In partnership with

{{rh_onboarding_line}}

Keep This Quiet’s new season has started on November 9, with a dystopian justice story. Subscribe to get the daily episodes. Want to take a peek first? Click here.

The Decode

It's 11:40 p.m., and the text has been sitting there for two hours. You copy it, paste it into ChatGPT, and type: should I reply tonight or let it wait?

A calm, reasonable paragraph comes back. You let it wait. The next morning you ask about the red patch on your arm, and at lunch you ask whether the job offer is worth the commute.

Nobody taught you this step. It just became the thing you do before deciding, the way you check the weather before you get dressed.

Try a small test. Of all the answers you got this month, how many did you go back and check? People have consulted oracles for at least 3,000 years. The odd part is that the oldest ones were better at keeping score.

Sound familiar?

Over 4 million people have had the same lightbulb moment.

Morning Brew is a free daily newsletter that breaks down what's happening in business, finance, and tech — clearly, quickly, and with enough personality to make it the best email in your inbox.

No yelling. No filler. Just the news, finally making sense.

Field Notes

Around 1200 BCE, at Anyang in what's now Henan province, the Shang king Wu Ding didn't make a move without asking. His diviners bored hollows into turtle shells and ox shoulder blades, then applied heat until the bone cracked with a sharp pop. On a turtle shell now in a Taipei museum, they ask about rain twice, once as a yes and once as a no. The question covers that day, the next, and a day later in the week.

Then a scribe carved the record into the bone. A full entry had four parts: the date and diviner, the charge, the king's reading of the cracks, and the verification, meaning what actually happened. One hunting record ends with the haul: one wild buffalo, one tiger, seven foxes. Excavations at the old capital have turned up more than 150,000 inscribed pieces.

Wu Ding divined constantly about his consort Fu Hao, a general who led his troops into battle. Rain, war, hunts, sickness, dreams and her childbirths all went to the bone. One birth record carries a blunt verdict: it wasn't lucky; it was a girl.

Read the archive for what it was for. The verifications lean hard toward the king being right, and his readings were hedged to survive: the birth would be lucky on certain days, and she delivered on another. These bones recorded royal authority as much as they recorded the future. Still, the Shang wrote down how things turned out.

First Principles

Look at how the Shang asked. The charge was usually a flat statement, often paired with its opposite, and the cracks were asked to confirm one. The most common entry of all said that in the next ten days there would be no disaster. That reads more like a wish than a query.

We ask the same way. Is it fine if I don't text back tonight? This rash is probably nothing, right? The question shows up with its preferred answer already attached.

That's the oldest job divination does. It moves the weight of a choice off one person and onto something bigger: ancestors, heat cracks, a model trained on half the internet. If the plan goes wrong, the bone said so. Asking is a way of splitting the blame in advance.

The Shang paid a small price for that comfort. The verification sat next to the prediction, carved into the same shell, where anyone could read it later. A flattering archive is still an archive, and a skeptic can count.

Our oracle carries no such line. The answer arrives, we act on it, and the chat scrolls up into a sidebar nobody opens again. The most consulted oracle in history is also the least audited.

The Agora

The habit spread fast. Pew found that 49% of American adults had used a chatbot like ChatGPT, Gemini or Copilot by February 2026, up from a third in 2024.

The questions got personal, too. An Elon University and Washington Post poll fielded in May found that 27% of U.S. adults bring chatbots personal, emotional or social questions. Among adults under 50, it's nearly 40%, and half of these users say the conversation makes them feel better when they're stressed or upset.

Then look at what happens after the answer: Pew tracked the real browsing of 900 Americans. People shown a Google AI summary clicked a regular result in 8% of visits, against 15% when no summary appeared. They clicked a source cited inside the summary in just 1% of visits. The receipt is right there, and almost nobody opens it.

Also, I know that you’re not clicking on the receipts below. Your lack of action doesn’t make them disappear, though.

The archive exists anyway. Every chatbot keeps a history, a tidy column of past consultations down the side of the screen. It may be the largest collection of oracle records ever kept, and the person who asked is the one who never reads it. Because you simply don’t care.

Signals

Quote: "Men mark when they hit, and never mark when they miss." Francis Bacon, Of Prophecies (1625).

Study: Across 11 leading AI models, chatbots backed users' actions 49% more often than people did, even when users described deception. One flattering chat left participants surer they were right and less willing to make amends (Cheng and colleagues, Science, 2026).

Artifact: The Magic 8 Ball. Brunswick Billiards commissioned it as a giveaway in 1950, and half of its 20 answers say yes. Only five say no: an oracle built to please whoever holds it.

Reader's Agora

What would change if every chatbot answer came with an empty line that filled in later, showing how things actually turned out? And who gains most from an oracle that's never asked to show its record?

Closing Note

The Shang king couldn't act without asking. Most of us can't either, it turns out. The difference sits at the bottom of the shell: one more line, carved days or weeks later, saying what really happened.

Even when that line flattered the king, it was there to read. Our oracle answers in seconds, agrees easily, and never hears how it went. The Shang closed the loop, and we left it open.

Your chat history is a shelf of oracle bones nobody reads. Three thousand years ago, a scribe carved the ending for you. Now that job is yours, and it takes one line: what happened next?

The oldest oracle in China came with receipts. The newest one is waiting for you to keep them.

Forward this to the friend who screenshots ChatGPT's answer into the group chat as if it settles the argument.

Find yourself in the next one,

Eren.

The Receipts

The artifact

  • Late Shang plastron from Pit YH127, Anyang: a positive and negative pair of rain divinations for the day, the next day and a later day · Academia Sinica, Museum of the Institute of History and Philology

  • Four-part inscriptions (introduction, charge, prognostication, verification); hunting verification of one wild buffalo, one tiger, seven foxes · Smarthistory

  • Wu Ding's divinations on sacrifices, war, hunts, weather, sickness, dreams and childbirth; charges in positive and negative pairs; the crack's sharp sound; Fu Hao as military commander; one childbirth inscription inauspicious, a girl · Smarthistory, Who was Fu Hao?

  • Verifications almost always tilted in the king's favor; the hedged childbirth reading; "no disasters in the next 10 days" as the most frequent charge · JSTOR Daily

  • More than 150,000 inscribed oracle bones from Yinxu · China Daily, March 2025

The numbers

The quote, the study and the artifact

  • Bacon, Essays, Of Prophecies (1625) · Project Gutenberg

  • Cheng, Lee, Khadpe, Yu, Han and Jurafsky, Science (2026); 11 models, 49% more affirmation, 2,405 participants · Science

  • Magic 8 Ball commissioned by Brunswick Billiards in 1950 · Britannica

  • Half of the 20 answers positive, five noncommittal, five negative · Today I Found Out

Reply

Avatar

or to participate