Eating

In brief

ChatGPT attracts more traffic than the next nine AI tools combined, with 5.5 billion visits crushing Gemini and Claude.
Chinese startup DeepSeek exploded from 33.7 million to 436 million monthly visits in just four months, operating at 1/30th the cost of Western models.
Traditional sectors are hemorrhaging users: freelance platforms down 14%, educational sites down 19%, and Chegg collapsed 64% as students turn to AI.

The latest traffic data from Similarweb reveals an uncomfortable truth about the generative AI market: While the world debates which AI model is technically superior, ChatGPT has already won the user adoption war by a margin so vast it defies conventional competition metrics.

With 5.5 billion visits in May 2025, ChatGPT commands roughly 80% of all generative AI traffic. To grasp this scale: ChatGPT gets more visits than Google’s Gemini, DeepSeek, Grok, Perplexity, and Claude combined. Then doubled. Then add another few millions for good measure.

Image: Similarweb

ChatGPT surged past 500 million weekly active users in late March, and ChatGPT’s mobile app already averaged more than 250 million monthly active users between September and November 2024. OpenAI’s partnership with Microsoft has certainly helped, but the scale suggests something more fundamental: ChatGPT has become the default AI assistant for hundreds of millions of users worldwide.

What’s particularly interesting is ChatGPT’s resilience. Despite a brief dip in traffic during early 2025, OpenAI quickly reversed the trend, surging back to new heights.

China’s DeepSeek: Neither gone nor forgotten

Another surprising data point: Chinese startup DeepSeek—banned on most U.S. government-issued devices and many institutions due to fears its feeding info to China—rocketed from 33.7 million monthly users in January to 436 million visits by May. That’s a 13x increase that would make venture capitalists weep salty tears.

Though OpenAI raised billions of dollars to accumulate as many GPUs as possible to fuel its models, DeepSeek operates at a fraction of the cost of Western models; its input tokens cost $0.55 versus OpenAI’s $15. The feisty upstart achieved this while being forced to use lesser Nvidia H800 chips due to export restrictions, the computational equivalent of winning a Formula 1 race with a Toyota Corolla.

Image: ArtificialAnalysis

DeepSeek’s geographic dominance tells the story: DeepSeek is building an empire in markets that Western AI companies have barely touched. The top three markets, according to Similarweb—China, India, and Indonesia—account for over 51% of its users.

The best of the rest: Google and Anthropic

For a $2 trillion company that processes 3.5 billion searches daily, Google’s AI performance is pretty pedestrian. Gemini’s 527.7 million visits barely edge out DeepSeek despite every conceivable advantage: billions in funding, integration across Google’s ecosystem, and access to more users than any platform on Earth—plus a monopolization of the AI browser and partnerships with major brands to boost the model’s adoption.

GenAI Traffic Share update —

🗓️ 6 months ago:
ChatGPT: 87.5%
Google: 5.4%
Perplexity: 2.0%
Claude: 1.6%

🗓️ 3 months ago:
ChatGPT: 77.6%
DeepSeek: 8.1%
Google: 4.9%
Grok: 2.7%
Perplexity: 2.0%

🗓️ 1 month ago:
ChatGPT: 80.2%
Google: 6.1%
DeepSeek: 5.9%… pic.twitter.com/y9bsckoyEF

— Similarweb (@Similarweb) June 10, 2025

Still, it should not be counted out: Gemini recorded 284 million visits in February, which means the model is growing in popularity—a good metric if it becomes a trend, considering its current state.

Perhaps the most unexpected finding is Claude’s poor performance. Despite backing from Amazon and Google, technical superiority in many benchmarks, and constant praise from AI researchers and people in Academia, Claude attracted a bit less than 100 million visits in May 2025.

Claude had 18.9 million monthly active users worldwide as of early 2025, and while the surge is definitely a major increase in traffic, these numbers pale in comparison to ChatGPT’s billions.

The disconnect between Claude’s technical capabilities and user adoption is interesting. Claude 4 Opus scored better than its competition at complex reasoning and even creative tasks, yet it has failed to translate this into mass adoption, showing yet again that the best product isn’t usually the one everyone loves.

While Meta AI exists, its current usage as a primary AI tool may be considerably lower compared to ChatGPT. Many users do not currently gravitate towards Zuck’s chatbot as their preferred choice for AI-related tasks.

Meta’s significant contribution to the open-source community, particularly through its Llama models, is notable. However, the nature of open-source usage is distinct from direct user interaction.

The AI-powered bloodbath

While AI companies fight for position, traditional internet businesses are watching their empires crumble. For example, Similarweb reports that Chegg’s traffic collapsed 64% as students discovered ChatGPT gives better homework help for free, Quora plummeted 51%, and freelance platforms like Fiverr dropped 14%.

Image: Similarweb

On the other hand, DevOps and code completion tools surged 41% year-over-year, voice generation platforms grew 14%, and automation tools jumped 12%.

The Similarweb data delivers a brutal truth to the AI industry: technical superiority means little without user adoption. Claude may impress researchers, but ChatGPT impresses everyone else. Gemini may have Google’s distribution, but distribution without differentiation is not interesting.

ChatGPT won by being first, being good enough, and most importantly, being what users actually wanted—a conversational AI that just works. While competitors focused on benchmark scores and safety protocols, ChatGPT focused on being useful. The 5.5 billion visits say that strategy worked.

So OpenAI may lose in some benchmarks, but in the race for global adoption, it has already crossed the finish line. The real competition now is for second place.

Edited by Andrew Hayward

Generally Intelligent Newsletter

A weekly AI journey narrated by Gen, a generative AI model.

Source link

After posting a “looksmaxxing” video, Imeh, who is Black, says he received racist comments, including ones telling him to “just be white.” Upon realizing that the community was “toxic and racist,” Imeh pivoted to anti-looksmaxxing content and then stumbled on “SkinnyTok” and pro-eating disorder communities on X.

“It’s way easier to find SkinnyTok, eating disorder TikTok, than recovery,” Imeh said. “I went for a different approach, telling them the side effects and what could happen if you’re not eating.”

Imeh’s videos mention more extreme potential health outcomes of eating disorders, like organ failure and hair loss. But he also sometimes mocks the messaging found in pro-eating disorder communities. In one TikTok, he’s eating with a text overlay that says “none of your friends are gonna be jealous that your Ed made you look like a skeleton baby pick up the fork.”

Imeh says he’s not trolling, but stating “literal facts” that he doesn’t sugarcoat. “One thing I noticed in Gen Z, especially my generation, is that they will only stop doing something if they’re embarrassed by it,” he says.

“I got a lot of people emailing me and DMing me like ‘Stephen you’ve helped me so much with my eating disorder.’” He has over 70,000 followers, many of whom began following him after he took on eating disorder communities.

Pillepich says she can see a modern, “chronically online” approach working to redirect attention and ideally get people who need it into recovery.

“Leading with nuance doesn’t get people’s attention. It does have to be more extreme, more funny, whatever it is,” Pillepich said. “If that gets someone to the first step of seeing a dietician, a therapist, working on the deeper issues, then that’s great, too.”

Breithaupt said that content that is too judgmental or makes people with eating disorders feel ashamed could make them less likely to get help. “The most effective anti-ED content tends to validate the pain beneath the disorder while still rejecting the behaviors,” she says.

“When content creators use humor or mockery to push back against pro-ED culture, there’s a real risk that viewers—especially those actively struggling—won’t just see the disorder being criticized, they’ll feel like they are being mocked.”

E said that TikTok content like Imeh’s helped her realize how “stupid” SkinnyTok was. She said she has started watching eating disorder recovery content, instead. But the algorithm still shows her “harsh motivation” for weight loss in addition to recovery videos.

In late 2024, TikTok banned a controversial weight loss influencer whose content glorified extreme thinness. E thinks TikTok should ban more of “SkinnyTok,” although pro-eating disorder communities have historically migrated to other platforms when that happens.

Eating disorder recovery practitioners say that posting anti-”SkinnyTok” videos on the same platform is likely helpful, but that it’s only a first step.

“What I work with most people on is limiting social media,” Breithaupt said. “Doing something else rather than engaging in social media is more helpful toward recovery, even if you’re watching recovery-oriented videos.”

The National Alliance for Eating Disorders Helpline provides support, resources and information about treatment options at 1-866-662-1235, Monday through Friday. You can also text “ALLIANCE” to 741741 if you are experiencing a crisis to be contacted by a trained volunteer. More information about eating disorders, including other free and low-cost support options, can be found on the National Eating Disorder Association’s website.