Full disclosure: As a writer of stories that are highly dependent on search traffic, I’m naturally biased about the value of links in search results.

But as someone who also conducts a lot of web searches, I’m also convinced that the new wave of generative AI search engines—most notably Microsoft’s ChatGPT-infused version of Bing—are missing the point. Instead of fixating on direct answers to users’ questions, they should be using their natural language understanding to serve better links.

If they can’t do that, web search will remain just as frustrating as it is today—only with distracting AI-generated conversation on the side.

IN DEFENSE OF LINKS

While I don’t yet have access to Microsoft’s new version of Bing, I’ve been playing around with several alternatives, including NeevaAI, You.com’s YouChat, and Perplexity.ai.

Similar to the new Bing, they answer your queries in complete sentences, combining information from multiple sources. YouChat and Perplexity.ai can also answer follow-up questions based on the context of your original search. While they all include links in the form of citations, their goal is to make clicking through less of a necessity.

This works really well for basic queries, such as “Who won the Rangers game last night?” or “How much does Netflix cost now?” The back-and-forth nature of generative AI is also good at providing encyclopedia-like knowledge, so you can ask about a new TV show, then ask follow-ups about the cast or critical response.

But the more in-depth your queries get, the more you just need some links instead.

Let’s say, for instance, that you’re having Wi-Fi problems. Asking Perplexity.ai for help brings up some obvious troubleshooting steps, like restarting your router and toggling your device’s Wi-Fi, but it’s ill-equipped to provide any more advice beyond that. Follow-up queries only lead to vague and unhelpful answers, like “uninstall the network adapter driver,” and it becomes clear that you’d be better off with links to actual articles on the topic.

Recipes are another example. There’s been lots of chatter about how ChatGPT recipes spell doom for SEO-cluttered cooking sites. But in my experience, browsing for recipes typically involves reading a few different versions, looking at photos of the finished product, and figuring out which approach works best with my ingredients and skill level. An AI-generated recipe might be faster to look up, but I’m not sure I’d want to eat it.

There’s also an entire category of searches in which the only thing you want is a link. If you want to buy an album on Bandcamp or watch a specific video on YouTube, conversational AI is just going to get in the way.