AI Reads 400 Headlines: Measles News Got Scarier, Blame Stayed Rare

Bangladesh had its worst measles outbreak in two decades in 2026: more than 97,000 suspected cases and 600 deaths across 61 of the country's 64 districts, arriving after a change of government and a vaccine shortage. A researcher gathered 403 headlines from seven national news outlets and used an AI model — a program trained to read and sort text — to tag each one as positive or negative and to identify who, if anyone, was being blamed. Human reviewers double-checked a sample and agreed with the AI about 89% of the time.

The pattern that emerged was striking. Early coverage was mixed, but as the outbreak worsened, headlines turned darker: negative stories rose from 56% to 88%, and "threat" framing — language that makes a disease sound terrifying and out of control — jumped from 44% to 84%. Importantly, the negativity didn't spike when cases spiked. It followed deaths, which climb more slowly. In other words, journalists reacted to the body count, not the case count.

Here's the surprise. Despite a charged political backdrop, blame was rare — about 9% of headlines — and when it appeared, 86% of it targeted "the system" (supply chains, health agencies, bureaucracy) rather than naming specific politicians. Coverage amplified danger far more than it assigned fault.

Why should you care? Scary coverage sells, and it shapes behavior: it can push people to vaccinate, or it can fuel fatalism, panic, and mistrust of health officials. This study also shows off a cheap new tool: AI can now scan thousands of headlines in hours, giving health agencies a real-time read on public mood during an emergency. That's useful anywhere — including wherever the next outbreak lands.