The great unbundling of attention

Human attention and the open web are getting a divorce. It's been happening slowly, then all at once. The settlement is straightforward: people get social platforms with their algorithmic feeds and video, bots get the open web with its stable URLs and crawlable text. Neither side is particularly happy about it, but the arrangement works because each audience found what it needed elsewhere.

This isn't a story about AI ruining the internet. Generative answers in search accelerated the timeline, but the underlying force predates ChatGPT by years. People were already leaving—not because they disliked reading, but because recommendation algorithms on social platforms learned to serve up interesting content more reliably than clicking around the web ever could.

The web didn't lose. It just changed jobs.

Why feeds won the attention economy

The numbers tell the story with unusual clarity. As of February 2025, the average person worldwide spends 141 minutes per day on social media—that's two hours and twenty-one minutes of scrolling, watching, and sharing. More revealing: this time now exceeds broadcast and cable TV viewing by 40 minutes.

TikTok's rise, in particular, represents one of the fastest attention migrations in computing history. Globally, users spend an average of 95 minutes per day on TikTok—more than any other social network. In the US, that number is a more modest but still staggering 52 minutes daily. The Washington Post tracked what happens when people start using the app: occasional users who initially spent around 30 minutes per day doubled their usage within months, reaching over 70 minutes daily, while power users now spend more than four hours per day scrolling.

The decisive advantage wasn't format alone, though video certainly helped. It was the combination of format and distribution. Social platforms cracked a harder problem: how to match content to interest at scale, continuously, without requiring users to know what they're looking for.

Search assumes you know your question. Feeds assume you don't, and they're usually right. The average person doesn't wake up with a list of queries. They wake up wanting to be interested, surprised, or entertained. Algorithmic curation turned out to be shockingly good at that job—so good that most people now spend more time in feeds than they do searching or browsing.

The open web couldn't compete on those terms because it was never designed to. It's a library, not a feed. Libraries are useful when you know what you need. Feeds are useful when you want someone else to decide.

The web's new primary audience

As human browsing declined, something else increased: machine reading. The collapse in web traffic has been swift and brutal. Zero-click searches—queries that don't result in any website visits—now account for approximately 60% of all Google searches. When Google's AI Overviews appear in results, click-through rates plummet to just 8%, compared to 15% for traditional search results.

The impact on publishers has been devastating. Of the top 50 U.S. news websites, 37 experienced year-over-year traffic declines in May 2025. CNN's traffic fell from around 440 million visits in 2024 to roughly 311-323 million by mid-2025. Forbes lost 50% of its traffic year-over-year in July 2025, Business Insider dropped 40-48%, and The Sun plunged 55-59%.

Even reference sites—the supposed bedrock of the web—aren't immune. Wikipedia's organic search traffic fell from 5.8 billion visits in January 2022 to 4.3 billion in March 2025, a 26% decline representing 1.5 billion fewer monthly visits.

But here's what matters: while human visits declined, machine reading increased. Public pages—documentation, product specs, comparisons, policies—are now crawled constantly by AI systems building the knowledge base that powers conversational answers. The open web has become infrastructure: fewer visitors, more citations.

That shift changes what publishing there means. You're no longer writing primarily for people who stumbled across your page via Google. You're writing for models that read everything you publish, synthesize it against everything else they know, and decide whether to recommend you when a person asks a question three conversational turns deep.

This is not a demotion. Infrastructure is valuable. But it requires different editorial instincts: clarity over cleverness, specificity over aspiration, freshness over evergreen. The goal is to be the default source when a model needs a fact it can cite with confidence.

Humans want curation, machines want coverage

People respond to narrative, pacing, personality, and parasocial connection. They binge video series, follow creators, and participate in community. That's why brand effort is shifting toward YouTube, TikTok, newsletters, and podcasts—the places where human attention actually concentrates.

Models respond to precision, structure, and breadth. They reward thorough FAQs, unambiguous claims, tables that compare options, and documentation that doesn't require interpretation. They don't care about your brand voice; they care whether you answered the question directly and whether other sources corroborate your answer.

Trying to serve both audiences with the same content is why so much of it fails at both jobs. The video essay that works on YouTube dies on your blog. The detailed implementation guide that satisfies a model bores a human. The medium and the audience have diverged; the strategy should follow.

What this means for commerce

If the web becomes primarily an interface layer for agents and bots, then the logical endpoint isn't just information retrieval—it's transaction. When someone asks ChatGPT to "find and book a flight to Denver under $400," the model doesn't just recommend options. Eventually, it completes the purchase.

That future is closer than most businesses realize. The question isn't whether agents will transact on the web; it's whether they'll know enough about your product to recommend it, and whether your site is structured so they can complete the transaction without friction. Visibility to models becomes table stakes. Transactability becomes the moat.

Brands that still think of the web primarily as a place to impress human visitors are preparing for a world that's already gone. The web is where you make yourself legible to machines—so that when people talk to those machines, they hear the right story about you.

Two jobs, two strategies

The implication is practical, not philosophical. Weekly screen time on social media crossed 18.8 hours globally in 2025, and that time isn't coming back to the open web. Meanwhile, Google processes between 9.1 and 13.6 billion searches daily—but the vast majority never result in a click.

Humans congregate where the feed is strong and the format is rich. Models congregate where the corpus is public and the structure is clean. Stop trying to make one content strategy serve both.

For people: invest in video, creators, serialized storytelling, and owned communities inside the walled gardens. Measure engagement, retention, and the strength of the relationship you're building.

For machines: invest in comprehensive, up-to-date, machine-readable reference on the open web. Measure crawler frequency, model citations, and the lift in branded search or direct traffic that follows when a model starts recommending you.

The earlier you accept this split, the less effort you waste trying to make blog posts do jobs they were never built for. The web isn't dying. It's just not for us anymore.