Where the ROI Is

The gap that matters is not attention growth, it's attention versus dollars

Most conversations about changing media habits stop at describing the change: more video, more AI, less TV. That framing misses the number that actually tells a business owner what to do next, which is not where attention is, but where attention and advertiser dollars diverge. eMarketer's comparison of US ad revenue against US adults' daily media time gives the cleanest single example: Facebook and Instagram together capture over 20% of US digital ad revenue while accounting for only about 7% of adults' daily time with media. That is not underinvestment in a growing platform, the story usually told about attention shifts. It is the opposite: a threefold overspend, dollars still chasing a channel well past the point where the audience's actual time has caught up.

That single data point reframes the whole exercise. The useful question for any business is not "where is attention growing" but "where does my audience's attention share currently outrun the ad dollars, mine or my competitors', actually going there." Overspending on a familiar, easy-to-buy channel because it is measurable and easy to buy is a real and common failure mode, not a hypothetical one, and Meta's own numbers prove it happens at platform scale.

Facebook and Instagram capture over 20% of US digital ad revenue while accounting for only about 7% of adults' daily media time, a threefold overspend, not an underinvestment story.

Search's quiet fracture: from blue links to zero-click and cited answers

Classic search has not disappeared, and the evidence does not support pretending otherwise. Google still holds roughly 90% of global search query volume as of 2025, per StatCounter data cited on Wikipedia, and that dominance has not meaningfully eroded. What has changed is what a search result does once someone reaches it. Pew Research Center analyzed 68,879 real searches from 900 US adults' actual browsing and found that when an AI summary appears above the results, a traditional link gets clicked only 8% of the time, compared with 15% when no summary is shown, and the session ends with no click at all 26% of the time versus 16%. Only 1% of visits click a link inside the AI summary itself. AI summaries appeared in about 18% of the Google searches studied and were seen by 58% of the study's participants, as of March 2025.

Ahrefs, working independently from a 300,000-keyword sample of aggregated Search Console data, found the same direction of effect at a starker magnitude: the click-through rate for holding the number-one organic position on keywords that trigger an AI Overview fell from 0.073 in December 2023 to 0.016 by December 2025, a 58% relative reduction, versus a smaller drop on keywords that never trigger a summary. Ahrefs itself is explicit that the 58% figure comes from a single vendor's aggregated panel data, not an independently replicated study, even though Pew's separately-fielded research corroborates the direction. Read together: ranking is not disappearing as a discipline, but the payoff for a top ranking, in raw clicks, is shrinking on the queries where an AI answer shows up, and that share of queries is only growing.

News found its new front door, and it is not a website

The Reuters Institute's Digital News Report 2026, drawing on a 48-market survey, recorded something that had not happened before: social media and video networks became the single most-used way people access news, ahead of both television and news websites and apps. The weekly figures are close, 54% for social and video networks against 52% for TV and 51% for news sites and apps, which matters as much as the ranking itself. This is not one channel collapsing while another one wins outright. It is three roughly comparable-sized doors into the same content, with the newest of the three now slightly ahead.

Pew Research Center's own August 2025 survey of US adults puts numbers behind which platforms specifically carry that traffic: 53% of US adults get news from social media at least sometimes, with Facebook the leader at 38% regular use, YouTube close behind at 35%, and Instagram and TikTok tied at 20% each. X trails at 12%. Underneath that reshuffling sits a faster and less-discussed decline: organic search referral traffic into publishers, per Chartbeat data cited in the Reuters Institute's own 2026 trends reporting, fell 33% globally and 38% in the US between November 2024 and November 2025. The Reuters Institute itself frames the AI-chatbot figure as small but fast-growing: 10% of respondents now use AI chatbots for news weekly, up from 7% a year earlier. None of these figures are large enough on their own to justify abandoning a channel. Together they describe a front door that has quietly moved.

Where the daily minutes actually go

Time-use data gives a blunter, more durable picture than any single platform's growth curve, because minutes in a day are a fixed budget in a way ad impressions are not. DataReportal's Digital 2025 Global Overview, drawing on GWI and Kepios data, put adult internet users online an average of 6 hours 38 minutes daily as of Q3 2024, with TV viewing at 3 hours 13 minutes, mobile internet at 3 hours 46 minutes, and desktop or computer use at 2 hours 52 minutes. Global ad spend sat at roughly $1.1 trillion in 2024, growing 7.3% year over year, of which social media alone captured about $240 billion, close to three in every ten digital ad dollars, growing at 15% year over year.

The follow-on Digital 2026 report, released in October 2025, sharpens the picture within "social" itself: weekly time on social and video feeds averaged 18 hours 36 minutes, and on a daily-app basis TikTok led at 1 hour 37 minutes, ahead of YouTube at roughly 1 hour 23 to 26 minutes, Instagram at 1 hour 13 minutes, Facebook at 67 minutes, and WhatsApp at 59 minutes, measured on Android devices in August 2025. That same report recorded the 80.3% figure for search-engine use in the past month, a record low for the metric. Audio tells a parallel story that gets less attention in most media-planning conversations: Edison Research's Infinite Dial 2026 found 58% of Americans 12 and older heard a podcast in the last month (167 million people) and 45% in the last week (130 million), with online audio reaching 81% of Americans monthly and 76% weekly. Reach among the 55-plus demographic specifically rose from 52% in 2024 to 70% in 2026, a segment most digital plans still underweight. Separately, the same survey found that 57% of Americans have now used a generative AI assistant.

The AI-answer layer is real, growing fast, and still not a channel you can buy

ChatGPT's growth curve is the cleanest evidence that a genuinely new attention surface has opened outside classic search and social. It reached 100 million monthly active users within two months of its November 2022 launch and scaled to 900 million weekly active users by February 2026. That scale is now large enough to matter to any business trying to be found, and it sits mostly outside the measurement and buying infrastructure that governs the rest of digital advertising.

That is the caveat this study has to hold onto. There is no independent, cross-platform census of the AI-answer layer, no equivalent of StatCounter for ChatGPT, Perplexity, Gemini, Copilot, or Google AI Mode query share. Every adoption figure in this study, including ChatGPT's own usage numbers, comes from a single vendor's or platform's self-report. That does not make the growth less real. It makes it currently unbenchmarkable against ad dollars the way television, digital display, or social spend can be benchmarked, because there is close to no addressable advertising inside the AI-answer layer today to benchmark against. A business cannot yet buy its way into a ChatGPT answer the way it can buy a search ad or a social placement. What it can do is earn a citation, which is a different discipline entirely, covered next.

There is no independent census of the AI-answer layer, no equivalent of StatCounter for ChatGPT or Gemini query share. That does not make the growth less real. It makes it currently unbenchmarkable against ad dollars, because there is close to no addressable advertising inside it to benchmark against.

Getting cited is a discipline now, not a guess

Generative Engine Optimization, the practice of restructuring content specifically to increase the odds it gets cited inside the answer an AI engine gives, is no longer speculative. The 2024 paper that introduced the term and the GEO-bench benchmark, from researchers at Princeton, Georgia Tech, and the Allen Institute for AI, and accepted at KDD 2024, found that GEO techniques boosted content's visibility inside generative-engine responses by up to 40% on that benchmark. The paper is explicit that effectiveness varies by domain, and no independent census exists yet to say how those benchmark gains translate into real query-share across live platforms in a given industry.

That combination, a peer-reviewed, benchmarked technique with a real effect size, sitting inside a measurement environment nobody can yet fully audit, is the accurate way to describe where things stand. It is not a reason to wait. Position inside an AI answer works the same way position on a search page has always worked: the mere-exposure effect does not care whether the exposure came from a blue link, a feed impression, or a citation inside a synthesized answer. Repeated, credible presence inside the surfaces an audience actually consults builds preference regardless of the mechanism. The absence of a census is a reason to start building citation-worthy presence now, while it is still early and comparatively uncontested, rather than a reason to treat the layer as unproven.

Where advertiser dollars are actually moving right now

Advertisers are not standing still while all this happens, and the direction their budgets are moving is instructive on its own. The IAB's 2026 Outlook forecasts total US ad spend growth of 9.5% in 2026, up from 5.7% in 2025, with the fastest-growing categories being social (+14.6%), CTV (+13.8%), and commerce media (+12.1%). The IAB and PwC's Internet Advertising Revenue Report puts 2025 total US internet ad revenue at $294.6 billion, up 13.9% year over year, of which social media alone was $117.7 billion, up 32.6% year over year.

That reallocation is not random. Commerce media and CTV both carry strong first-party, purchase-intent signal, which is exactly the kind of measurable, attributable inventory that survives budget scrutiny in a climate where marketers are increasingly asked to defend spend. It also means the surfaces gaining share of advertiser dollars are not, in every case, the same surfaces gaining the most share of raw attention. Messaging apps, for instance, hold real daily time, WhatsApp alone accounts for 59 minutes of daily Android use, but none of the ad-spend sources reviewed for this study track a comparable advertising category for messaging at all. That is a structural blind spot in any dollars-versus-attention index built only from paid-media data, and it is one more reason a business needs its own read of where its specific audience's time goes, not just the aggregate advertiser trend.

Why raw attention converts into preference at all

None of the reallocation argument in this study works without a mechanism connecting attention to commercial outcome, and that mechanism is well established, not new or speculative. The mere-exposure effect, documented across decades of replicated social-psychology research beginning with Zajonc's 1968 experiments, shows that repeated exposure to a stimulus, independent of its content or message quality, reliably increases liking and preference for it. That holds across words, faces, images, and brands.

The practical implication is that attention share is convertible into preference regardless of which surface carries it, a feed impression, a search snippet, a podcast mention, or a citation inside a synthesized answer. That is precisely what makes the attention-versus-dollars gap actionable rather than academic. If exposure builds preference regardless of format, then a business's return on a given unit of visibility effort is highest on the surfaces where its audience's attention share is large relative to how much competitive noise, paid or earned, currently occupies that same surface. That is the surfaces flagged throughout this study: short-video and audio on a raw-minutes basis, and the AI-answer layer on a competitive-crowding basis, since almost nobody has built deliberate citation-worthy presence there yet.

What this means for where you invest next

Put together, the evidence supports a specific and disciplined answer, not a call to chase every new platform. First, audit for overspend before chasing new spend: Meta's threefold gap between ad-revenue share and time share is a warning that familiar, easy-to-buy channels can absorb a business's whole budget well past the point of diminishing return, simply because they are measurable and habitual to buy. Second, classic search discipline still matters, Google still carries roughly 90% of query volume, but the payoff for ranking well is shifting from clicks toward citations, and a business's SEO investment increasingly needs to also answer for how it shows up inside an AI summary, not only on the results page beneath it. Third, the surfaces with real daily-minute weight and comparatively light advertiser competition right now, short-video, streaming and podcast audio, and messaging, deserve a harder look than most current budgets give them, particularly among older audiences where Edison Research recorded online-audio reach climbing from 52% to 70% in two years.

Fourth, and most important for the medium term: the AI-answer layer cannot yet be bought, measured against a census, or benchmarked the way the rest of this study's channels can. That is not a reason to wait. It is the argument for building citation-worthy, GEO-informed presence there now, while the competitive field is still thin, using the one behavioral mechanism, mere exposure, that has never required a purchased placement to work. None of this replaces knowing your own market specifically. The figures in this study are global and US averages; what your buyers actually do, and where their attention currently outruns anyone's investment in reaching them, is a question only a direct read of your own market's attention terrain can answer.