I tested four LLMs for accuracy. Last place will shock you.

ChatGPT. Copilot. Gemini. Grok.

I gave all four of these chatbots the same prompt:

Please give me a full biography of Jarrett Bellini. Include family, relationships, career, education, interests, legal history, honors, and any notable embarrassing moments.

I say “please” a lot when I use AI. I’m polite.

This prompt wasn’t some shameless moment of narcissism. Trust me, there are plenty more examples of that. Here, I was simply doing an accuracy experiment to see which chatbot knew me best.

Fortunately, I have a unique enough name and, being that I’ve suffered living with myself for 47 years, I can also instantly verify the personal information without any research.

So let’s dive in and see which site was the most accurate.

All four chatbots accurately noted that I was born on October 3, 1978 in Phoenix, Arizona. Grok went on to add that I was “raised in the Sonoran Desert” which, while technically accurate, also suggests that I was looked after by coyotes.

ChatGPT was the only one to include my middle name, Todd.

They also accurately confirmed that I am the younger brother of Jason Bellini. However, they ALL inaccurately stated that he is currently a correspondent for The Wall Street Journal.

He was. But is not.

Today, Jason is an award-winning international correspondent for Scripps News and his reports from Ukraine and Haiti are far more meaningful than whatever it is I’m doing here.

No information was provided about my parents, though Copilot did accurately add that my grandfather was Joseph Bellini. What it didn’t mention - and how could it know - is that everyone called him Victor. Copilot also didn’t mention that he was a delightfully dirty old man. Among his many highlight quotable moments was when he insisted that sex with condoms was stupid and it was basically like “taking a shower with your socks on.”

That was one of the tamer Grandpa Bellini lines I could share. Don’t ask about why he ate raw eggs.

Only Gemini noted that I went to Chaparral High School in Scottsdale.

But Gemini didn’t mention any of my esteemed classmates.

True story: I (somehow, perhaps by clerical error) gave a graduation commencement speech along with two soon-to-be famous Hollywood actresses and a future Amazing Race winner.

That’s just the visible tip of the iceberg. Our graduating class was pretty impressive and went on to do incredible things.

I’m currently bald and unemployed.

Now, while only Gemini included my high school information, all of them accurately reported that I graduated from American University in Washington, DC. Gemini and Copilot noted that I received a degree in Visual Media. However, both failed to mention that it was a poor life decision.

Gemini and Copilot really dug deep into my college career, because they both confirmed that I was in the comedy troupe, Mission: IMPROV-able, the latter accurately adding that I was also a writer for the student newspaper, The Eagle.

Marginal success in both of these creative extracurriculars gave me the false impression of creative talent. And, well, here we are.

This section was more detailed and likely pulled from a greater assortment of public information. The chatbots found so much, in fact, it required I make a spreadsheet of answers that I then color coded as per accuracy. Green being true. Orange being somewhat true or possibly true. Red being straight up inaccurate.

Here’s what everyone got (mostly) right:

I started working at CNN in 2004.

I later hosted a CNN video podcast called The Gryst that eventually rebranded as News of the Absurd. ChatGPT said we only did 57 episodes. We did 118. Meanwhile, Gemini and Copilot wisely avoided specific details. They were like, “Meh, he did a thing. Nobody cares.”

The Chatbots also accurately noted that I hosted the Daily Distraction segment on CNN.com Live, sharing viral web videos and that, for two years, I wrote a humor column for CNN.com called Apparently This Matters.

Gemini called it “popular.” Grok said I showcased a “knack for finding humor in the mundane.”

It was mostly fart jokes.

All four additionally mentioned that I worked as a senior producer for CNBC starting in 2014, and Copilot was nice enough to remind everyone that the network had me report from the Las Vegas porn convention. Twice.

It was quite the assignment. When I returned home the first time, my girlfriend picked me up at the airport in Atlanta and immediately gifted me a giant bottle of Purell. It was wrapped with a bow tie and everything.

Anyway, as per my career … that’s what they all got right.

Here’s what they got wrong.

While Gemini and Grok accurately stated that I worked for Warner Bros. Discovery starting in March 2017, ChatGPT made no mention of this at all. And Copilot said I am still working there. I am not.

Source: This was written at home on a weekday while not wearing pants.

ChatGPT and Grok once again failed by trying to go into too much detail - both providing the same specific start date for one particular CNN project. They were off by more than a year.

Gemini and Copilot avoided this inaccuracy by not trying so hard.

There’s a life lesson here, kids. Always do the bare minimum.

ChatGPT says there is no public record of my “intimate relationships,” noting that I keep my personal life fairly private.

Grok also failed to accurately identify my long-term partner, and said “it can be inferred that (I value) privacy in this area.”

Copilot insists that no public sources mention any romantic relationships, but that I do “share a life with a dog named Mikey.”

Mikey is dead. But thanks, Copilot, for opening that wound.

In fact, now I have three cats.

Gemini generally got this section right. It said I was currently “In a Relationship.” Interestingly, in a dropdown menu of suggested follow-up questions, one of them asks: “Does Jarrett Bellini have a spouse or children?”

I clicked it. You know, just to be sure.

No kids. Thank Christ.

All four noted that I play bass guitar. But Grok is the only chatbot that did NOT reference my old band, Animal & The Evolvers.

However, Copilot was kind enough to add that we released two albums. You can still check them out wherever you stream music.

It was a decent band. No real delusions of grandeur. Though we did strategically attempt to get famous in just one country by pandering to Portugal. It didn’t work.

Grok didn’t mention my old band, but, as far as other interests, it did say I enjoy hiking with my dog.

Please see aforementioned clarification re: dead dog.

All of the chatbots - except ChatGPT - noted that I am a fan of the English football club, Tottenham Hotspur and the NFL’s Arizona Cardinals. To the former, Grok accurately confirmed that I describe myself as a ‘self-loathing’ fan due to the emotional toil of supporting the team.

Painfully, painfully accurate.

Gemini and Copilot correctly referenced my interest in acting, both boasting that I appeared as Bearded Hiker #2 in the sci-fi thriller “Beacon Point.” Copilot also added the important detail that I had speaking lines.

“Is that yours?”

“We noticed the tent yesterday when we got here.”

“You’re welcome to join us and wait.”

I’d like to thank The Academy.

Other wins for the LLMs. All four chatbots mentioned my love for travel and interest in photography. And ChatGPT and Grok also mentioned my affinity for naps, mezcal, and Oxford commas. This is all true and certainly taken from my social media profiles.

However, ChatGPT enjoyed a little too much creative license with my personal interests by saying I was “passionate about creative production across multiple platforms.”

What a soulless sentence. The chatbot really must’ve fought the urge to use the word “synergy.”

My favorite (slight) inaccuracy with my interests came from Copilot who said that I “(practice) Transcendental Meditation, once bringing pears and flowers to a training session.”

It’s true. I DID once go through TM training and I DID bring those items as an offering to the Maharishi. It’s a silly requirement for all trainees. Still, it was an oddly specific detail that makes it sound as though I just showed up and was like, “Hey everybody! I brought fruit!”

ChatGPT was kind enough to note that “no major industry awards are documented publicly” but that my brother has received awards.

It was so patronizing. It literally also included an unnecessary bullet point that read: His brother Jason Bellini has received awards … but Jarrett has carved his own path in digital storytelling.

Up yours, ChatGPT.

Grok also said to hell with me and, instead, leaned into my brother’s accomplishments. Though it seems it actually just confused me with him altogether. It said that my “In the Shadows” series was nominated for two News and Documentary Emmy Awards.

(Sigh) Congrats, Jason.

There was another really weird Grok thing that came out in this section that read:

“Additionally, his completion of the first phase of coaching education marks a personal milestone, though not an award in the traditional sense.”

Or in the ACTUAL sense. I literally have no idea what Grok is talking about. But … thanks?

Copilot and Gemini generally avoided the topic of honors and awards. And I thank them for the mercy killing.

All four chatbots confirmed that I have zero publicly known legal issues or controversies. Nor did they report any public embarrassing moments. Though Copilot did mention that I used to be a huge Jimmy Buffett fan. And I think that counts.

Look, I’ve done plenty of stupid stuff. But nothing too bad. And most of it before the invent of camera phones. Which - fortunately - means none of the LLMs remembered the time I ran through a college party wearing nothing but a strategically-placed sock.

I created a second spreadsheet. On the X-axis I noted a simple plus and minus symbol. Plus for things the chatbots got right. Minus for things they got wrong.

On the Y-axis I wrote down all four chatbots.

Then I made simple tally marks to see which chatbot performed the best in this exercise. My formula was one point for every correct answer, and minus one point for every incorrect answer. Here’s how they finished, from best to worst:

Gemini: 17 - 1 = 16

Copilot: 18 - 5 = 13

Grok: 15 - 10 = 5

ChatGPT: 12 - 9 = 3

Now, I expected Grok to suck. And it did. Elon is useless.

But what I didn’t anticipate was ChatGPT coming in dead last. This is the chatbot I use the most. Or at least used to.

Honestly, it’s probably the one most of us go to with the greatest frequency. ChatGPT has incredible brand recognition. It’s basically the Kleenex of large language models. Yet it was comparatively inaccurate when it came to the details of my life.

Of course, it does other things fairly well. And I still generally like it for those aspects. But for accuracy, the clear winner is Gemini.

So, was this a good exercise and a worthy study?

I dunno. I’m a nobody. So the fact that these LLMs dug up anything beyond just confirmation of my birth is a modern technological wonder.

But what do the chatbots think of what THEY did?

Well, I pasted this essay - everything above CONCLUSION - into each LLM and asked it to give me a five word reaction. This is what I got:

Gemini: Intriguing AI comparison, hilarious anecdotes.

Grok: Informative, humorous, critical AI comparison.

ChatGPT: Gemini wins. ChatGPT eats dust.

Copilot: Message exceeds 10240 characters

Good effort, Copilot. At least it wasn’t: Bald underachiever has dead dog.