X’s Grok chatbot has no reason to exist

Illustration: The Verge

Grok, the AI bot built into the platform formerly known as Twitter, wasn’t supposed to be like other chatbots. Grok was pitched as fun. It was pitched as funny. And it had all of Twitter — errr, X — to tap into for relevant answers. So it was supposed to be useful and responsive, too. But after trying Grok out, I’m left with one basic question: why does it exist?

Technically, of course, Grok exists because Elon Musk wanted an answer to ChatGPT. The service is a paid add-on to X; it rolled out in beta last month to Premium Plus subscribers, who spend $16 a month for it and other perks. For X, it’s supposed to ease a transition away from advertising. For users, it’s supposed to offer the kind of up-to-the-minute information with which pre-Musk Twitter was synonymous. Where most chatbots rely on data from across the web, Grok can pull information directly from X posts (formerly tweets), linking them at the end of each answer. It’s a setup that could lead to more timely answers, particularly since it’s built straight into the X interface.

I’m one of the many people who once used Twitter as a search tool, especially for local events. I’d go on Twitter to find out what that random explosion I heard (fireworks around New York City) was about or why I couldn’t suddenly text my friends while in Puerto Rico (T-Mobile went down in San Juan in 2021). I wondered if Grok could provide me with a similarly useful experience. So I paid $17.42 including tax for X Premium Plus status, resurrecting my long-defunct blue check, and gave it a try.

My first problem with Grok was, actually, its deep interface integration with X — which ended up being more messy and distracting than useful. To use Grok, you have to go into the X platform or the app; there’s no standalone site for the chatbot. Once you tap the Grok icon, you can start asking it questions, but the sidebar and the trending bar are still there. ChatGPT and Bard have sidebars that let users go back to a previous chat; Grok required me to scroll up or lose previous queries.

Grok is supposedly “rebellious” and irreverent if you toggle an optional setting, but as my colleague Liz Lopatto has noted, humor is distinctly not its strong point. When I asked it general knowledge questions, its answers were often about the same as those of Bard or ChatGPT — a query about fugu sushi, the poisonous blowfish dish, resulted in a disappointingly non-edgy warning that eating and preparing fugu can be dangerous and a suggestion to find a licensed fugu chef. When it did try to give funny answers — usually because I’d asked it something funny or cursed at it — they felt forced. I’ve had more chuckles playing around with Bard.

Grok is one of the few AI chatbots without a free version, and for general-purpose searches, there’s no clear reason to use it over free options from Bing, Bard, and ChatGPT. It answered the simple queries I tested quickly and about as accurately as the competition, but it doesn’t include features you’ll get with other paid services, like ChatGPT Plus’ text-to-image prompts. (If you ask it to generate a photo, Grok will remind you that it cannot draw anything yet.)

That means the stakes are high for Grok’s big special feature: up-to-the-minute searches. One of Grok’s unique features is a kind of X drama decoder ring. When you log in, you can get suggested prompts based on trending topics on the platform. Sometimes these perform perfectly fine: I saw Figma was trending one day, asked what the latest news was, and got a summary of the company’s deal with Adobe falling through. Sometimes, though, I still came away feeling out of the loop. Grok suggested I query “Actress Jisoo’s Upcoming Movie and Friendship with Hyeri. Search for latest news and updates from last 24 hours,” and I gave it a try, recognizing Jisoo from K-pop group Blackpink. But I had no idea who Hyeri was, and Grok didn’t fill me in — it just told me Jisoo was reading scripts and her friendship with Hyeri was “going strong.”

To its credit, Grok cleared things up when I asked a follow-up question, but I wish it gave me more context instead of repeating what posts said. The answers Grok gave assumed I would have prior knowledge of something I wasn’t even intending to really search for, and I just wanted more information because I was curious about it.

Even if my experience was mixed, Grok may be good for helping people navigate X. A co-worker told me he finds himself using Grok to get quick context about trending news he sees while on the platform. Grok’s integration into X drives people already looking at posts to get more information about what people on X are discussing. This is similar to how Copilot plans to get more users using AI-powered apps by bringing ChatGPT features into workflows with other Microsoft products.

But it also means that Grok is only as good as the underlying X ecosystem, and that increasingly seems like a problem. Under Musk, X has antagonized significant parts of the Twitter community, including some of the news organizations Grok relies on for breaking news. Twitter is still a central hub for information online, but conversations have been gradually shifting to alternatives like Threads, Bluesky, and Mastodon. Even if other chatbots don’t have access to a social network like X, they can deliver results from general web searches without worrying about nurturing their own user base. X is betting Grok doesn’t need to tap into the whole internet, but soon, it may not be enough to simply learn from people who refuse to leave X. Grok’s ideal was to be the anti-ChatGPT, but the result is that it seems to live in a bubble — trying to justify why it and its parent platform still exist.

Posted from: this blog via Microsoft Power Automate.