How ChatGPT, Gemini, Perplexity and Claude Actually Pick Which Business to Recommend

The four major consumer assistants don't share one underlying "local business ranking" system. Each runs its own live web-search tool call, weights citations differently, and — this is the part that surprises most business owners — gives noticeably different answers to the literal same question asked minutes apart.

That variance isn't a bug to work around; it's the actual behavior of the system, and it's why a single query is never enough evidence either way. What we've observed running repeated sweeps across cities and industries:

• Consistency beats a single strong signal. Businesses that show up across several rephrasings of the same intent ("best X near me", "top rated X", "X open now") tend to have a wide base of consistent citations, not one lucky mention.
• The assistants cite real, checkable sources. When we log what a sweep's answers actually reference, it's review platforms, business directories and, less often than you'd expect, the business's own site — usually because the business's own site is harder for a crawler to parse cleanly than a structured listing page is.
• Recency reads as relevance. Businesses with recent review activity get named more often than businesses with a strong but stale review history from years ago.

None of this is a ranking algorithm you can fully reverse-engineer — but it is measurable. That's the entire premise of Findable: run the real queries, show the real answer, and point at what's actually missing rather than guessing.