Ten million dollars bought Google 100 million emails, 500 million items out of Microsoft Teams, 30 million recorded customer service calls, 1.2 million fuel slips, and the purchase records for 787,452 aircraft parts. Spirit Airlines grounded itself permanently in May; its liquidation is now selling the exhaust. The underbidder, at $7.5 million, was Mercor — a company whose entire business is producing training data for AI labs.

Two of the day's other stories are the same story. An AirTag hidden in a rare book traced a bulk order to an Amazon warehouse in Las Vegas, where a team called VGT3 tears books from their spines and scans them. And a think tank that published more than 100 reports in its first eleven days turns out to be the product of a firm on a $900,000 contract with the Israeli government. Buy it at auction, cut it out of the binding, or write it yourself. Scraping is finished as the primary input. Everything after scraping is procurement.

The Amazon detail that matters isn't the T. rex logo on the warehouse door. It's that the warehouse ran out. Workers said the book supply went to zero earlier this year and they expected to be shut down, and that they were trained to scan the barcode before the pages. That is how you work through the list of every ISBN ever issued rather than a list of books worth reading. It isn't a library. It's an inventory count against a finite universe, and Amazon can see the bottom of it.

Buy it at auction, cut it out of the binding, or write it yourself.

Which is what makes the fake think tank the important one. Google's purchase and Amazon's scanning are both bounded by the same hard fact: the data has to already exist. Piro, Inc. is not bounded by it. The firm sells what its own site calls "AI Story Optimization" — content "engineered for how LLMs evaluate credibility." A NewsGuard analyst called the site perfect mimicry of a credible American think tank, down to the red-white-blue color scheme. Responsible Statecraft ran twelve of the reports through GPTZero; eleven came back machine-written with high confidence. A hundred reports in eleven days is not research output. It is manufacturing to a published spec.

When someone asks ChatGPT, Gemini, or Perplexity about your category, an answer comes back in one confident paragraph. Most brands have no idea how that paragraph gets built. So we spent months reverse-engineering it.
Responsible Statecraft

The obvious objection is that none of this is new. Public relations has always written for whatever the era's credibility oracle happened to be — the wire desk, then the search index, now the model. That's fair, and as advertising Piro's work is unremarkable. What's new is which side of the ledger it sits on. Google and Amazon are buyers, competing for scarce supply at auction prices and running out. Piro is a supplier, feeding the same pipe at a fraction of the cost with no scarcity constraint on it whatsoever. A corpus that pays $10 million for a dead airline's fuel slips and accepts a hundred contracted reports for nothing does not have a moderation problem. It has a pricing problem.

Every frontier lab is running a procurement desk now, and procurement desks get quoted prices. Spirit's exhaust went for $10 million because two bidders wanted it. The Hanover Institute's hundred reports cost the corpus nothing and took eleven days. The cheapest supplier is always going to be the one willing to write it for you.