← ALL DATA STUDIES

Who do AI assistants recommend for restaurants in Miami?

Generated 2026-08-12 · window 2026-07-15 → 2026-08-12 (28 days) · n = 157 sampled answers across 5 engines

Key findings

  • AI engines named 597 distinct businesses/products for this niche in 28 days (across 157 sampled answers) — but recommendations concentrate hard on the top of that list (see the share table below).
  • The #1 answer ("Chug's Diner") appears in 17.2% of sampled answers (27 of 157 answers), recommended by 5 of 5 engines measured.
  • 50.1% of everything AI recommends here is named by only ONE engine (299 of the 597 distinct entities recommended in the window) — the engines materially disagree.
  • The most-cited source for this niche is vertexaisearch.cloud.google.com — it backs 27 of the 157 sampled answers (17.2%).

Top 5 most-recommended

#NameShare of answersAnswersAvg. list positionEngine agreement
1Chug's Diner17.2%27/1579.35/5
2Enriqueta's Sandwich Shop10.2%16/1575.85/5
3Versailles10.2%16/1577.55/5
4CVI.CHE 10510.2%16/1579.15/5
5Mandolin Aegean Bistro (mandolinrestaurant.com)9.6%15/1577.55/5

Share = % of all 157 sampled answers in the window that recommend the entity. Engine agreement = engines recommending it at least once / engines measured.

Engine disagreement

Across the 5 engines measured (Claude (Anthropic), Gemini (Google), Grok (xAI), ChatGPT (OpenAI) and Perplexity), 597 distinct entities were recommended at least once. 299 of them (50.1%) were named by only a single engine (n = 597 entities). Ask a different AI, get a different answer.

What AI reads: most-cited sources

116 of 157 sampled answers cited at least one source.

SourceCitationsAnswers citing it% of answers
vertexaisearch.cloud.google.com21727/15717.2%
theinfatuation.com11168/15743.3%
reddit.com9541/15726.1%
miaminewtimes.com4734/15721.7%
timeout.com4639/15724.8%
miami.eater.com3222/15714%
opentable.com3224/15715.3%
hungrypost.com2727/15717.2%
instagram.com1914/1578.9%
tripadvisor.com1918/15711.5%

Chart data

Who do AI assistants recommend for restaurants in Miami? — recommendation share, last 28 days
Chug's Diner 17.2%
Enriqueta's Sandwich Shop 10.2%
Versailles 10.2%
CVI.CHE 105 10.2%
Mandolin Aegean Bistro 9.6%
% of sampled answers · 157 sampled answers

Methodology

Sampled from the Orbator AI Recommendation Index: 157 answers collected 2026-07-15 → 2026-08-12 (28 days) across Claude (Anthropic) (n = 30), Gemini (Google) (n = 40), Grok (xAI) (n = 14), ChatGPT (OpenAI) (n = 28), Perplexity (n = 45) via their official APIs, with web search enabled where the engine supports it.

Category sampledn (answers)Note
Best breakfast in Miami11small sample (<20) — directional only
Best brunch in Miami13small sample (<20) — directional only
Best cuban food in Miami11small sample (<20) — directional only
Best italian restaurant in Miami11small sample (<20) — directional only
Best mediterranean food in Miami11small sample (<20) — directional only
Best pancakes in Miami3small sample (<20) — directional only
Best peruvian food in Miami11small sample (<20) — directional only
Best pizza in Miami11small sample (<20) — directional only
Best ramen in Miami11small sample (<20) — directional only
Best seafood in Miami11small sample (<20) — directional only
Best steakhouse in Miami11small sample (<20) — directional only
Best sushi in Miami11small sample (<20) — directional only
Best tacos in Miami9small sample (<20) — directional only
Best thai food in Miami11small sample (<20) — directional only
Best vegan restaurant in Miami11small sample (<20) — directional only
  • Prompt design. Each category carries buyer-shaped prompts — the questions real buyers type ("best [category] for [audience]", "what should I use for [job]"). Prompts are strictly neutral: they never contain product or business names and never steer toward or away from anyone.
  • Entity resolution. Every business/product named in an answer is extracted and resolved against a canonical registry, so name variants and domains count as one entity. Generic phrases are filtered and never become entities; suspected duplicates are human-reviewed.
  • "Known" entities. The never-recommended stat counts every canonical entity the Index has ever extracted from this niche's answers (all time) and asks how many were not named even once inside the study window.
  • Minimum sample. Categories with fewer than 20 sampled answers in the window are marked as small samples and should be treated as directional, not ranked — we publish stable numbers, not noise.
  • Independence. Orbator sells AI-visibility tooling; measurement is identical for every business and product, customer or not. Rankings cannot be bought.

Full methodology: https://www.orbator.io/ai-index/methodology · Data © Orbator — free to cite with attribution to orbator.io.

Want to know whether AI recommends your business? Run the free check →