Evaluating LLM Spatial Grounding: A 100-City Audit of 7,000+ Restaurant Recommendations vs. Google Places for Ground Truth [R]
We evaluated the spatial grounding capabilities of ChatGPT, Gemini, and Perplexity (API) by querying 100 US cities and 5 cuisine types. Using the Google Places API as ground truth, we measured hallucination rates, "permanently closed" retriev…