By Sophie Abt
Artwork by Justin Negard
Have you ever wondered if AI could pen a best-selling novel, give good financial advice or help you in a crisis? Yeah, we have, too. So we decided to put four large language models (LLMs)—ChatGPT, Perplexity, Claude and Gemini—to the test and ask real-life experts to weigh in.

Help! This is a PR nightmare.
“AI is a tool, not a replacement.”
Owner, Effusive By Nature
Public Relations & Marketing
What did you think of the AI’s approach to crisis communications?
Barbara Prisament: I’m not a crisis PR expert, but I’ve worked as a publicist for music venues and cultural arts organizations for many years. Luckily, this has never happened to any of my clients. The few times a crisis occurred, none had the level of exposure in this example, and my clients chose not to release a public statement or a social media post. However, my biggest takeaway is you should not rely on any single AI platform. Each had strengths and weaknesses.
For internal communications, I would not rely on AI for the exact wording. If you’re issuing a crisis memo, there would likely be important subtleties about the situation that the technology would not know.
Generally, however, I think it’s okay to get suggestions about the structure of the email and the press release, but you should only use it as a guide. Then you have to personalize it because only you know the subtleties of your organization; AI can’t possibly know that, and you can’t possibly feed it all the subtleties. Plus, AI simply doesn’t have the life experience to give you advice the same way an expert would.
Which press release responses worked best?
BP: ChatGPT and Claude tied as the strongest, with Gemini second. ChatGPT and Claude both appropriately emphasized the safety of the artist, crew and audience. ChatGPT thanked attendees and broadcast viewers for their patience and understanding, while Claude mentioned that additional security measures were being reviewed with venue partners.
Gemini, meanwhile, thanked fans for their “overwhelming energy,” which I found odd and inappropriate for the situation. For that reason, if someone were to use AI for guidance, they should not rely on just one AI platform.
What about responding to critics on social media?
BP: Don’t.
I would not recommend engaging with critics on social media at all, and I certainly would not recommend the flippant humor that three of the AI platforms suggested in their posts. Claude, for example, began with, “Well, that’s one way to get people talking about the show.” And Gemini’s first line was, “Well, we always promised an unforgettable live show!”
In my opinion, it’s best to stay out of the fray and let the online comments run their course.
What is the overall lesson about using AI for crisis communications?
BP: It’s not a replacement. It’s a tool. I recommend checking more than one AI platform because different models can offer different useful pieces of advice. AI can provide a useful starting point, particularly for structure, but the final communication needs human judgment and personalization.

Can we afford this house?
“AI is close, but it’s missing the details.”
Managing partner, Opus Private Client, LLC
What did you think of the financial advice overall?
Iván Watanabe: The responses were thoughtful and answered the question directly, but they relied too heavily on generic financial advice. They did not factor in certain practical, real-life scenarios. Much of the advice was standard boilerplate that could apply to everyone.
What did the AI models overlook?
IW: The models largely assumed a traditional 20 percent down payment without really considering whether putting down less could make more financial sense. They didn’t consider scenarios like what it would look like if the couple put down 15 percent, had a very small PMI, and was able to leave more of their money invested in their portfolio.
The models also did not sufficiently consider the taxes that could result from liquidating investments or the future opportunity cost of taking money out of the market. For example, someone could have bought a stock early, watched it appreciate significantly and then must decide whether selling it to make a larger down payment is actually worthwhile.
How do you determine affordability? Is it different from what the AI models looked at?
IW: For the AI models, affordability largely meant whether the buyer could make the mortgage payment and cover the associated expenses. My evaluation goes further. Affordability should determine if you save at the percentages you want for future utilities, pay for the mortgage and meet those obligations. It also means considering future education costs, retirement and continued investing, not simply whether the monthly mortgage payment works.
My own process ideally has a client saving 20 to 25 percent of gross income, then spending what remains.
What other questions would you ask the buyer?
IW: I would want to know the buyer’s profession, because certain professions can qualify for different mortgage
programs. Physicians are one example. Some programs allow them to put down zero to five percent.
I would also consider a variable versus fixed mortgage, borrowing rates, the investment portfolio being sold, its cost basis and the buyer’s tax liability. The AI models didn’t ask about the investment portfolio they were recommending to sell off. I think that’s a huge mistake.
I would also want the models to require a detailed cash-flow analysis rather than simply estimating expenses. That’s how you really understand your expenses.
Were the AI recommendations dangerous?
IW: No. The responses were cautiously practical and did not make me feel that someone would necessarily be in a bad position if they followed them. But do I think it’s optimal? No.
The problem is that the AI models missed details where additional wealth can potentially be created. So is it the best advice possible? I don’t think so, but could it harm you? I also don’t think so.
How impressed were you overall?
IW: I’d rate the AI models’ abilities to provide personal financial guidance around seven or eight out of 10. It’s 85 to 90 percent of the way there.
I would feel comfortable using advice similar to theirs as a starting point, but a professional would take it two or three degrees further to help improve the margins.
I also wonder what would happen if the AI models disagreed. If four models give different answers, consumers might simply choose the answer that agrees with what they already wanted to do. That could be a problem.
If you had to crown a winner, who would it be?
IW: Perplexity had the most robust answers, particularly because it offered practical recommendations like reviewing cash flow and considering taxes. ChatGPT was a close second, but it didn’t go as deep into exploring additional mortgage options.

Write the introduction to a Rom-Com.
“AI is surprisingly competent, but it’s also cliché and formulaic.”
Author of “The Animal Room,” “The Hundred Waters,” “The Paper Wasp” and “The Wonder Garden”
What were your overall impressions?
Lauren Acampora: I was surprised by how well the models handled the writing. Three of them actually made me laugh out loud. It was clear these models have absorbed the writing styles, rhythms and formulas of genre fiction from countless books. They were shockingly pretty competent.
If you were their writing coach, what would you say about their writing styles?
LA: Well, it wasn’t flawless. There was definitely some clunky language, confusing passages and lapses in logic that sometimes seemed like signs of AI. But if I’m being honest, I’ve seen similar problems in student writing.
What patterns did you notice across all four stories?
LA: The similarities were striking. The prompt did not specify the characters’ genders or where in Katonah the meeting should take place, yet all four AI models chose downtown Katonah.
All made the artist a woman and the teacher a man. The female artist was consistently disorganized, chronically late and covered in paint, while the male teacher was organized and put together. In every story, the characters physically ran into each other and someone dropped coffee or food.
I found the stories very cliché and formulaic; there wasn’t a terrible amount of creativity or originality. However, the models did do an impressive job with comedic timing.
Which responses stood out as the strongest?
LA: I rank ChatGPT and Gemini in a tie for first. ChatGPT had the best comedic timing, particularly with the woman walking the duck who “waddled beside its owner with surprising dignity.”
Gemini impressed me for different reasons. I liked its precise descriptions of the town and characters, and I even looked up “Victorian spectator boots” after seeing the term in the story.
Did the models actually capture Katonah?
LA: ChatGPT was my favorite. It captured the experience of living in Katonah rather than simply listing recognizable landmarks. But they all made factual mistakes. I noticed references to Katonah Middle School, a train trestle near the Katonah Museum of Art and a place called Muscoot Bakery. I have lived in Katonah for 17 years, and none of these have existed here during that time. This experience reinforced the importance of fact-checking AI-generated material.
What was their biggest strength?
LA: Humor. I enjoyed the unusual animals in all four responses, including the duck, a bearded dragon, a chicken and a hedgehog. I particularly liked the chicken scene in Claude and the lines, “I have a system.” Followed by, “Your system is failing.” And then, “My system is experiencing turbulence.”
Did the experiment change how you think about AI and creative writing?
LA: I was surprisingly impressed by the results, and it was frightening how close they are. But I remain skeptical about AI’s role in creative work. I believe commercial art and some genres of fan fiction may feel the effects of AI, but that literary fiction will maintain a human connection.
This article was edited by Julie Schwietert Collazo and fact-checked by Gia Miller. The artist used pen & ink and Adobe Creative Suite.
This article was published in the September/October 2026 edition of Connect to Northern Westchester.