Six in the morning
At six in the morning a woman reaches for her phone before she pours coffee, and she skips the television while the radio stays dark. She types a question in her own words, and a machine responds to her, in her language, about her particular day.
Anthropic sampled Claude conversations hour by hour for its June 2026 Economic Index, and the measurements recovered the exact shape of that morning across millions of ordinary users. People request news early in the day, and they request advice about sleeplessness late at night. American tax questions climbed to roughly eight times the May average on 14 and 15 April, then collapsed the morning after the filing deadline.
That same report classified the output of every conversation in its sample, and the categories deserve attention. Ninety-three percent of Claude conversations finish with a recognisable artifact, most often an explanation (17 percent), a document or report (15 percent), or practical guidance (11 percent).
Read those numbers as a schedule, and then ask who now writes the schedule. For roughly a century, commercial media operated on schedules that broadcasters wrote, published, and controlled. This one operates on a schedule that ordinary people compose for themselves, one question at a time. Nobody programmed that woman’s six o’clock: she programmed it herself, and she programmed it as a conversation rather than a transmission from an institution.
What the old order lost
Count the losses of the old order in sequence. Nielsen’s Gauge measured streaming at 48.5 percent of American television time in June 2026, while cable dropped to 19.5 percent from 23.4 percent a year earlier. Nielsen flagged that particular edition as its final report under the previous measurement methodology. Hold the decimal points loosely and the direction firmly, since that direction has held steady for more than a decade.
The Reuters Institute interviewed almost 100,000 news consumers across 48 markets for its 2026 Digital News Report, published on 16 June. For the first occasion in the survey’s history, social media and video networks (54 percent) overtook the news industry’s own websites and applications (51 percent) as the most common route to information. Television news consumption has fallen 13 percentage points since 2020, and trust in journalism dropped to 37 percent, the lowest figure the Institute has recorded since it began measuring trust in 2015. Deliberate news avoidance climbed to 42 percent, and among people aged 18 to 24, a majority of 52 percent now identify social media, video networks, or AI chatbots as their principal route to the news. The same survey measured the creator layer itself, reporting that 27 percent of respondents take some news from individual news creators, while only 3 percent depend on creators alone.
The old order, measured
- 48.5% : streaming share of US TV time, June 2026. Cable held 19.5%, down from 23.4% a year earlier.
- 54% vs 51% : social and video networks now outrank publishers’ own sites and apps as a route to news.
- 13 points : the fall in television news consumption since 2020.
- 37% : global trust in news, the lowest reading since 2015. News avoidance now stands at 42%.
Nielsen, The Gauge, June 2026 · Reuters Institute, Digital News Report 2026, 16 June 2026
Radio holds firmer ground than any other legacy medium, and that number deserves a place on the page. Edison Research at SSRS surveyed 2,050 Americans aged twelve and over in January 2026 for The Infinite Dial, and 73 percent of drivers reported listening to AM/FM inside their vehicle within the previous month. Yet Edison’s Share of Ear study handed podcasting the leading share of spoken-word listening in 2026, at 40 percent, a fraction past radio itself. Reach survived. Primacy slipped away.
The influencer pyramid
The influencer economy arrived as the remedy for all of that, then rebuilt the identical pyramid it had promised to demolish. CreatorIQ published its State of Creator Compensation on 21 January 2026, drawn from payment volume across its own creator network. The top one percent of creators captured 21 percent of all creator advertising payments during 2025, up from 15 percent two years earlier, while the top ten percent took 62 percent, up from 53 percent. The median campaign paid $3,000, a decline from $3,500 the previous year, even as the average payment climbed to $11,400. Read that paradox with care, because both directions run at the same time: payments into the creator economy expanded 59 percent during 2025, while the typical creator collected less money for each individual campaign.
Goldman Sachs Research counted roughly 50 million creators worldwide and estimated that only about 4 percent of them clear $100,000 a year. That estimate dates from 2023, and the firm expects the proportion to hold steady even as the creator population multiplies.
Engagement, meanwhile, runs backwards from audience size, a relationship the measurement firms confirm year after year. Social Cat analysed more than 100,000 Instagram influencer posts for its 2026 report, and measured nano creators at a 2.61 percent average engagement rate against 1.68 percent for micro creators. The rate declines further as the accounts scale upward toward genuine celebrity. Buy a bigger megaphone, and fewer people answer it.
Mass media sold many eyes at once, and two things happened to that product together: the supply exploded, and the audience stopped answering.
The scarce good moved
I spent thirty years building machines for reach. I helped found Kaieteur News in Guyana, which grew into a national daily newspaper, and I wrote a weekly column for Stabroek News under the title Ways of Looking and Feeling. I put documentary television on air in Guyana, and I started two community newspapers in Canada. Every one of those ventures answered a single commercial question, and every balance sheet turned on the answer: how do we place one message in front of the largest possible number of people at the lowest possible cost?
That question lost its edge, and the reason sits in the capital accounts. A printing press once cost more than a house, and a broadcast licence cost far more than a press. Those capital requirements manufactured scarcity, and scarcity determined the price of human attention. Today a seventeen-year-old reaches ten million people from a bus seat, and a language model drafts a thousand posts before breakfast, so supply of reach ran toward infinity and price collapsed behind it.
So the scarce good moved, and it moved from the impression to the turn. An impression travels one direction and lands on a stranger, while a turn travels both directions and lands on a particular person who handed you a question first.
In September 2025, OpenAI’s economic research team and the Harvard economist David Deming published the largest study yet undertaken of consumer chatbot behaviour, classifying roughly 1.1 million de-identified conversations through a privacy-preserving pipeline. About 49 percent of messages ask a question, about 40 percent request a task, and about 11 percent express a feeling or an opinion. Non-work messages expanded from 53 percent of total usage to beyond 70 percent, and by July 2025 some 700 million weekly users sent 18 billion messages every week, a population approaching one adult in every ten on earth.
Anthropic recorded the same migration inside its own traffic, where personal conversations expanded from 35 percent of Claude.ai usage in November 2025 to 42 percent by February 2026. Read the verb. People ask. Nobody asks a billboard. A question establishes an obligation, and that obligation carries commercial value which an impression never possessed.
OpenAI reported 900 million weekly ChatGPT users on 27 February 2026, alongside 50 million consumer subscribers, and press reports through the middle of 2026 placed the figure near or beyond a billion. Treat the counts themselves with caution, since the company controls the meter, and treat the direction with confidence.
The new unit, measured
- 49 / 40 / 11 : the split of ChatGPT messages into asking, doing, and expressing. Half of all traffic asks a question.
- 53% to 70%+ : the rise of non-work messages as a share of consumer usage.
- 18 billion : messages sent per week by roughly 700 million weekly users, July 2025.
- 35% to 42% : personal use as a share of Claude.ai conversations between November 2025 and February 2026.
Chatterji, Cunningham, Deming, Hitzig, Ong, Shan & Wadman, NBER Working Paper 34255, September 2025 · Anthropic Economic Index, March and June 2026
A fair critic answers here, so let the objection stand at full strength. The Reuters Institute also found that only 10 percent of people worldwide consulted an AI chatbot for news in a given week, and a bare 1 percent named artificial intelligence as their main news source. That distribution still places conversation far behind television and social video in raw population terms. Mass audiences still gather on cue, and World Cup coverage lifted American broadcast share during June 2026. Podcasting, newsletters, and creator video still operate as genuine mass channels for particular audiences, and honest analysis grants them that reach. Call the old order wounded rather than buried: its reach persists as a cheap commodity available to anybody who wants it, while its authority, its pricing power, and its claim on a person’s morning have gone.
Two hundred million living rooms
Community migrated along the identical path, and the platform operators announced the migration out loud. At Social Media Week in May 2025, Discord’s vice president of sales, Adam Bauer, told an ADWEEK audience that 90 percent of the platform’s servers hold between five and twenty people, on a service counting roughly 200 million monthly users worldwide. Bauer named the shift without hedging: community operates as the product now, and follower counts no longer determine the commercial outcome for anybody involved. He added the warning that most brands ignore: Communities die quickly without ongoing care.
Two hundred million people, organised into living rooms: that sentence describes the architecture of the coming decade with more precision than any chart of monthly reach. Answering costs labour, patience, and accountability, which explains why the market has started paying for it.
Six practical moves follow from that evidence, and every one of them costs far less than a conventional media buy.
- Collect the questions. Write down the ten questions your people actually ask you, phrased in their words rather than yours, because those questions form your genuine editorial plan. A machine can mine them from your correspondence, your comments, and your call notes in a single afternoon.
- Answer one person by name. A direct reply carries more weight than a public post, and it leaves behind a record you can publish afterward. Ten questions answered by name outrank a hundred broadcasts delivered to nobody in particular.
- Build the room of twelve. Twelve people who answer you outperform ten thousand who scroll, so start the room before you chase the audience. The room survives every algorithm change, while the audience vanishes with one.
- Own the asset. Move your answers into a book, an archive, a mailing list, a course, a smart book that carries your name. Platforms rent you a crowd for as long as they choose, while a book keeps your work permanently, along with the relationships that the book creates.
- Run AI as cognitive executive. Let the machine retrieve, verify, remember, structure, and draft, then keep authorship inside your own hands. Sign nothing you cannot defend in a room full of informed opponents.
- Change the dashboard. Count replies, questions, returning readers, and referrals passed from one person to another. Retire the impression from every performance report your organisation circulates.
Then sign it
I write this as a prototype rather than a prophet. I run my own work through the practice I recommend here, and I carry the identical risk I ask a reader to carry. I author the future instead of inheriting it, and I put my own name on the result.
Go back to six in the morning, where that woman will ask her question tomorrow, and the day after, and for years afterward, and somebody will answer her. The entire contest of this decade fits inside that one sentence.
Build the thing that answers her by name. Then sign it.