AI rarely cites an award. Two in five are Michelin

Short answer: when AI recommends a business, it cites an award in 1.2% of cases, and two times in five it is a Michelin star. In 16,022 passages where ChatGPT or Gemini recommend a business in four countries, 191 give a prize, a guide or a ranking as a reason. The Michelin Guide alone accounts for 76. For restaurants, a distinction comes up in one passage in six; for plumbers, electricians, locksmiths, garages and driving schools, in 6 passages out of 4,809. AI does not reward awards: it repeats the ones from a judge it already knows.

An award travels into an AI answer when it comes from a judge the model already knows.

What was measured

The corpus of the 1 and 2 October editions, cut the same way: our four studies (United States, United Kingdom, France, Brazil), buyer questions put to ChatGPT (gpt-4o-mini) and Gemini through their programming interfaces, without web search - 3,600 answers, and within them 16,022 passages each devoted to a named business. In every passage we looked for a distinction: a prize or award (“award-winning”, “primée”, “premiado”), a guide (Michelin, Gault&Millau), a title (Meilleur Ouvrier de France), a ranking (Chambers, Legal 500), a readers' vote (“Best of”). The detector was proven on 15 positive and 10 negative cases, then corrected three times on rereading: the French “se distingue par” (stands out for) is not a distinction, “the best of both worlds” is not either, and a piece of advice (“look for attorneys who have received awards”) says nothing about a business.

A prize comes up for restaurants, almost never for tradespeople

Restaurants carry most of it: 74 of their 452 passages cite a distinction, one in six, from 22.5% in the United Kingdom to 11.1% in Brazil. Next come French bakeries (8 passages out of 102, Meilleurs Ouvriers de France among them) and hotels. In the trades that come to your home, a distinction barely exists in the answers. Gemini cites them 4.4 times as often as ChatGPT: in 2.3% of its passages against 0.5%.

Two in five come from a single guide

Michelin accounts for 76 of the 191 mentions: two in five. The other awards (71) mix named prizes - a chef described as a “James Beard Award winner”, a “Forbes Five-Star” hotel - with a bare “award-winning”, for agencies or hair salons for instance. “Best of” lists and readers' votes, which so many directories sell, appear in 8 passages out of 16,022.

What it means if you are the business

A distinction enters an AI answer when it comes from a judge the model already knows: a guide, a national title, a ranking the trade recognises. In trades that have such a judge - restaurants, bakeries, corporate law - getting that judgement is a visible lever, and the AI names it. In trades that do not, a “Best of” badge bought from a directory is unlikely to be what the assistant repeats; other proof carries, such as a licence or a clear price (our editions of 19 and 29 September). And if you hold a real distinction, write it out in full on your own pages: who awarded it, for what, in which year. “Award-winning” on its own gives nothing to check, to a customer or to a model.

What this measurement does not prove

• It counts wording, not intention. Reread on 60 passages drawn at random, the detector was right 57 times; the 3 errors are two pieces of general advice and a metaphor (“the Michelin guide of the profession”).
• The classification by kind keeps the first word found: an “award-winning” passage that goes on to name the James Beard counts among the “other awards”.
• Answers from memory, without web search, one model per assistant, at one date per country (United States 13 September, United Kingdom 31 July, France 30 July, Brazil 6 August 2026). We did not check that the distinctions cited are accurate or still current: a lost star can survive in a model's memory.
• It does not prove that a distinction gets you recommended: it measures how often AI gives one as a reason, among businesses it already recommends. No customer was observed.