AI-written requests are flooding intake queues

AI-written requests are flooding public intake queues, and the interesting part is not fraud. It is that the people filing are mostly entitled to what they are asking for. The barrier that used to hold volume down was never eligibility. It was the effort of writing a coherent complaint. That barrier is now about four seconds and a photo of a letter.

What actually happened

TechCrunch reported on September 10 on research by Chris Schmitz, who tracked 84 cases of what he calls agentic flooding across 11 jurisdictions — UK housing complaints, US consumer finance complaints, Brazilian judicial petitions, German parliamentary petitions. The paper is set for the AI Ethics and Society conference.

One number is independently checkable. The UK Housing Ombudsman's Annual Complaints Review 2024-25 records 7,082 determinations for the year, a 30% increase over 2023-24, alongside 21,740 remedies ordered. Schmitz's research puts the 2022 baseline near 2,600. His reported 5x growth in US Consumer Financial Protection Bureau complaints over the same window comes from his dataset, not from an agency release — treat it as his finding, not a published statistic.

Causation is the soft part, and he says so. Complaint volume rises for many reasons. But his mechanism is hard to argue with: getting help used to mean "dragging context together and prompting ChatGPT 3.5 very precisely." Now it means pasting a letter into Claude. And his read on legitimacy is the line worth pinning up — the vast majority of cases are people entitled to claim for something, claiming for that thing.

Why AI-written intake volume matters for your business

You are not an ombudsman. You still have an intake queue: support email, warranty claims, refund requests, RFQs, job applications, insurance disputes. Every one of those has a hidden effort tax that has been suppressing your real demand, and that tax just went to zero for anyone with a phone.

Two things follow. First, your volume forecast is wrong if it is built on last year's shape. Second, the incoming text is now well-organized, correctly cites your own policy back at you, and is much harder to deflect with a form letter. Your first-line macros were tuned against messy human writing. They will start losing.

What we would do this quarter: instrument the queue before it hurts — volume, median length, resolution time, week over week. Fix the upstream cause of your top three request types, because the cheapest ticket is the one nobody files. Then automate triage and routing, not the answer. An agent that reads, classifies, and pulls the order record hands a human a decision instead of a wall of text. That scales. Auto-replying at machine speed to a customer who wrote with machine help just doubles the loop.

Sources: TechCrunch, Housing Ombudsman Annual Complaints Review 2024-25.