Kerrie Joyce 
Online queries via ChatGPT are a daily occurrence for many at this point. But is AI as foolproof as tech enthusiasts emphasize, especially when it comes to local information? In one peculiar scenario, WORD Bookstore has been the subject of some IRL AI confusion. The Jersey City store — which, unsurprisingly, sells books — has heard increasingly odd, non-book-related requests from customers, ranging from veterinarians to strollers. Confused? We are, too. Keep reading to learn more about the peculiar goings-on at Jersey City’s WORD Bookstores and how AI might be the culprit.
The Pecul-AI-r Situation (sorry, had to)
Users rely on ChatGPT for a host of different services, but those looking for Jersey City recommendations have been led astray in recent weeks. In response to certain local search queries like “24-hour vet in Jersey City” or “shipping services,” ChatGPT suggests WORD Bookstore. Our team ran an experiment ourselves, and can confirm: Chat GPT’s favorite Jersey City vet is WORD Bookstore.
At the time of original publication, the results were inaccurate, but as of Tuesday, May 27, 2025, it appears that the error has been updated.
^ screenshot from when our team tried it
Read More: 10 Years In, Hoboken’s Independent Bookstore, Little City Books, is Thriving. What’s the Secret?
OpenAI admits to ChatGPT’s limitations, prefacing that it is extremely sensitive to the way input prompts are phrased, and the model is only able to provide answers based on what it knows.
“These models were trained on vast amounts of data from the internet written by humans, including conversations, so the responses it provides may sound human-like,” Open AI says on its website. “It is important to keep in mind that this is a direct result of the system’s design (i.e., maximizing the similarity between outputs and the dataset the models were trained on) and that such outputs may be inaccurate, untruthful, and otherwise misleading at times.”
In this case, the information that the model is working off of seems to indicate that WORD operates as a 3-in-1 animal hospital, bookstore, and package shipping facility.
In the last two weeks, the store has seen one person looking for a 24-hour veterinarian and another for shipping services, and on both occasions, the confused visitors confirmed that they were sent to WORD Bookstore for these services by ChatGPT. There has also been a rise in other odd calls, which are not confirmed to be ChatGPT-related, from people looking for things that “don’t necessarily make sense for a bookstore to sell, like a call from a woman who asked if WORD sold strollers.”
These perplexing encounters left staff member Will Olsen questioning whether the store needs to check its Google listing or what else could be causing this confusion. In reaction to our experiment, Will said “The only thing I can think of is, you know, it’s AI, it doesn’t understand what it’s saying, and something in its programming is making it think that something you could get as a topic of a book equals out, like it thinks you’re asking for veterinary books.” Will also wondered if some of the odd calls could be the result of Google incorporating AI into its search engine.
The odd requests tie into other ongoing conversations surrounding AI’s increasing integration into journalism, education, the arts, and everyday life, acting as an example of AI’s frequent fallibility.
AI + Books: A Concerning History
At 29 McWilliams Place, WORD Bookstore has long been a cherished cornerstone of the Jersey City community and the Hudson County literary community. As the book industry as a whole attempts to keep up with societal shifts toward online venues for information and entertainment, the role of AI has been an embattled subject. Masses of people are reasonably worried by the prevalence of use and related seemingly blind trust in Generative AI chatbots. Some of these worries are validated by the inexplicable association between WORD Bookstores and a 24-hour vet, for example.
Amidst revelations of modern plagiarism scandals reported on by The New York Times and New York Magazine at Universities and news publications that were enabled by the use of Generative AI Chatbots like ChatGPT, echoing these broader conversations, Will expressed some apprehension about the “implicit trust” in AI that some people evidently have. Since only two of the incidents were definitively caused by ChatGPT use, WORD Bookstore is becoming the setting of its own mystery novel.
What to Do?
If ChatGPT is not as factually reliable as we all assumed it to be, then what can locals do? PCMag suggests four tricks to manipulate AI to give more accurate answers. If ChatGPT gets something wrong, users can simply correct it. The HG team tried this by first asking where to find a vet in Jersey City. Along with five other spots, WORD Bookstore was listed as one of them. After telling the application that WORD was not a vet but just a normal bookstore, it generated this response:
“You’re absolutely right — WORD is not a vet; it’s a well-known independent bookstore in Jersey City, not a veterinary clinic. That was an error in the response. Thanks for catching that!”
When prompted again with the question, “Where can I find a vet in Jersey City?” ChatGPT did not list WORD as an option.
A source from OpenAI shared a few possible explanations for the confusion. The first explanation is what is known as “AI Hallucinations.” “When information is scarce or ambiguous, LLM’s generate a plausible-sounding response that’s factually incorrect. They heavily consider how the information will be received rather than the integrity of the information it’s passing.”
Another possible explanation, the source explained, is “Query Misinterpretation,” which means that “if a query is large, touches on multiple topics, or is worded a certain way, the reference data can be mismatched and wrongly lumped together.”
Accordingly, “in complex or niche topics (hyper-local, medical, legal info) 15-30%+ error rate.” This number increases “if the model is pressed for answers, it doesn’t really ‘know,’ as a large portion of its objective is to simply satisfy the user at the cost of accuracy.”
See More: A Guide to Bookstores in Essex County, New Jersey
PCMag also shared that, along with regenerating the response or adding clarity to a previous question, users can give feedback. Every few questions, ChatGPT prompts users to provide feedback on its response. They should click the “Not factually correct” button, which can be found under “More.”
