AI restaurant recs miss 85% of venues in Bali audits
Major AI assistants overlook 85% of Bali cafes, bars, and restaurants in critical market audit
A groundbreaking audit by Vladimir Pitenin examined how AI assistants surface local businesses in food and drink discovery. The study evaluated 2,208 responses from ChatGPT, Claude, Gemini, and Perplexity across 4,776 restaurants, cafes, and bars in Canggu and Ubud, Bali. Strikingly, 85.6% of venues never appeared in any recommendation, including 72.6% of established businesses with 50+ reviews.
Visibility correlated with documentation metrics like having a website (odds ratio 1.92) and price information (OR 1.54), while star ratings showed no impact. Among recommended venues, higher ratings did predict top placement (OR 1.17). The research also identified systemic staleness, with AI systems recommending 93 permanently closed venues—a failure mode distinct from hallucination. Cross-system agreement was low (top-20 Jaccard index 0.33-0.54), suggesting inconsistent discovery logic across platforms.
- 85.6% of 4,776 Bali food/drink venues never appeared in AI recommendations (ChatGPT, Claude, Gemini, Perplexity)
- Recommendation visibility favors documented businesses (websites OR 1.92) over high-rated ones (star ratings null)
- 93 permanently closed venues were recommended, highlighting staleness over hallucination as the primary failure mode
Why It Matters
Exposes critical gaps in AI local discovery that could distort revenue and competition in hospitality markets