FINAL TALLY MUST READ? That's my opinion AI versus AI, here is a tainted review. Multi-AI research pipeline experiment — here's what actually worked Been expanding my Shipaton app (Naughty Knitters, a BC yarn shop/fiber farm map) into Washington State, and ran the same research brief through Grok, Gemini, and Perplexity to see who'd do the best job finding real yarn shops, farms, and guilds with actual published contact info. Findings, in case it saves someone else a few rounds of guessing: - Grok (Expert mode) was the strongest single pass by far — broad, honest about its own gaps, math checked out. - Gemini (3.1 Pro) covered less overall but independently found several businesses Grok missed, right in the towns Grok flagged as thin. - Perplexity was weak by default (leaned on one directory site instead of checking businesses individually) — but giving it a tip to search DuckDuckGo-style instead of generic "yarn shop + town" queries turned its 4th attempt into the best pass of all four. Order that worked: Grok first for breadth → Gemini to fill gaps → Perplexity with a search-strategy nudge to round it out. Went from 29 usable results to 101 across three passes. (I did not use claude.ai for actual testing, I should have, but, I used sonnet to run these AI tests, they are all the lowest pro levels I pay for.)