The Ghost That Wrote Itself: How a 'Thought Exercise' Wiped $700 Billion

A Scenario, Not a Prediction

On February 22, 2026, a 7,000-word essay appeared on Substack. It was written as a dispatch from June 2028 — a fictional retrospective on an economic crisis that hadn’t happened. The authors, James Van Geelen and Alap Shah of Citrini Research, called it a “thought exercise.” They stamped it with a disclaimer: “a scenario, not a prediction.”

By Monday morning, the scenario had become the prediction anyway.

The Dow dropped over 800 points. DoorDash, American Express, KKR, and Blackstone fell more than 8%. Visa, Mastercard, Uber, Capital One, and Apollo shed at least 3%. Indian IT stocks lost roughly $50 billion in market capitalization. Total damage in the first hour of trading: approximately $700 billion.

Van Geelen later told Bloomberg he was shocked. If he’d known the article would trigger a selloff, he said, he “would not have released it for free.”

That quote tells you everything about what actually happened here — and what didn’t.

What the Essay Said

The thesis is straightforward. AI replaces white-collar workers. Those workers stop spending. Consumer demand collapses. Companies cut more jobs to protect margins, accelerating the cycle. The S&P 500 falls 38%. Unemployment hits 10.2%. A concept called “Ghost GDP” — economic output produced by machines that spend nothing — inflates national accounts while the real economy hollows out.

The essay names names. Visa and Mastercard lose relevance as AI agents handle transactions directly. DoorDash and Uber are built on “friction” that AI eliminates. ServiceNow, Blackstone, KKR — each is cast as a casualty of the intelligence transition. The specificity is deliberate. The essay doesn’t argue about abstract categories. It puts crosshairs on tickers.

This is important.

The Disclaimer That Didn’t

“A scenario, not a prediction” is doing a lot of work in this story. It functions as a legal and reputational shield — the authors can point to it when questioned. But it’s worth examining what the essay actually does versus what the disclaimer says it does.

The essay is written in retrospective past tense. It describes events as though they’ve already occurred. It uses specific numbers — 10.2% unemployment, 38% drawdown, labor’s share of GDP dropping from 56% to 46%. It names real companies and describes their specific failures. It models a causal chain with enough internal coherence to feel like analysis.

None of this is how people write thought exercises. This is how people write forecasts. The form contradicts the disclaimer. And when form and disclaimer conflict, the form wins — because the form is what people actually read.

Noah Smith, writing on Noahpinion, called it “just a scary bedtime story.” He noted that the essay lacked “an explicit macroeconomic model” and couldn’t explain its own assumptions about how the economy actually works. He pointed out that companies like DoorDash have network effects that “Claude Code can’t simply conjure up out of nothingness.”

He’s right about the economics. But he’s describing a different problem than the one that matters here.

The Influence Architecture

What makes this case worth examining isn’t whether the Citrini thesis is correct. It’s how its construction produced real-world effects independent of its accuracy.

Several techniques compound:

1. Specificity as Credibility

Abstract fear is easy to dismiss. “AI might disrupt the economy” is a think-piece. “DoorDash’s moat is made of friction, and friction is going to zero” is a trade idea. By naming companies, citing specific percentage declines, and modeling precise unemployment figures, the essay converted vague anxiety into actionable alarm.

Investors don’t sell on vibes. They sell on scenarios that map to their portfolios. The essay provided the map.

2. The Fictional-Retrospective Frame

Writing from the future is a narrative power move. It inverts the burden of proof. Instead of arguing “this could happen” (which invites counterargument), it describes “what happened” (which invites only denial). Psychologically, it’s harder to argue with a story told in past tense — your brain processes it closer to memory than hypothesis.

This is a well-documented technique in persuasion research. Pre-mortems work because imagining a future failure makes it feel more plausible. The Citrini essay is a pre-mortem for the entire U.S. economy.

3. Riding Existing Fear

The essay didn’t create anxiety. It crystallized it. Software stocks had been declining throughout February in what traders called the “SaaSpocalypse.” Anthropic had just released Claude Cowork, sparking fresh fears about AI replacing specialized services. The market was already nervous.

The essay gave that nervousness a name — Ghost GDP — and a narrative structure. It took diffuse unease and organized it into a compelling story with a beginning, middle, and catastrophic end. The timing wasn’t accidental. Narratives gain traction when they explain feelings people already have.

4. The Sticky Concept

“Ghost GDP” is a masterpiece of narrative packaging. Two words that make you see an economy producing output that no human benefits from. It’s vivid, it’s memorable, and it’s terrifying — regardless of whether it describes anything real.

Concepts that stick tend to share a structure: they name something people sense but can’t articulate. “Ghost GDP” does this for the fear that AI growth metrics mask human displacement. Whether the economics hold up is almost beside the point. The concept now exists. It will show up in earnings calls, policy debates, and future market panics. Once a sticky concept enters circulation, it doesn’t need to be true to be influential.

5. The Self-Fulfilling Architecture

Here’s the part that deserves the most attention.

The essay describes a crisis caused by belief — specifically, by companies believing AI can replace workers, acting on that belief, and creating the conditions the belief predicted. The essay calls this a “negative feedback loop with no natural brake.”

Then the essay itself became a negative feedback loop with no natural brake.

Investors read a scenario about AI destroying value. They sold stocks of the named companies. The stocks dropped. The drops were reported as evidence that markets are pricing in AI disruption. The reporting drove more fear. More selling.

The essay about a self-fulfilling crisis became a self-fulfilling event. The Ghost GDP memo is its own ghost.

Van Geelen didn’t intend this. His surprise appears genuine. But intent is irrelevant to influence architecture. A well-constructed narrative produces effects through its structure, not through the author’s wishes.

The Influence Tactics Breakdown

Scoring this through a manipulation detection lens:

Fear Appeal: Very High. The entire piece is engineered to activate loss aversion. Specific dollar amounts, named companies, and a fictional timeline create urgency. The retrospective frame makes the fear feel like memory rather than speculation.
Deceptive Framing: High. The “thought exercise” disclaimer contradicts the essay’s form, tone, and level of specificity. The framing says hypothesis; the content says forecast. Most readers processed the content, not the caveat.
Missing Information: High. The essay omits counterarguments almost entirely. No discussion of historical precedent for technology-driven job transitions. No mention of new job categories AI might create. No acknowledgment that previous “this time it’s different” theses have a mixed track record. The omissions aren’t lies — they’re architectural. The story works because contrary evidence is absent.
Source Credibility Exploitation: Moderate. Van Geelen is an ex-Citadel analyst. Shah is an AI entrepreneur. The pedigree lends authority to a piece that, by its own admission, isn’t making a claim. Citadel Securities later issued a formal rebuttal — an unusual move that itself signals how seriously the market took the piece.

What Actually Happened Here

A well-written essay, published at the right moment, organized existing fears into a compelling narrative, named specific targets, and triggered a $700 billion selloff. The author didn’t intend it. The disclaimer said it wasn’t real. The market made it real anyway.

This is influence without intent. Manipulation without a manipulator. The architecture of the narrative did the work — the specificity, the framing, the timing, the sticky concept.

It’s also a preview. As AI-generated analysis becomes more common, the gap between “thought exercise” and “market-moving event” will continue to shrink. Anyone can write a detailed, plausible-sounding scenario about any sector’s collapse. The tools to produce this kind of content are getting cheaper and faster. The markets that react to it are getting more algorithmic and sentiment-driven.

Ghost GDP might not be real. But ghost narratives — stories that move markets without being true, without even claiming to be true — are already here.

The next one won’t come with a disclaimer.

This article is part of Decipon’s Manipulation Breakdown series, where we analyze real-world examples of influence tactics using the Decipon Influence Tactics Score methodology. Decipon doesn’t tell you what’s true — it shows you how content is trying to influence you.

Sources: