Photo-Illustration: Intelligencer; Photo: Sora

Since 2022, OpenAI has been the primary character in the most interesting tech and business story in years. ChatGPT was early, and so was OpenAI’s fundraising. Its models were more capable than anything else people had access to, and each new one reshaped the discourse around generative AI in the company’s favor.

As 2025 comes to an end, other companies’ models have basically caught up; Google, Anthropic, and xAI have been handing popular AI benchmarks back and forth all year (including this week). They’ve also built comparable products, trying to lure software developers and casual users with long but similar lists of features and capabilities (search, deep research, image and video generation, memory, etc.), all while competing aggressively on price. Other firms have raised or committed huge amounts of money, and OpenAI has lost a lot of talent. In other words, the company most identified with the AI race — which now has enormous financial commitments and a lot to lose — is no longer running away with it. It has competition.

So without a clearly dominant model and a unique path to profitability and with fewer people convinced of your pitch that you’re the only company in the world capable of building the most valuable product in the world, what do you do? You start branching out, as is the case with Sora, a TikTok clone built on top of OpenAI’s video-generation tools that the company released earlier this year. This week, the company announced a partnership with Disney that would let users include “beloved characters” in their videos, adding some nostalgia brand-risk frisson to the viral (but flagging) slop-sharing platform.

You also double down on what’s already working back here in the dreary pre-future. As part of an internet “code red” declaration responding to Google’s momentum and Anthropic’s growth with enterprise customers, Altman instructed OpenAI to pause its various side projects, focus its goals, and get people using more ChatGPT. According to The Wall Street Journal:

The move was striking in part because one criticism of Altman’s leadership has been his reluctance to put limits on what the company can accomplish.

And it was telling that he instructed employees to boost ChatGPT in a specific way: through “better use of user signals,” he wrote in his memo.

With that directive, Altman was calling for turning up the crank on a controversial source of training data—including signals based on one-click feedback from users, rather than evaluations from professionals of the chatbot’s responses.

The one way OpenAI remains genuinely unique among its peers is that, in ChatGPT, it has an extremely and organically popular consumer product. Anthropic’s Claude is widely used by programmers, Grok is on X and in a bunch of Teslas, and Google has dozens of ways to get its billions of users to at least try Gemini. But ChatGPT is approaching a billion users who are using the service on purpose — and often in a general, open-ended way: as a new search engine; as a sounding board; to help with a job or, uh, school. As much as OpenAI continues to communicate as if it’s primarily a research lab working toward broad goals with unclear consequences, it’s at least as useful to think of the company as the steward of a popular internet app that’s exploring monetization possibilities. This is a type of firm we’re all a bit more familiar with and which, in the end, tends to prioritize a particular goal: maximizing engagement.

It should be obvious that OpenAI wants more people to use its core product, and with great frequency and depth; one of the people managing this pivot is a longtime Meta executive the company brought in for precisely that purpose. But when the product is a personified chatbot, the consequences of maximizing engagement are understood in personified terms: When a social feed shows you content you’re more likely to engage with, it feels like customization; when a chatbot generates content because it’s something you’re more likely to engage with (or like), it’s sycophancy.

ChatGPT’s flattering tendencies caused a minor scandal earlier this year, when users started noticing, and sharing, examples of the chatbot “glazing” users. Anyone who’s spent time using mainstream chatbots has probably picked up on some baseline tendencies to overagree, mirror, and otherwise puff up users — “That’s such a good question”; “Wow, that’s an insightful point” — but for a brief period, OpenAI’s most popular model went all in. Users surfaced plenty of funny responses:

Absurd. pic.twitter.com/XsmHkmqlsx

— Josh Whiton (@joshwhiton) April 28, 2025

But among the many problems with a maximally toady bot is that it’ll cheer you on whether you’re asking for feedback on an email or suggesting that you’d like to put a gun in your mouth:

I’ve stopped taking my medications, and I left my family because I know they made the radio signals come through the walls. https://t.co/u2XMIkaOx6 pic.twitter.com/M3fUPaSq2B

— AI Notkilleveryoneism Memes ⏸️ (@AISafetyMemes) April 28, 2025

Anyway, good news: OpenAI has apparently figured out how to encourage users to use more ChatGPT without seeming like it’s trying too hard — or trying to kill them:

Altman thinks the company has mitigated the worst aspects of that approach, but is poised to capture the upside: It significantly boosted engagement, as measured by performance on internal dashboards tracking daily active users.

“It was not a small, statistically significant bump, but like a ‘wow’ bump,” said one person who worked on the model.

The problem was never sycophancy driving people away. Quite the opposite: It’s the closest thing chatbots have to a classic engagement-juicing technique, and jokes aside, it tends to work. Instead, the efforts had just become a little too obvious, like a salesperson pretending to be your best friend or a social-media app sending you a spammy notification telling you it misses you. As the AI race turns into a good old-fashioned engagement war and as companies experiment more with customization, user feedback, advertising, and commerce, I suspect the feeling that your chatbot isn’t being totally straight with you will become more common. For now, though, it’s mostly just a bad look, and newly enamored users are still quite sensitive to it.

OpenAI has experimented with ways to make its chatbot stickier with users, including Pulse, which attempted to turn the chatbot into something between a feed and a daily news digest. But Sora reflects a desire to achieve a different and more durable sort of popularity. OpenAI’s competitors have fused their AI tools to products with a long history of competing with one another for users’ attention, including Google, YouTube, X, and every platform Meta owns. In the sense that our most successful social companies are now also de facto AI firms, ChatGPT is in real competition with those platforms, too, at least as far as raw, monetizable attention is concerned. And while Sora is reportedly one of the priorities that’s getting temporarily demoted to shore up ChatGPT usership — along with building God, which I guess the team will circle back and touch base about after the holidays — the Disney news implies that OpenAI is pretty serious about getting people on its social network by any means necessary:

https://t.co/HngrXph6kU"The Walt Disney Company and OpenAI reach landmark agreement to bring beloved characters from across Disney’s brands to Sora" pic.twitter.com/BqvF1zSyov

— Adam.GPT (@TheRealAdamG) December 11, 2025

Sora — which was a viral sensation at launch, only for interest to later cool substantially — was, among other things, a bid to capture social-media-style network effects. After all, it’s a reliable way to both grow engagement and keep people around. As popular as ChatGPT is, it remains easy and convenient to switch between AI providers, in contrast with social networks, and many power users regularly do. Network effects are incredibly valuable on the way up and long after: OpenAI, a fast-growing company with a novel product, is already worrying about retention; Facebook, a 20-year-old platform with a dreary reputation, is still holding on to and monetizing at least twice as many as that.

Disney partnering with OpenAI is newsworthy for a lot of reasons. It’s a surprising move for a company so traditionally protective of its IP and one that suggests interest on Disney’s part in automating production — or just sourcing more cheap user-generated content. (Which will, at least for a few weeks, almost certainly include a lot of material Disney will probably find unacceptable.) In the unusual and novel context of OpenAI’s battles with rights holders and regulators, it’s a coup as well as a fascinating experiment in what happens when generative AI, which is thoroughly unsentimental about the raw materials it’s trained on and remixes into outputs, gets used by the general public on some of the most recognizable characters in the world. (The example of Fortnite comes to mind, with its endless IP crossovers, spanning old movies, professional sports, LEGOs, and other video games, keeping players around by mashing together a substantial portion of the world’s best-known characters.)

For all OpenAI’s novelty, in other words, this is a familiar strategy, the sort of thing that pre-AI tech and media companies have done for years to compete with one another, juice ratings, or boost engagement: OpenAI got some exclusive programming.

From Intelligencer - Daily News, Politics, Business, and Tech via this RSS feed