Google's AI lied and said a restaurant served brunch
What's the big deal? Well...
When Google debuted AI overviews in search results about a year ago, the company conceded that AI can and will make mistakes. To paraphrase: It’s new, it’s somewhat experimental, and we’re all sort of figuring out how this thing works. But recently I heard a story about the impact on a real restaurant when the bots get it wrong. I’m sharing it here, because I think it balances our excitement for technological progress while flagging the imperfections that cause trouble for restaurants.
A few weeks ago, I moderated a conference panel about artificial intelligence in restaurants. During that session, one audience member1 shared a story about their experience with AI:
Recently, they said, their one-location restaurant got some frustrated feedback from diners trying to visit for weekend brunch. The restaurant’s staff were confused because the business had never, in its years-long history, offered brunch. It couldn’t parse the sudden interest.
Eventually, the team realized Google’s AI overview, appearing at the top of the results page for some searches, was confidently telling searchers that the restaurant served brunch. That discovery led them to the actual culprit: a years-old, incorrect blog post from a local visitor’s bureau promoting the restaurant’s nonexistent brunch service.
Immediately after hearing the story, the vibe among the restaurant pros seated in the conference theater shifted to something like, We didn’t ask for this. Why is it happening?
As Google weaves artificial intelligence further into its search results, we’re all subject to its experiments and errors.
I’d argue we expect it from a new and rapidly evolving technology. Google’s not a lone offender here; most (if not all) companies rolling out generative AI, including OpenAI and Anthropic, acknowledge the tech doesn’t always get its facts straight. It’s all part of the process.
But as Google, which handles a reported 90ish percent of the world’s searches, adds more AI into its mix, businesses know they must adapt to continue using the decades-old search engine to their advantage. That doesn’t mean it’s easy or efficient; in this case, one restaurant’s staff spent valuable time hunting down a previously irrelevant source of bad information suddenly impacting their operations and reputation.
Google’s blog post announcing AI overviews in search is over a year old, but in it, Google’s search lead Liz Reid shared some early observations of the young tech. People visited a wider variety of websites than they would have previously thanks to the source links included in AI overviews, she wrote. Additionally, links included in those overviews got more clicks than they did in Google’s organic search results. (According to a recent report, news publishers might disagree.)
Based on this info, I’ll infer this is what happened to the brunch-less restaurant in question. Google’s AI wanted so badly to produce a relevant answer to that brunch query that it went deep inside its own search results to find a definitive response to the searchers’ prompt. Unfortunately in this instance, the source material was wrong.
At my conference, the restaurant pros assembled in the room seemed worried. What can a restaurant do to insulate itself against further misinfo on the internet, attendees asked. Is it now the restaurant’s responsibility to monitor every Google-able channel to intercept and correct bad info?
I asked Google about this, looking for guidance about how restaurants might get on top of future AI searches as the tech changes. A company representative pointed me toward an eight-page overview of AI searches that says, in part:
“AI Overviews are built to surface information that is backed up by top web results, and include links to web content that supports the information presented in the overview. People can use these links to dig deeper on a topic, because we know people come to Search to find a range of perspectives and to explore the open web.”
And also:
“We approach improvements to AI Overviews in the same way that we approach Search more broadly: we learn from examples and feedback to identify areas of improvement. We come up with solutions that we believe could help not just one query, but a broad range of similar searches. People can provide feedback on the AI Overviews they see in Search, which helps us improve.”
In my own search for info before the rep responded, I checked Google’s help pages for the company’s explanation of AI overviews, the same place a restaurant trying to understand the newish tech might look. It suggested a fact check.
“Think closely about the responses you get from generative AI tools,” the company advises. Then, almost ironically, Google suggests using Google search — old Google search, presumably — to confirm AI results. Fair enough, I guess; in the case of the fake brunch listing, a human peek at the restaurant’s operating hours or online menus — or even a primitive phone call — would’ve cleared up the confusion.
Of course, encouraging a human confirmation relies on our likelihood to… double-check the computer’s work. I unfortunately couldn’t find any relevant data or research confirming people’s willingness to do this. I did, however, discover a recent analysis by a SEO company that says over a third of searchers click through to the first link in Google’s AI-generated overview. Interestingly, almost exactly the same percentage will click on the first organic search result. And as we learned a long time ago, people rarely look past the first few results.
Whether or not we choose to do it, individual fact checks seem at odds with the top-of-the-page “quick answer” Google says it’s using AI to offer. One industry expert panelist onstage with me that day helpfully offered suggestions to ready a restaurant for future changes: Clean up and update your digital footprint, remove outdated info and menus from circulation, and add as much detail as you can — highchairs? gluten-free offerings? late-night service? — to your own restaurant’s online platforms.
Has your restaurant run into trouble as tech companies introduce new products and updates? I’d love to hear about it — just hit reply!
Related coverage:
I regret that I did not ask the person who shared this story where they worked. If this was you or someone you know: 1. Thanks for freely sharing your experience working with an evolving and imperfect technology, and, 2. Please tell me who you are!