内容来源：https://www.wired.com/story/when-you-should-and-shouldnt-use-ai-to-search-the-web/

内容总结：

AI聊天机器人网络搜索功能引发关注：便利性与准确性如何权衡

随着Claude、ChatGPT和Gemini等AI聊天机器人不断升级，这些工具不仅能对话生成内容，还能替用户搜索网络。与传统搜索引擎不同的是，它们将传统网页搜索与训练数据中的知识相结合，再以生成式AI特有的流畅格式呈现答案。这种附加的AI处理环节，既带来了便利，也带来了新的问题。

何时适合使用AI进行网络搜索

对于简单、广为人知的信息查询——例如帝国大厦的高度——AI搜索是相当可靠的选择。这类信息在网上易于获取，且大语言模型倾向于“自动纠错”的概率模式，使结果包含正确信息的可能性极高。

对于复杂查询，AI搜索同样表现出色。例如“我需要知道旧金山市适合五口之家、花费不超过200美元的餐厅”这类传统谷歌搜索难以解析的问题，AI能够同时查询并整合多个来源，给出综合性的回答。而传统搜索往往只是按基本条件列出旧金山餐厅的蓝色链接列表。

但信息来源始终是关键。AI机器人从未去过旧金山，也未在那里用过餐，它们依赖的是网上由亲历者撰写的公开内容。当没有完全匹配的页面时，它们会寻找相关词汇（如“实惠”“家庭”）来生成答案。

斯坦福大学近期研究的一项发现是：当AI找到准确来源时，它非常擅长从中提取正确信息；其容易出错的地方在于未能找到准确来源（可能是因为来源不存在，或查询被误解），却仍然自信地给出答案。

因此，AI网络搜索最适合那些网上存在大量准确、最新信息的常见查询。但前提是用户始终应进行某种形式的复核。

何时不适合使用AI进行网络搜索

生成式AI机器人和大语言模型的工作方式决定了它们最擅长改写、综合和总结，但这些过程也可能引入错误。如果搜索需要精确细节——如名人引语、报告数据或设备规格——传统谷歌搜索可能更为稳妥。

如果用户希望对信息来源拥有完全掌控，传统搜索方式同样更优。将网络搜索交给AI虽然便捷，但用户依赖AI正确解读查询、找到正确来源并从中提取准确信息。这节省了时间，但过程更加不透明。

加州大学圣迭戈分校的研究人员发现，AI网络搜索有时会选择与查询更相关的来源，而非排名更权威的来源。这意味着个人博客可能因被AI判定为更匹配用户实际搜索内容而排在专业医疗网站之前——而在浏览传统搜索结果列表时，用户通常会更倾向于选择自己最信任的来源。

传统搜索仍然占优的其他领域包括：突发新闻（AI提取的数据在时效性上可能存在滞后），以及健康、法律或财务等不容出错的高风险搜索。

如果确实需要使用AI方式，务必反复核查聊天机器人在回答中引用的来源链接。如有疑问，可以要求AI说明信息来源，以确保内容并非大语言模型凭空捏造。

当然，传统谷歌搜索同样可能在网络上遇到虚假信息，但在没有聊天机器人介入的情况下，用户更容易辨别信息的真伪。对于那些准确性比便利性和速度更重要的查询，传统搜索方式仍有其不可替代的位置。

中文翻译：

如果你用过Claude、ChatGPT和Gemini这类工具，你会注意到，除了它们所做的其他一切之外，这些AI聊天机器人还能替你搜索网络。取决于你在找什么，通过聊天机器人可能更容易找到更相关的结果，这或许可以解释为什么谷歌如此急于为其搜索引擎推行AI模式。

当被要求搜索网络时，这些机器人的最新版本会将传统网络搜索的元素（它们本质上只是替你在谷歌上搜索）与它们能够从训练数据中提取的知识结合起来。然后，一切都被打包并以如今熟悉的、精致的生成式AI风格呈现出来。

差异恰恰出现在传统谷歌搜索之上附加的AI元素，这意味着AI驱动的搜索未必总是最佳选择。

何时使用AI进行网络搜索

如果你查找的是简单且广为人知的内容——比如帝国大厦的高度——那么AI搜索是相当稳妥的选择。这类信息不仅在网上容易获取，而且大语言模型所偏好的“自动纠错”概率模式意味着，你看到的结果几乎可以确定会包含正确的信息。

如果你只需要快速查一下，AI是可以的。然而，问题越冷门，你就越需要仔细审视答案。如果你问的是不太为人所知的事情，比如某部冷门电视剧中排在第五位的演员，网上可能只有一两条相关记录，而AI可能会选错。

斯坦福大学近期研究的一个要点是：当AI找到准确来源时，它非常擅长从中提取正确信息。它容易出问题的地方在于找不到准确来源（要么因为根本不存在，要么因为查询被误解），却仍然自信地给出一个答案。

当你有一个标准谷歌搜索无法解析的复杂查询时，AI搜索也很有帮助：“我需要知道旧金山适合五口之家、花费不超过200美元的最佳餐厅。”AI能够针对这样的搜索查询并整合多个来源，然后给出一个综合得当的答案。传统的网络搜索会给你一串蓝色链接列表，很可能只是按更基本的标准对旧金山餐厅进行排名。

不过，来源仍然至关重要。AI机器人从未去过旧金山，也没有在那里吃过饭。它们依赖的是网上由去过的人撰写的公开描述。如果没有完全匹配的页面，它们会寻找相关词汇（如“实惠”或“家庭”），然后生成一个答案。

对于网上信息丰富的查询，比如旧金山餐厅，AI网络搜索会让人觉得非常有用。对于不太热门的目的地和更具体的请求（“星形餐桌”），你需要更加谨慎地对待结果，因为AI串联信息的方式此时会变得不那么可靠。

因此，在始终应该做某种复核的前提下，AI网络搜索最适合那些常见的、可能在网上找到大量准确且最新信息的查询。无论是在处理搜索查询的方式上，还是在整合与综合答案的方式上，AI聊天机器人都比谷歌搜索有了显著升级。

何时不要使用AI进行网络搜索

生成式AI机器人和大语言模型的构建方式决定了它们在改写、综合和总结方面表现最佳。但这些过程也可能引入错误。如果你的搜索需要精确的细节（比如名人名言、报告中的数据或设备规格），传统的谷歌搜索可能效果最好。

如果你想完全掌控所使用的来源，去谷歌用老办法搜索也可能更好。将网络搜索交给AI很方便，但你是在依赖它们正确解读你的搜索、找到正确的来源，然后从这些来源中提取准确的信息。这节省了时间，但更加不透明。

例如，如果你知道自己最信任哪些科技新闻网站，你可以在传统网络搜索中优先考虑它们，而AI可能无法做到这一点。加州大学圣地亚哥分校的研究人员发现，AI网络搜索有时会选择某些来源，是因为它们与查询更相关，而不是因为它们被列为更具权威性。

这意味着，如果AI认为某个个人博客更接近用户实际搜索的内容，它可能会选择该博客而非专业的医疗保健网站。而在浏览传统搜索结果列表时，你做出这种选择的可能性要小得多，因为你通常会选择自己最信任的那个。

传统搜索仍然占优的其他领域包括突发新闻（AI提取的数据在时效性上有时可能存在滞后），以及你绝对不能出错的高风险搜索——例如与健康、法律或财务相关的搜索。

如果你确实觉得需要采用AI方式，请务必对聊天机器人所使用的来源进行双重甚至三重核查，这些来源应该链接在答案中。如果你有任何疑问，可以要求AI说明某个信息的来源出处，以确保它不是大语言模型的凭空捏造。

当然，传统的谷歌搜索仍然可能在网络上翻出虚假信息，但没有聊天机器人从中介入，你处于更有利的位置来辨别什么是真实准确的、什么不是。对于真实性比便利性和速度更重要的查询，老式的网络搜索方式仍有其用武之地。

评论

返回顶部

英文来源：

If you've interacted with the likes of Claude, ChatGPT, and Gemini, you've noticed that in addition to everything else they do, these AI chatbots can search the web for you. Depending on what you’re looking for, it can be easier to find more relevant results through a chatbot, which may explain why Google is so keen to push AI Mode for its own search engine.
When asked to search the web, the latest versions of these bots are combining elements of traditional web search (they're essentially just googling on your behalf) with knowledge they are able to pull out of their training data. Everything is then packaged and formatted in the now-familiar, polished generative AI style.
It's the added AI elements on top of the traditional googling where the differences come in, and that means AI-powered search may not always be the best option.
When to Use AI for Web Search
If you're looking up something simple and widely known—like the height of the Empire State Building—then AI search is a pretty safe bet. Not only is information like this easy to access online, the “autocorrect” probability patterns favored by large language models mean there's a near-certain chance that the results you see will include the right information.
If you just need a quick look-up, AI is OK. The more obscure the question, though, the closer you need to look at the answer. If you've asked something less well known, like the fifth-billed actor in an obscure television show, there might only be one or two references to it on the web, and an AI might choose the wrong one.
One of the takeaways from recent Stanford University research is that when AI finds an accurate source, it's very good at pulling the right information out of it. Where it tends to fall down is in not finding accurate sources (either because there aren't any or because the query has been misinterpreted) and presenting a confident answer anyway.
AI search is also helpful when you have a complex query that a standard Google search wouldn't be able to parse: "I need to know the best restaurants in San Francisco for a family of five who want to spend no more than $200." AI is able to query and collate multiple sources in response to a search like this, and then present a well-synthesized answer. A traditional web search would give you a list of blue links that most probably just rank San Francisco restaurants by more basic criteria.
Again though, sources are crucial. AI bots have never been to San Francisco or eaten there. They are relying on public accounts on the web, written by people who have. Where there are no exact page matches, they'll look for related words (like "affordable" or "family") and generate an answer.
For queries where there's lots of information online, like San Francisco restaurants, AI web search can feel incredibly useful. For less popular destinations and more specific requests ("dining tables shaped like stars"), you need to be more careful with the results, as the way that AI joins the dots then becomes less reliable.
So, with the caveat that you should always do some kind of double-check, AI web search works best for common queries that are likely to lead to a lot of accurate, up-to-date information online. In both the way they process search queries, and package together and synthesize answers, AI chatbots are a significant upgrade on googling.
When Not to Use AI for Web Search
The way generative AI bots and LLMs are put together means they work best when paraphrasing, synthesizing, and summarizing. But these processes can also introduce mistakes. If your search needs precise details (like a celebrity quotation, a report stat, or a device specification), a traditional Google search may work best.
Heading to Google and searching the old way is also likely to be better if you want full control over the sources used. Offloading web searches to AI is convenient, but you're relying on them to interpret your search correctly, find the right sources, and then extract accurate information from those sources. It saves time, but it's more opaque.
If you know the websites you trust most for technology news, for example, you can prioritize them in a traditional web search in a way that AI might not be able to. Researchers from UC San Diego have found that AI web searches will sometimes pick sources because they're more relevant to the query, rather than because they're ranked as more authoritative.
That means a personal blog might get chosen above a professional health care site if the AI deems it to be a closer match to what the user is actually searching for. This is a choice that you would be much less likely to make when scanning a traditional list of results, as you'd typically pick the one you trusted the most.
Other areas where traditional searching still wins out are for breaking news (there can sometimes be a lag in the timeliness of the data AI pulls up) and for high-stakes searches where you absolutely cannot make a mistake—searches related to health, the law, or your finances, for instance.
If you do feel like an AI approach is needed, be sure to double- and triple-check the sources the chatbot has used, which should be linked inside the answer. If you're in any doubt, you can ask the AI for clarification on where something was sourced from, to make sure it isn't an LLM invention.
Classic Google searches can still turn up false information on the web, of course, but you're in a much better position to discern what's true and accurate and what isn't without a chatbot in the middle. For queries where that's more important than convenience and speed, there's still a place for the old way of searching the web.
Comments
Back to top