Product1 publisher3 min readPublished
Diners are citing ChatGPT to overrule their server's shellfish warning
Madison, a New York server, told The Verge that guests with shellfish allergies dispute her warnings because ChatGPT says the dish is safe. Staff have improvised polite lines for invented wines, but allergens need a written house rule that holds when the guest disagrees.
The Product Desk · Product desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- At the Michelin-recommended Italian restaurant where Madison worked until recently, more and more guests sent the sommelier away so they could ask ChatGPT instead.
- A federal river ranger in the Mountain West said visitors arrive with AI itineraries listing invented campsites and days that would require rowing upstream.
- A New York after-school teacher estimated that a child on a ChatGPT-built bedtime routine needed two to three more hours of sleep a night.
- The Verge also heard from movie theater workers, baristas and Apple Store staff whose customers asked for products or features that do not exist.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- exposure When the chatbot wins the argument, the guest with the allergy carries the risk, because the server's spoken warning is the last check before the broth reaches the table.
- decision Allergen training that ends once the server has asked the question now has to cover what a server says, and who steps in, when a guest's phone contradicts the kitchen.
- precedent The Verge's reporter expects Meta's new Muse agent to add to the problem, so the share of guests arriving with a chatbot's answer in hand is more likely to grow than to fade.
"You're allergic to shellfish, and ChatGPT says there's no shellfish in this, and we are telling you, yes there is!" said Madison, a New York server who asked The Verge to withhold her last name [4][1]. "They just want to talk to ChatGPT," she said [14].
Madison asks every table about allergies when she greets it [1]. That question assumes the guest will carry the answer through to the order. What guests actually do, by her account, is tell her "they have a shellfish allergy," then "order a fish dish that comes with a broth that is made of shellfish," without asking her any questions [2].
In the accounts The Verge collected, guests use the chatbot to outrank the staff member who has the actual information [11]. Jason, the river ranger, said: "The craziest part is that even with a ranger explaining this while pointing at a map, we've had multiple groups remain skeptical. A certain type of AI user is inclined to trust ChatGPT over rangers, maps, and printed guidebooks." [11] Customer service workers on Reddit and X complain about customers who take ChatGPT's word as gospel, according to The Verge [15].
Staff already have a script for the low-stakes version. Some guests ask for bottles the restaurant does not carry, or that do not exist at all [6]. To "soften the blow," Madison and her colleagues say, "We have this really hyper-specific wine list, maybe it's showing you a French wine, and we don't carry French wine," instead of telling the guest the chatbot made up the brand [7]. The line lets the guest keep their dignity. It costs the restaurant nothing. The Verge's account does not describe any employer policy for these disputes, and the lines in it are ones staff worked out for themselves [7].
That softness is wrong for an allergen. A face-saving line leaves the guest's belief intact, and a guest who still believes the chatbot can still order the broth. I think the allergen response belongs in a written house rule that a manager owns. The server states what is in the dish from the kitchen's ingredient record and does not debate the chatbot. If the guest declared the allergy, the kitchen does not send that dish to that seat. The cost is a guest who feels contradicted in front of their table, and some will leave unhappy. A manager should accept that cost before the next shift, so a server does not have to weigh it alone in the middle of service.
Two questions sort anything a guest brings from a chatbot. The first is who pays if the guest is wrong, their evening or their health. The second is whether staff can check the claim on the spot. Wine that is not on the list costs an evening and can be checked against the list, so the soft line is fine. A shellfish broth costs health and can be checked against the kitchen's record, so the written rule applies and the guest does not have to agree. A child's sleep schedule costs health and cannot be checked at the door. Casey, the after-school teacher, had only her own estimate, and the parents "just insisted that their ChatGPT was telling them that their schedule was fine," she said [13]. There, the job is to say it once and write it down for whoever owns the decision. Low cost and uncheckable is the quadrant to let go.
What to watch
- Whether restaurant groups write a chatbot-dispute response into their allergen procedures instead of leaving it to individual servers.
- Whether chatbot makers start telling users to confirm allergens with restaurant staff when asked if a dish is safe to eat.
- Any reported allergic reaction tied to a guest choosing a chatbot answer over a server's warning, the case that would bring lawyers and insurers into it.