Stefanina's Wentzville learned the modern local-business lesson: you can publish your own menu, run your own Facebook page, and tell customers your actual specials, but a search box can still walk in wearing a fake coupon mustache and start negotiating on your behalf.

In August 2025, the Missouri restaurant posted a warning asking customers not to use Google AI to find its specials. The restaurant said Google AI was not accurate, was telling people about specials that did not exist, and was causing angry customers to yell at employees. It also made clear that it would not honor the AI-generated specials.

First Alert 4 reported that the issue involved misleading or false information in AI-enhanced search results. Eva Gannon, part of the family that owns the restaurant, told the station customers were asking for deals or offers the restaurant was not actually running. One example was an AI Overview saying Stefanina's would offer a large pizza for the same price as a small. The restaurant said some generated menu information was wrong too.

This is a smaller story than a court sanction or a production database wipe. Nobody died. No one lost millions. But it is exactly the sort of everyday AI failure that turns platform slop into someone else's customer-service shift.

The fake deal problem

A fake restaurant special is not just a wrong fact. It creates an expectation. A customer sees a generated answer from Google, assumes it reflects the restaurant's current offers, and walks in or calls expecting the deal. When the employee says no, the customer experiences that as the restaurant refusing to honor something that looked official.

The restaurant did not make the promise. Google did not have to stand at the register and explain it. The employee did.

That is the asymmetry. AI search systems increasingly present answers in a voice that feels authoritative, but the businesses named in those answers absorb the fallout when the answer is wrong. A platform can call it an imperfect generated summary. A server dealing with a customer demanding a nonexistent pizza discount calls it Tuesday, apparently.

First Alert 4 said Google did not respond to the station's messages about the situation. The station noted that Google's own guide warns that AI can sometimes be inaccurate because it may make mistakes or misunderstand the information it is searching through. That disclaimer may be true, but it is cold comfort when the false information is displayed to customers before they ever reach the restaurant's actual page.

Why AI search is bad at specials

Restaurant information is a nasty target for automated summaries. Menus change. Specials change faster. A "Tuesday special" might be seasonal, expired, dine-in only, location-specific, or posted by a similarly named restaurant somewhere else. Old Facebook posts, third-party menu aggregators, customer photos, delivery-app pages, and local news mentions all float around the web with different timestamps and different levels of reliability.

A traditional search page can still mislead users, but it at least shows separate sources. A user might see the restaurant's official page, a third-party menu, and a Facebook post, then choose which to trust. An AI Overview turns that messy pile into a single answer. If it blends an old post, a wrong menu item, and a pricing assumption into a clean sentence, the user sees a deal.

That clean sentence is the danger. It removes the provenance. A generated answer saying "large pizza for the price of a small" does not feel like a rumor from a stale page. It feels like Google summarizing the current truth.

The platform's incentives make this worse. AI search products are built to answer, not to shrug. "Check the restaurant's official page" is less flashy than inventing a tidy deal list. But for local commerce, refusal and routing are often the safer product behavior. If the system cannot verify an active special from an official source with a current date, it should not answer as though it can.

The small business blast radius

The measurable harm here was staff time, customer confusion, and public reputation. The restaurant had to publish a warning. Employees had to explain that the Google AI specials were not real. Some customers reportedly yelled at staff. The business had to refuse false offers without looking like the bad guy.

That is not catastrophic, but it is not nothing. Local restaurants operate on thin margins and thin patience. A fake discount can pressure employees to comp food, argue with customers, or absorb bad reviews from people convinced the business reneged on an offer. Even when the restaurant stands firm, the conversation costs time during service.

There is also a broader brand risk. Search is often the front door for local businesses. If the first thing a customer sees is wrong, the business may never get a chance to correct it. Worse, the correction can make the business sound defensive. "We will not honor Google AI specials" is a perfectly reasonable statement, but it is not the kind of sentence a restaurant wants to have to post.

Same pattern, different counter

Stefanina's belongs next to the other AI search and chatbot customer-service failures because the underlying mistake is the same: a platform put generated text between a business and its customers, then the generated text made a promise the business did not make.

Air Canada's chatbot misstated a bereavement fare policy and the airline was held liable. Google's AI Overviews told users to glue pizza and eat rocks, which was funny because the examples were absurd and horrifying because the product was default search. Overland Park Farmers Market had shoppers directed to a construction site because AI summaries mangled a location transition. Stefanina's had customers demanding fake deals because Google AI turned local business information into coupon fan fiction.

The lesson is not complicated. If an AI system speaks with apparent authority about prices, policies, hours, locations, or discounts, it is doing operational work. Operational work needs current authoritative sources, timestamps, and a low tolerance for improvisation. A restaurant special is not a creative-writing prompt.

For local businesses, the defensive playbook is tiresome but practical: keep official pages current, use clear dates on specials, remove stale posts where possible, and monitor search results. But the burden should not sit entirely with the business. Google and other platforms are the ones choosing to synthesize local commerce facts into answers. They need to treat "active offer" as a high-risk claim, because customers will try to redeem it.

The Stefanina's incident is small in dollar terms, which makes it easy to dismiss. That would be a mistake. Most AI failures in public life will not look like science fiction. They will look like employees getting yelled at because a platform invented a pizza discount and then left the restaurant to mop up the sauce.