The economics of scraping AI answers
Scraped AI answers cost real money: residential bandwidth runs $0.8–2/GB retail, so an unrendered 3MB session costs roughly $9 per thousand in bandwidth alone. Aggressive resource blocking, stream-terminator detection instead of fixed waits, and deduplicating identical public queries are what bring cost per query down toward workable margins.
Bandwidth is the whole game
Residential bandwidth is $0.8–2/GB at retail. An unblocked browser session pulling 3MB costs around $9 per thousand sessions in bandwidth alone, before compute, before proxies, before failures and retries.
Three levers that actually move the number
- Block everything you don't parse: images, fonts, media, analytics. Most of a page's weight is never read.
- Detect the stream terminator instead of waiting a fixed thirty seconds. Browser time is the second cost line after bandwidth.
- Dedupe identical public queries across customers. Head terms collapse toward zero marginal cost as the base grows.
The pooling compound
n=10 is expensive for one client and nearly free for the fortieth. As the customer base grows, samples get cheaper and confidence intervals get tighter on the same input. Cost falls while quality rises, which is a genuine economy of scale rather than a slide.
Be suspicious of very low prices
See this for your brand
Nostimates shows you how your brand shows up across twelve AI engines and search — in a dashboard, or as data in your own tools. Tell us what you want to measure.