The Internet’s Biggest Mystery

How many people use YouTube? A lot, obviously. Probably most people who use the internet in countries where it isn’t banned. But how many is that, specifically? If you ask Google Search, which shares a parent company with YouTube, you’re presented with different answers. “YouTube has more than 2.70 billion monthly active users as of 2024,” claims a Google Featured Snippet, citing a Dubai-based marketing firm that derived the number from its own “research.” In its AI Overview, Google offers a different number. “As of March 2024, YouTube had 2.49 billion monthly active users,” the AI-generated answer claims, citing a blog post from a social-media-management company that cites a report from a small media consultancy that cites “platforms’ self-service advertising resources,” so that’s an outside estimate as well.

You won’t find a YouTube user number in the company’s regular financial filings, nor will you see top-level stats for Google Search, Maps, or Gmail — or, for that matter, useful stats about how many people are intentionally using Google’s new AI tools. If you email Google and ask, a spokesperson will tell you that “YouTube has billions of monthly logged-in users,” and that the company won’t be getting more specific than that. Maybe it’s not the most important fact in the world and isn’t of much use on its own. But it’s nonetheless a basic and knowable figure — Google knows it! — that is not known by the people contained within.

This isn’t just a Google quirk, although the company ditched top-line user numbers earlier than many of its peers. It’s becoming standard industry practice. In 2024, we know a bit less about what people are doing online, and where they’re doing it, than a casual but reasonable observer might expect. How many users does Instagram have? Meta doesn’t regularly share such a number, instead emphasizing “family daily active people” — the number of people who interact with any of its distinct but overlapping products, some of which (infinite ad-conveyer Facebook) are substantially more monetized than others (encrypted messaging platform WhatsApp). In April, Netflix announced that, starting in 2025, it would no longer be reporting quarterly membership numbers, instead focusing on “engagement” — how much content its users are consuming. How many total customers does Amazon have? None of our business, except for infrequent announcements about Prime-subscription milestones.

These are some of the biggest tech companies in the world, all of whom, have, at various opportune times in the past, bragged about how many people use their products. As profitable, publicly traded firms, their preference for different metrics is easy enough to explain. Top-line user figures are a helpful metric when you’re growing fast but losing or not making much money — it’s a suggestion to investors that a great harvest will come, or a request for more patience. Snap, another publicly traded social platform, recently reported that its Daily Active users were up 10 percent year-over-year, which is certainly a better story than just “Snap loses another $305 million.” When you’re actually making money, though, you can just talk about that instead. In April, Netflix said as much: “In our early days, when we had little revenue or profit, membership growth was a strong indicator of our future potential. But now we’re generating very substantial profit and free cash flow.”

Netflix really has become a more complicated product with different price tiers and ad-supported options. It’s true that subscriber numbers don’t tell the whole story and that there is more illuminating data that it can share. But it’s also true that subscriber numbers both never told the whole story and remain an important indicator, they’re just more annoying for Netflix now that the company has reached something close to market saturation. The first time Netflix announced a subscriber loss, in 2022, the profitable company watched its narrative flip in an instant, after which it instituted a password-sharing crackdown, growth resumed, and the narrative flipped back. But in the same way that basic user growth was a leading indicator for Netflix’s coming success, persistent user losses — even if overall viewing time kept climbing — would be a leading indicator of a very basic problem: a company that’s losing customers. They still mean something, which is both a reason they should be shared and a reason they’re not. If it starts happening, Netflix won’t be letting investors, or the public, know about it.

At Meta, raw user numbers are even more important, as is their direction of movement. Netflix sells access to movies and shows. It has to worry about competition, churn, and fatigue — about slow, mounting change. Meta, despite its best efforts, is still a social-media company. It sells access to its users. Facebook, its oldest, most popular, and best-monetized product, started plateauing years ago, and previously reported brief periods of user decline. This is cause for alarm. If users leave, Meta doesn’t just make less money, its product starts getting worse with little chance of getting better — the only reason that the users are there, after all, is to interact with other users.

As for whether Facebook is currently bleeding users, it’s hard to know for sure. Facebook certainly seems like a topped-out and neglected platform tipping into decline, overflowing with AI-generated slop and sporadic updates from people with whom its users have otherwise lost touch. Instagram’s multiyear effort to clone TikTok — a product with the more recent luxury of bragging about massive user growth — doesn’t scream confidence, either. But by measuring Facebook, Instagram, WhatsApp, Messenger, and Threads together, Meta’s been able to tell a neat and simple story anyway: The mother metric is up, and the company is making money. While public companies are required to share relevant financial information with the public, disclosure-bound tech giants have gradually arrived at a communication style that has a lot in common with their privately owned peers. X’s post-takeover metric inventions might be especially egregious, but are they that much less opaque or potentially misleading than “family daily active people.” This isn’t far from what Netflix is pivoting to:

This platform is seeing incredible usage growth! pic.twitter.com/zIOk0wNCsd

— Elon Musk (@elonmusk) January 4, 2024

The disappearance of the Active User metric from conversations around tech platforms has corresponded with the proliferation of smaller, more local bullshit metrics, like visible view counts for posts and videos, which seek to replace a waning sense of platform-wide growth — everyone is here, and more people are arriving — with lots of nearby numbers always going up. They imply platform growth without explicitly claiming it, and can also be contrived however the platform needs.

In other words, mature companies are telling the stories of aging products in the most favorable ways they can. It’s individually rational behavior with the consequence that it’s sort of hard to tell where the internet is going. Is the bizarre but defining socio-commercial environment of the 21st century — social media, search, the app economy — on the verge of total collapse? Holding steady ahead of years of incumbency? We’ll only find out for sure if and the EBITDAs go south. In the meantime, it seems fair to interpret big tech’s collective coyness as a fuzzy but bearish signal. They privately record and analyze incredible amounts of broad and granular user data. If they had the numbers they wanted, they’d share them.

Big Tech’s turn away from growth metrics makes for an unusual dynamic in the world of AI. Mostly, the new crop of AI start-ups doesn’t talk about big user numbers because it doesn’t have them. In 2009, the best news a start-up could share was that it had somehow hustled its way to a million users; in the 2020s, AI start-ups courted investors and appealed to the public with tech demos, videos, and benchmarks. There are a few start-ups with top-line figures to brag about, most notable among them OpenAI, which boasted that it hit 100 million monthly users within a year of launching ChatGPT, while search engine Perplexity claims to have 10 million monthly users. There are a few throwback vanity metric boasts, too, recalling social-era hockey-stick projections: Character.ai, a genuinely popular chatbot personality platform, recently claimed to serve “around 20,000 queries per second — about 20% of the request volume served by Google Search, according to public sources,” which, yeah, sounds like a lot!

True to recent form, though, the biggest tech companies have been mostly quiet about how many people are using their new AI products. In the context of products like Google’s Gemini or Meta AI, which is now deployed across the company’s “family” of apps, lack of disclosure becomes less a background fact than a conspicuous absence. Do they have 100 million intentional, eager users? Or do they just have “billions” by default, as hastily deployed AI features confront users like ads across Gmail, Facebook, Search, and Instagram? The tech giants are banking their futures on AI and spending tens of billions of dollars on it, so big and credible early user numbers for flagship chatbots would be, to borrow Netflix’s formulation, a “strong indicator” of their “future potential,” not to mention the viability of a whole category of software, and the ability of incumbent firms to compete within it. Individually, they have a limited but useful sense of how the first wave of new AI deployment is going; together, they could paint a fairly full picture. For now, instead, the industry’s direction and velocity are indicated indirectly and imperfectly by hints and abstractions like, for example, Nvidia’s fluctuating stock price.

Perhaps there’s a constitutional difference to account for, here. User growth is always a good endorsement of a product, but it’s doubly good for social-media platforms, where more people tend to make the product better. They’re selling a human experience. A lot of AI companies, in contrast, are selling automation: solitary efficiency, new tools, work-oriented solutions, and profit margins. The broadest pitch for social media is to join up with other people; the broadest pitch for AI is that soon you won’t need other people. Still, it seems fair to interpret a lack of basic information about how the tech giant’s AI efforts are landing — how many people are using your product — critically. Secrecy serves a purpose. Top-line user figures were already a compromise, an inflated metric that combined casual app-checkers with hard-core users, and yet now they’re treated like trade secrets. Perhaps secrecy has become an instinct, a norm, and a way of doing things. But I think the basic assumption still holds: If they had the numbers they wanted, they’d share them.