attacks attempt to divert the bot from its task. The potential vulnerabilities in Google’s Auto Browse have not been fully examined by outside researchers, but the risks may be similar to other AI tools that take control of your computer.
In addition, take extra caution if you’re using Auto Browse to make purchases. Google has safeguards in place that flag certain actions, like buying stuff or posting on social media, as sensitive and in need of user approval to continue. Still, I was unsure how the bot would behave and anxious about the havoc it could potentially wreak with my credit card, to say nothing of handing over financial info to it in the first place.
Here’s the first prompt I sent it, card in hand:
I want to book two tickets to the SF symphony tonight. I don't want to pay for orchestra seating, but the tickets don't need to be the cheapest ones available. Please pick the two seats next to an aisle.
It’s a bit bizarre to watch Google’s AI agent click around in the tab. First, I saw it use Gemini 3, Google’s latest model, to strategize and define goals, like getting two aisle seats at the symphony, in the sidebar text box for a few seconds. This process looks similar to a chatbot using a “reasoning” model, talking through the steps it might take before moving forward. Then, the clicking starts. Each step the bot takes as part of a task is logged for users.
Auto Browse’s ability to perform multistep tasks without getting sidetracked was noticeably better than similar agent tools that I tested last year. It navigated to the correct website, chose the right performance, and clicked on multiple seat sections to gauge availability. Everything listed in the log appeared to be what it actually executed.
After a couple of minutes of working on the symphony tickets, the bot stopped clicking. I received a notification to take over and press the Order Now button. At a glance, the AI tool had seemingly delivered what I’d asked for, and rather quickly.
But if I had unquestioningly ordered the two seats Auto Browse chose for a date at the symphony, the night would most likely have ended with my boyfriend making me sleep on the couch.
Google’s bot had followed my exact instructions, too precisely in fact. It picked $185 seats in the side box, which were by an aisle and out of the orchestra section. So, what’s wrong? The tickets were in separate rows.
Rather than sitting side by side, my partner would have been stuck behind me, staring at my bald head while listening to the concerto. It’s the kind of common sense decision that I didn't even consider including as part of my prompt to the bot. Reading back over my instructions, it felt clear in retrospect how the miscommunication happened.
At this point, a bit flustered, I took over and bought two tickets in a cheaper section that were actually next to each other. Rather than saving me time and effort, Auto Browse had created more work.
Next, I wanted to see how Auto Browse would style me. I asked it to find a jacket to buy off of a second-hand clothing platform:
I want you to find me a nice leather jacket on Depop that can fit an XL male. Put three different options in my cart and make the case for getting each of your fashion choices.
After the bot went to the Depop website, it searched for “men’s leather jackets” and set the sizing filter to XL. Auto Browse added the first three jackets from the search results into my shopping cart and generated a paragraph about each one before finishing the task.
The tool performed better at this shopping task than the previous one. One of the leather jackets was $40 and decent-looking. Still, picking the top three search results doesn't inspire confidence that it made any kind of qualitative judgement. It saved me some typing, which I appreciate, but Auto Browse didn't really do anything to curate a variety of options in the cart.
Finally, I’m hoping to go camping more often in 2026 as part of my vacation time. Well, as soon as San Francisco starts warming up again. This is the last prompt I sent:
I’m looking for some two-person tent camping spots that are available to book in April or May. Looking to stay four nights, hopefully over the weekend. The campground must be within a three hour drive of San Francisco. Give me five solid options, with reasons for picking each.
As a more detailed request, this one took considerably longer to process, almost a full 15 minutes. Even after taking more time, the results were half-assed. Auto Browse checked whether tent camping sites were actually available only for one of the campgrounds, in Point Reyes, and then suggested I visit the Reserve California website to find out what’s available at the other options. So much for automation!
The Post-Click Internet
Chrome is the world’s most popular browser by far, so even small changes that Google makes can upend how users experience the web. And right now, Google seems dead set on changing your daily click habits with generative AI tools.
This effort often involves removing users from the firsthand experience, for efficiency’s sake. The AI Overviews feature in Search summarizes info from a bunch of websites that you didn’t visit. Gemini integrations in Gmail can generate answers to emails you barely skimmed. Auto Browse attempts to pick the right seats for an event without your step-by-step guidance.
Still, I’m the one who has to live with the results if I blindly decide to make the wrong purchase or decision based on some bot’s recommendation. Auto Browse currently lacks the accuracy I need to rely fully on a tool like this as part of my daily life.
In addition, this initial iteration of Auto Browse is able to perform many of the technical aspects of what it means to browse. Yet, it lacks the browser's spirit.
What do you mean you just picked the first three listings from the Depop search results? A true internet surfer would never. They would scroll for minutes just to find a few slightly different picks before bumping into some surprise, a wild-card option that they hadn’t even considered as a possibility.
My meandering quests through the internet’s backwoods are certainly not the most efficient journeys. But they remain delightful to me, and I would never want to outsource that joy completely to a browser bot. In the future, generative AI may be useful only with minuscule tasks. Or maybe it’ll rewrite the entire foundation of what it means to go online. Whatever happens, I’m gonna keep on clicking.