All case studies
Matchmaking · Discovery
A matchmaker found out which AI tool is actually best at finding people online
Five tools, 69 real candidates, one question: which AI finds the right LinkedIn, Instagram, and Facebook profiles? The answer was not one winner. It was a ranked pipeline that uses each tool where it performs best.
The problem
A matchmaker needed a repeatable way to research candidates from public rosters. Given a university athletic roster or a professional directory, she needed each person's LinkedIn, Facebook, and Instagram profiles in a clean spreadsheet. The question was: which AI tool does this best, and can any of them actually be trusted?
What we built
- Five tools ran head-to-head on the same 69 real candidates: Claude Sonnet 4.6, Claude Opus 4.6, Perplexity Pro Deep Research, Manus AI, and Gemini. Results were scored by channel so the matchmaker could see exactly where each tool wins and where it fails.
- A single-file benchmark report showed accuracy scores side by side. No spreadsheet math needed to read it.
- A best combined pipeline came out of the test: Sonnet to scrape the source roster, Perplexity to find LinkedIn, Manus to find Instagram and Facebook, a second Perplexity pass for anyone with no links, then Sonnet to merge into the final spreadsheet. No single tool was best at everything. The pipeline uses each one where it performs.
- Expected coverage gave the matchmaker a clear picture before committing: LinkedIn 75-80%, Facebook 25-30%, Instagram 65-70%, at least one link on 90% or more of candidates.
- A guide to where each tool makes things up. Gemini fabricates LinkedIn URLs that look real but return 404 errors. Manus overclaims confidence on private profiles. Knowing the failure modes lets you catch errors before they waste time.