I'm an ex-barista and I challenged ChatGPT to a latte-making contest — here's who won

Tom's Guide

7 hours ago

Picture this: it's the final of the World Barista Championship (yes, a real thing, and no, I've never been). There's a human barista, who's spent years honing their craft, researching their passion, and inventing brand new drinks to hopefully take the title and be crowed the World's #1 Barista. Their coffee is nuanced, innovative, and clever.
Then there's a robot arm churning out coffee after coffee to a precise, scientific recipe programmed into its motherboard. Its coffee is caffeinated, comfortable, and most importantly: consistent.
There's nothing inherently wrong with either. When I lived in South Korea, I visited these robot-staffed cafes, because, well, I was curious. And while the coffee is drinkable, it's missing that oomph that human-made coffee does. Taste is mostly why I stay away from automatic espresso machines — I just don't think they can brew coffee as well as a human can. The best espresso machines are the ones that need a human hand. Prove me wrong.
This is the espresso machine I used in this challenge. I use this machine every day, and while it's not perfect, it's an awesome all-in-one machine that stops coffee getting everywhere thanks to the internal tamper.
At the weekend, I walked past one of my favorite cafes in my city. As the smell of freshly-ground specialty coffee permeated the air and meandered into my nose, I had to stop myself from going in and ordering my old faithful: an oat milk flat white.
You have oat milk flat white at home, I told myself. So I moseyed back to my house and got brewing.
Then I remembered those semi-cute-semi-scary robot arms in Korea, and the idea hit me: What if I ask ChatGPT how to make the best coffee?
Although I love flat whites, I understand lattes to be a more common beverage, so I decided to pit two lattes head-to-head.
Get instant access to breaking news, the hottest reviews, great deals and helpful tips.
I opened up ChatGPT on my laptop and prompted: "How do I make a perfect latte?"
This is what ChatGPT said (note: scroll to see both images).
If you know anything about espresso, ChatGPT's first mistake will be glaringly obvious.
While it recommends 18g-20g of coffee (correct), it goes on to say a double shot is 2oz of espresso in 25-30 seconds. 2oz is around 56g, which is almost double the size of a double shot. This is more like a lungo shot, which I wouldn't recommend for a latte — it'll weaken the flavor.
Generally, the optimum ratio is 1:2, which means I'd get 36g of espresso out of 18g coffee.
And then to recommend I get 56g of espresso in 25-30 seconds? That means my coffee would need to be ground way too coarse to be espresso-fine. I was skeptical, but I committed to the bit and followed ChatGPT's advice exactly.
The next issue was ChatGPT telling me to "let[ting] air in [the milk] for 1-2 seconds". This is simply not true. I knew just from reading these instructions that my milk would be way too wet.
I also don't like medium or dark roasted coffee (much prefer a juicy light-roasted African bean), but I persevered for science.
Still, I committed to the bit and followed the instructions exactly. Here's what happened in the battle of ex-barista vs. AI. Spoiler alert: I won.
It resulted in a bitter, over-extracted taste. I was not impressed.
I followed ChatGPT's instructions exactly. Using my Wacaco Exagram Pro scale (one of the best coffee scales, IMO) I weighed out 18g of finely-ground coffee (notch 3 on my De'Longhi La Specialista Opera).
I tamped with the Opera's smart tamper and slotted the portafilter into the group head. I placed the scale underneath my latte glass and started extraction.
By 25 seconds, I only had 31g of coffee. Still, I waited, adamant on following ChatGPT's guidance. I stopped extraction when the scale read 56g, which took a massive 40 seconds.
This extraction time is about 15 seconds longer than I'd usually pull a shot for, and resulted in a bitter, over-extracted taste. I was not impressed.
Now, the milk. ChatGPT told me to aerate for just 1-2 seconds, then plunge the steam wand into the milk until it reached around 140°F, so that's what I did.
Reader, it was not good. I know from experience that you should probably aerate milk (gently) for around 5 seconds, or until it feels right. A lot of coffeemaking relies on feelings and vibes, which obviously an AI can't relate to.
Because the milk wasn't aerated enough, I wasn't able to get latte art.
Here's a photo of the ChatGPT latte.
Now I'll walk you through my process.
Using the same Wacaco Exagram Pro scale, I weighed out 18g of the exact same dark roast coffee in my portafilter. I tamped with the same tamper and slotted the group handle into the machine.
After placing the latte glass and scale under the portafilter, I started extraction. I timed the shot with my scale. When the espresso reached 36g, at 26 seconds, I stopped extraction.
Then I poured just under 8oz of the exact same whole milk and steamed it my way. For this, I made the little tss-tss-tss sound for about 5 seconds and created a whirlpool in the center of the jug. I moved the jug up and down to ensure my milk was steamed evenly and had a nice layer of microfoam.
Here's a photo of my latte.
No, the latte art isn't my best work. You'd have to check out my other reviews (like the De'Longhi Stilosa) to see some really nice art.
However, in terms of both looks and taste, my latte was much more pleasant. The coffee was sweet and balanced with a little touch of bitterness, whereas the ChatGPT latte was overwhelmingly bitter thanks to the lungo shot.
My latte was also more pleasing to drink thanks to the textured microfoam.
Me. Duh.
I joke — I actually wanted ChatGPT to win, to prove me wrong. I wanted to rise to the challenge.
But there wasn't really much of a challenge. I was simply left disappointed by how misguided ChatGPT's recipe was.
After crowning myself the winner, I thought maybe I shouldn't have been the only judge. Of course I'd give myself the title.
Thankfully, I have a completely unbiased test subject in the form of my boyfriend (sarcasm intended). I gave him both lattes and asked him which was better.
"This one looks better," he said (direct quote), while pointing to mine. He took a sip and said, "It tastes good." After I asked him which once tasted better, he shrugged and offered, "They both tasted good."
Hm.
He's the kind of person who wants hot bean juice and wants it now, whereas I'm the kind of person who'd rather drink no coffee than bad coffee. So maybe I am biased. Maybe ChatGPT's recipe isn't so bad after all.
But I think for other semi-coffee-snobs like me, human touch can never be replicated.