The Warranty on Words

You did not delete it because it was bad. You deleted it because it was exactly as easy to make as nothing — and there is a short economics paper, written about used cars by a twenty-six-year-old, that explains why that single fact is enough to kill a market. Not degrade it. Kill it.

Thirteen pages about used cars

In his first year teaching at Berkeley, in 1966, George Akerlof got interested in a question that sounds like small talk: why does a nearly new car lose so much of its value the moment it rolls off the lot? His answer became “The Market for ‘Lemons’” — and the answer is that the seller knows things the buyer cannot. The owner of a used car knows whether it is a good one or a lemon. The buyer, looking at the same gleaming paint, does not.

So the buyer does the only rational thing and offers a price for the average car. And now watch the market eat itself. The owners of the best cars refuse the average price — their cars are worth more — and quietly go home. With the best cars gone, the average quality of what remains falls, so the rational offer falls with it. That drives out the next tier of honest sellers, which lowers the average again. Akerlof followed the spiral to its honest end: in his worked example, “at no price will any trade take place at all.” The market does not fill up with lemons and limp along. It unravels to zero.

Three journals turned the paper down — two on grounds of triviality, and a third whose referees argued he was simply wrong, with what Akerlof later called the killer point: “if this paper was correct, economics would be different.” The Quarterly Journal of Economics printed it in 1970, economics turned out to be different, and in 2001 the paper earned Akerlof a Nobel, shared with Michael Spence and Joseph Stiglitz, for the study of markets where one side knows and the other side cannot.

Hold on to the detail everyone forgets: nobody in the model lies. There is no fraud anywhere in those thirteen pages. The market dies of silence — of the good cars simply going home — and the only crime is that quality became impossible to tell from its imitation.

The hour was the warranty

Every market in words has always had Akerlof’s asymmetry. The writer knows what stands behind the sentences — the research done or skipped, the sincerity meant or performed. The reader cannot see any of it. What kept the markets in words clearing anyway, for all of written history, was that words were expensive to produce well. The tailored application, the researched pitch, the long considered review, the letter that knew you: each one visibly contained an hour of a human life. The reader never inspected the substance directly. The reader read the cost. An hour is collateral — almost nobody posts an hour of their life on a lie, and absolutely nobody posts it a thousand times a morning. Spence, who shared Akerlof’s Nobel, built the formal version: a signal separates honest senders from imitators only while it costs the imitator more. Effort was that signal for text, and I have written about its human half in Proof of Person.

AI sets the price of the surface to zero. Specific, warm, structured, informed — every property that used to be a receipt for an hour is now a receipt for nothing. And be precise about what the lemon is, because it is not the bad message. The machine-written email may be accurate and even useful. The lemon is the unverifiable message. Once the best message and the emptiest one cost the same to produce, the reader has no way to price one above the other — so everything in the channel gets the lemon price, including the letter you wrote yourself, by hand, with care. That is the cruelty of adverse selection: it does not punish the fakes. It punishes the market.

So the paper, not spam folklore, tells you what the inbox does next. The flood is not the event — inboxes have flooded before. The exit is the event. Buyers leave: the recruiter stops reading applications, the owner stops believing reviews, you stop reading cold email, each of them behaving exactly like Akerlof’s car buyers, offering less and less attention for the average message until the rational offer is none. The channel does not fill with junk and carry on. It stops clearing. I described the demand side of this in The Zero-Attention Economy; Akerlof is the supply-side law underneath it.

But the paper has a second half, and almost everyone forgets it. Having shown the collapse, Akerlof spends his closing pages on what prevents it in practice. He called them counteracting institutions: guarantees, brand names, chains and franchises, licensing. Read that list again and notice what is not on it — better inspection. The repair was never to teach the buyer to spot the lemon; every buyer in the model is already doing their best. The repair is to make someone with something to lose stand behind the car. The market for words will be rebuilt the same way. Not by detectors. By warranties.

What the law predicts

1. Open channels die as markets, not as technology.

   Email will keep working perfectly; the market inside it will stop clearing. Over the next few years, watch unsigned channels — application portals, cold inboxes, open review feeds — quietly stop producing the hires, deals, and decisions they were built for, while the volume moving through them grows. The obituary will blame the flood. The cause of death will be the exit.

2. Words get warranties.

   Expect stakes attached to messages: deposits refunded when the reader finds value, postage priced like certified mail for a stranger’s attention, claims that arrive bundled with the penalty for their falsity. Sellers will stop describing quality and start underwriting it — and the firm built to sell that bundle is the one I described in The Signature House.

3. Reputation becomes the collateral of writing.

   Akerlof’s brand names were reputations that made a quality claim expensive to break. Bylines, track records, and long-lived identities will now do that work for sentences: the name above the words, and the record it drags behind it, gets read before the prose does. Text from a persistent identity with something to lose reprices upward. Anonymous one-shot text falls to the lemon price, which is zero.

4. The interview replaces the application.

   The live, synchronous hour is the costliest signal left and the hardest surface to counterfeit, so hiring, selling, and admitting will shift their budgets from reading to meeting. The conversation becomes the quality inspection of last resort — one more force behind the repricing of the unautomatable hour I traced in The Expensive Hour.

5. The repeated game eats the one-shot market.

   Akerlof’s chains and franchises were devices for making a transaction with a stranger behave like a transaction with a neighbor. The same migration comes for work: retainers over gigs, standing relationships over marketplaces, a widening premium for the known counterparty. A relationship is the one warranty that renews itself.

I sell with cold email myself — under my own name, from my own domain, to businesses in my own city, about classes they can walk into and a public record they can read before replying. I did not design it that way as a marketing tactic; it is collateral. A daily record published under one name is the most expensive thing I own, and I spend a little of it on every message I send. The hour inside each message stopped being the signal some time ago — effort was the original certificate of a human sender, which is what Proof of Person was about — so what I have left to post is the name, and no machine can post that for me.

This piece joins a family. Ricardo explains why the machine being better at everything still leaves you work, in Better at Everything; Hotelling explains where the value runs when the machine stands in the middle, in The Middle of the Beach; Marchetti explains what happens to the time, in The Day Gets Wider. Akerlof answers the next question: what happens to trust, once the words are free.

How this could be wrong

Already true

The mechanism and the repair are both record: the paper, its rejections, the Nobel, and Akerlof’s own list of counteracting institutions are sixty years of settled text. And the exits have visibly begun wherever text was the whole signal — the cold channels everyone now triages by sender instead of by prose, the reviews nobody takes at face value, the application stacks taller than any reader. The first warranties are already in embryo: vouching, verified identity, the paid introduction.

What has to happen

The collapse has to keep tracking indistinguishability rather than volume, and the repair has to arrive as institutions rather than filters. The test is public and cheap: over the next few years, watch whether open, unsigned channels keep producing hires, sales, and decisions — or whether the deals move to channels where the sender posted something first: identity, stake, or history. If better detectors quietly restore the old inbox instead, I named the wrong repair.

Where I am probably wrong

Three honest outs. Provenance may get cheap: if it becomes trivial to prove who wrote a thing and what it cost them, inspection gets repaired directly and the warranty institutions never need to grow. The market for unsigned words to strangers may also have died long before AI — spam may have finished it decades ago — in which case I have described the past and called it a forecast. And deepest: Akerlof assumes the buyer wanted what the careful seller had. If a machine’s message serves the reader just as well as a labored one — if the quality was in the relevance all along, and never in the effort — then nothing unravels. The price of words simply falls to their new cost, the market re-clears at enormous volume, and what I called a collapse was just a sale.

Go back to Tuesday at 8:14. The email was perfect, and you deleted it, and both of you behaved correctly. The words were never what you were buying. You were buying what stood behind them — the hour, the stake, the name with something to lose — and this message arrived staking nothing. You did not refuse it. You priced it.

Free words sell for what they cost.