The Feed Is Leaking From Both Ends: What the Scroll Decline Actually Means for Your Media Plan

For over a decade, time spent on social media moved in exactly one direction, and then it stopped. GWI's tracking of 250,000 users across 50 countries shows daily social time peaked in 2022 at about 151 minutes and has slipped every year since, with the Financial Times putting developed markets down nearly 10% from the peak. Deloitte's Australian research found the same drop and named the causes: low-quality AI content and mental-health concerns. A popular explanation making the rounds says AI assistants did it, that scrolling minutes are simply migrating to chatbots. That's half right, and the half it misses is the one your media plan should be worried about. AI shows up on both sides of this ledger: it's pushing people out of the feed and pulling them somewhere else at the same time.

The push: slop broke the scroll

The decline began in 2022, before mainstream assistants existed, driven by quality fatigue and deliberate self-regulation; Ofcom found 48% of young adults saying they spend too much time on social platforms and 68% actively managing it down. Then generative AI poured accelerant on the quality problem. Feeds filled with synthetic content, engagement rates fell year over year across Facebook, Instagram, TikTok, and X in Rival IQ's multi-million-post benchmarks, and the surprise that once powered the scroll (real people, real moments) got diluted by an infinite supply of the fake kind.

Here's the uncomfortable part for marketers: we're funding the leak. Surveys show 94% of marketers plan to use AI for content creation, largely to feed algorithms that demand volume, and the early performance data already punishes it; on LinkedIn, AI-assisted posts gained engagement while fully AI-generated posts lost it. The industry is mass-producing the exact substance driving users off the channel it's buying. If that sounds like our tasteslop essay wearing a media plan, it is.

The pull: assistants hired away the jobs

The other half of the story is the substitution the popular take gets right. Attention is zero-sum inside sixteen waking hours, and the jobs people once hired the feed for (a quick answer, ten minutes filled, what should I buy) are, in Clayton Christensen's jobs-to-be-done terms, being hired away by a tool that does them on demand without the ads or the doom. The migration concentrates precisely where marketing value concentrates: the high-intent moment. Recommendation queries that once meant scrolling reviews or asking the feed now go to an assistant that returns one answer, and even classic search is bleeding the same way; pages ranking at the top of Google now lose most of their expected clicks when an AI Overview answers above them, with reported click-through drops around 58%. Meanwhile 46% of Gen Z already prefers social platforms to search engines for finding information, which tells you the deeper truth: discovery goes wherever answers live, and answers are moving again.

The honest read of the data

Before anyone eulogizes the feed, the counter-evidence: DataReportal's global average has hovered near two and a half hours for three years, suggesting a plateau (an attention ceiling) more than a collapse, and the declines concentrate in developed markets and younger cohorts. Social isn't dying; community, culture, and top-of-funnel discovery still live there, and for some categories they will for years. What's changed is the direction of travel and where the leak is sharpest: the developed-market, high-intent minutes that considered-purchase brands actually pay for. A channel can keep its audience and still lose its decisions.

What this does to the media math

The strategic shift underneath the minutes is a change in the unit of competition. The feed sold attention by the impression: many exposures, rented at auction, forgettable by design. The assistant awards attention by the answer: one recommendation, winner-take-most, earned by whether machines can read you, quote you, and trust your receipts. You could dominate the old game and still be absent from the new one, because nothing about impression share transfers to citation share.

The playbook follows from the diagnosis. Measure your answer-layer position first, because almost nobody has; our Beacon audit runs your category's real buying prompts across five AI engines free and shows where you stand. Fix machine legibility (the majority of AI crawlers can't even render JavaScript-heavy sites) before spending anything on content. Publish answer content with evidence instead of feed content at volume, since the volume game is now literally the slop that's emptying the room. And keep social for what it still does (discovery, community, culture), just stop treating it as the default line item for decisions it no longer hosts. The scroll built the last decade of marketing. The answer is building the next one, and it's early enough that showing up counts double.