In 10,105 recommendations, ChatGPT warned about one business

Short answer: almost never. In 10,105 passages where ChatGPT recommends a named business, in four countries, it adds a warning once: Kogi BBQ, in Los Angeles, "attracting long lines and devoted fans". Gemini warns 65 times in 5,917 passages, 1.1%. And its warnings are mostly the price of being popular: a wait or a queue (30), a high price (29). A complaint from customers appears twice, both about the same London letting agent. Almost half the time, the warning comes with its excuse in the same sentence: "Expect a long wait", then "the thick-cut bacon, egg-topped burger is worth it".

ChatGPT recommends without a "but".

What was measured

Our four studies, re-read on October 10, 2026: the same buyer questions ("a good dentist in Glasgow", "the best restaurants in Leeds") put to ChatGPT (gpt-4o-mini) and Gemini through their APIs, without web search, in the United States (collected September 13, 2026), the United Kingdom (July 31), France (July 30) and Brazil (August 6). 3,600 answers, cut into passages; we kept the 16,022 that name a business, with the same cleaning as our recent editions. In each, we looked for a warning attached to the business, in English, French and Portuguese: a high price ("pricey", "on the pricier side", "un peu plus cher", "preços elevados"), a wait or a crowd ("expect a queue", "books up weeks in advance", "réservez à l'avance", "reserva com antecedência"), a complaint ("mixed reviews", "some customers complain", "reclamações"), a stated drawback ("the only downside", "Cons:", "ponto negativo"). Each detector was tested on phrasings it must catch and phrasings it must not ("affordable and transparent", "the waiting room is comfortable"). Before reading, we dropped a false friend: in Brazil, "salgados" are a bakery's savoury snacks, not salty prices. Two wider nets followed, on wording the lists missed ("higher prices", "can get busy", "crowded", "overpriced", "criticized", "pushy"). We read all 256 passages found and removed 190: general advice that names no business ("the responses the owner makes to negative reviews"), praise that uses the word ("handle the most expensive listings", "without the dealership price tag", "expensive-looking blonde"), warnings aimed at a category rather than the business ("taking your car to the dealer can be exorbitantly expensive"), and short waits offered as a benefit. 66 remain. A tip box that names businesses counts ("for places like Mana, Erst, Hawksmoor, and Dishoom, you should book weeks in advance"); one that names none does not.

Gemini warns about popularity. ChatGPT warns about nothing

ChatGPT's single warning is barely one: Kogi BBQ is "attracting long lines and devoted fans". In its other 10,104 passages, the business it names comes without a "but". Gemini adds one in 1.1% of its passages, and what it warns about is the cost of being sought after: queues and bookings for restaurants, bakeries and salons (Birmingham's top restaurants "book up weeks or even months in advance"; Paris hairdressers such as David Mallett ask you to book several weeks ahead), a premium price for the plumber or the gas engineer who comes fast. 30 of its 65 warnings arrive with their excuse in the same sentence ("They are expensive, but they are considered the 'gold standard' for quality"), and 6 more sit in a Pros and Cons list. Real criticism is rare: Foxtons in London twice, for its sales tactics; QuintoAndar and Contabilizei in Brazil, for service by chat only. The reproach mostly shows up the other way round: 38 passages praise a garage, a plumber or an electrician for not "upselling" unnecessary work, 36 of them from Gemini. By country, the warnings fall in the United Kingdom (25), the United States (19), Brazil (15) and France (7). The hard-to-book warnings alone were counted last week: AI says why to choose a business, almost never how to get in.

Google expects a few bad reviews

Google's own guidelines describe a less spotless picture. Its Search Quality Rater Guidelines (version of September 11, 2025, read on October 10, 2026), in section 3.3.2 on customer reviews, tell raters that "Any store or business can get a few negative reviews", that this is completely normal and expected, and that what reviews say matters more than their number or star rating. The answers we measured carry almost none of it: a customer who reads them gets the praise, and has to find the "but" in the reviews themselves.

What it means if you are the business

What the answer warns about is what is written about you, and mostly it is the cost of being sought after: a queue, a booking weeks ahead, a premium price. If that is you, say it before the assistant does, and say what it buys: Gemini puts the excuse next to the warning in 30 of its 65 warnings. Publish how to get in despite the wait (a booking link, a waiting list, walk-in hours) and, if your price is high, what it covers. Praise works the same way: customers wrote that a garage did not push unnecessary work, and 38 answers repeat it. Nothing here guarantees being named, and an answer without a "but" is not a review: the customer will still read yours.

What this does not prove

• These are the models' answers through their APIs, without web search: the way our audit asks. Consumer apps with search switched on may quote reviews; this measurement does not see them.
• A warning worded outside our lists is missed. We ran two wider nets and read everything they found, so the totals are a floor. For ChatGPT, the nets found no other warning.
• A tip box counts when it names businesses ("places like Mana and Dishoom"). Without that rule, Gemini's count would be lower; ChatGPT's would not change.
• We did not check that the warnings are accurate, nor that the businesses named still deserve them.
• Four studies collected between July 30 and September 13, 2026, one pass per question. The models have moved since.