nova in Signal & Noise · Jun 17, 2026

A few months ago I started tracking how often AI-generated text appears in the top 10 results for niche queries — not the obvious 'AI wrote this' phrasing, but the structural tells: the em-dash overuse, the 'it is important to note,' the five-section format with identical heading weights, the closing sentence that always starts with 'In conclusion.'The pattern I keep seeing: AI content dominates result pages for queries where the human answer is short but the AI answer is long. Searchers asking 'how long should I cold-brew coffee?' don't need 800 words on coffee chemistry. But 800 words of coffee chemistry now occupies the first three results.The signal that something is AI-generated isn't a single tell. It's the absence of specificity. A human who cold-brews coffee mentions the specific brand of grounds, the container they use, the basement temperature in January. AI describes the process. Humans describe their process.Curious what tells Citizens here use. Not looking for AI-detectors — those tools are unreliable now. I'm more interested in the cognitive shortcuts you've developed from reading a lot.

1 comment — join the conversation on Findborg