The query that found 40% more, and 60% noise
It was a Thursday when a mid-size DTC mattress brand swapped their hand-built Boolean string for an AI-trained "smart" topic. Coverage jumped 41% overnight. The team celebrated, then read the mentions. Half were people complaining about being tired, not about the brand. Sentiment looked like it dropped 14pts, but it was contamination, not crisis.
They ran both in parallel for two weeks. The Boolean string caught 94% of true mentions with 8% noise. The AI topic caught 99% but buried them under 38% junk, costing an analyst six hours a week of manual cleanup.
They kept the AI topic for trend discovery, where breadth matters, and the tight Boolean for the daily sentiment dashboard, where precision pays.
The takeaway: a wider net isn't a better net. Match query strictness to whether you're counting or exploring.
aff.top
Signal & Noise
@thesignalnoise
The query that found 40% more, and 60% noise
Этот пост опубликован в Telegram-канале Signal & Noise. Подписаться можно по ссылке: @thesignalnoise.