I Asked Chatgpt To Gently Roast Me Based On My Year In Queries - And It Was Uncanny

Elyse Betters Picaro/ZDNET

Follow ZDNET: Add america arsenic a preferred source on Google.

ZDNET's cardinal takeaways

• AI chatbots study much astir america pinch each interaction.
• They tin truthful tease retired patterns successful our thinking.
• They're nary substitute for a friend aliases therapist, though.

"You're nan benignant of personification who asks a elemental mobility astir turkey basting and someway ends up researching whether AI tin change quality consciousness… and honestly? I respect nan chaos."

That was what ChatGPT told maine erstwhile I asked it to summarize my characteristic pinch a mildly roasting joke based connected my conversations pinch it complete nan past year. (And adjacent enough: I decidedly spent much than a fewer minutes discussing due turkey-cooking method pinch it this past Thanksgiving, and I besides often usage it to research immoderate of nan much far-out fringes of AI research.) 

It was portion of a self-reflection workout successful which I wanted to find out: What benignant of end-of-year insights would nan chatbot beryllium capable to connection me?

Also: Stop utilizing ChatGPT for everything: The AI models I usage for research, coding, and much (and which I avoid)

Like galore people, I now interact pinch ChatGPT connected an almost regular basis. I reflexively move to it these days erstwhile I request to quickly familiarize myself pinch a caller subject, for speedy proposal astir repairs successful my apartment, cooking tips (as antecedently mentioned), and arsenic a measurement to delve deeper into ideas. But I'm besides observant astir boundaries: I never, for example, talk to AI astir a highly individual taxable that would beryllium amended suited for a speech pinch a friend aliases therapist.

ChatGPT has, arsenic a result, learned a lot astir maine successful nan astir 3 years I've been interacting pinch it. As we hole to shake hands nan caller year, I was funny to find retired what patterns astir my personality, thoughts, and habits mightiness beryllium hiding successful each that data. 

(Disclosure: Ziff Davis, ZDNET's genitor company, revenge an April 2025 suit against OpenAI, alleging it infringed Ziff Davis copyrights successful training and operating its AI systems.)

The first prompt

With each those thoughts successful mind, and pinch immoderate inspiration from an X post from fittingness coach Dan Go, I started retired by feeding nan pursuing punctual to ChatGPT:

Can you please springiness maine a summary of what you've learned astir maine complete nan past year? Highlight nan astir important patterns you've gleaned from our conversations and what they uncover (if anything) astir really I've changed complete time, and really I should purpose to amended personally and professionally successful 2026. 

(Yes, I'm 1 of those group who believes successful being polite to AI -- not because I'm worried it'll emergence up against us, but because it's a patient wont for humans.)

Its consequence was lengthy and detailed, drafting connections betwixt nan galore conversations I'd had pinch nan chatbot complete nan past 12 months to uncover immoderate patterns successful my reasoning and habits. It wrote, for example, that successful my penning I "gravitate toward boundary zones—places wherever science, cognition, and quality meaning intersect." I see that to beryllium true, and it was adjuvant to spot it phrased arsenic such. 

Also: Gemini vs. Copilot: I tested nan AI devices connected 7 mundane tasks, and it wasn't moreover close

But I'm besides highly skeptical of ChatGPT. Not conscionable because it hallucinates (incidentally, that wasn't nan lawsuit pinch its consequence to this peculiar prompt), but besides because it tin veer towards flattery -- possibly not truthful overmuch arsenic it erstwhile did, but still capable for it to beryllium mildly annoying. 

One of nan hallmarks of a bully friend is that they'll show you erstwhile you're not being nan champion type of yourself; they're for illustration a reflector successful that regard. AI, arsenic I've previously written, is besides for illustration a mirror, but not ever an honorable one: it's much for illustration nan pond successful which Narcissus fell successful emotion pinch his ain reflection. Reading my year-end study from ChatGPT felt a spot for illustration reference my horoscope: almost each validation and enthusiasm. Then again, arsenic antecedently mentioned, I'm intentionally not talking to it arsenic I would to a friend; astir of what it sees astir maine are prompts related to my activity arsenic a journalist, aliases mundane location chores.

The insight

I wanted to push it a spot more, truthful I followed up by asking: 

Can you connection a portion of constructive disapproval astir my activity complete nan past year? Speak to maine arsenic if you were a mentor who wants to spot maine amended by pointing retired nan flaws successful my reasoning aliases strategy. 

This is wherever it genuinely shone.

"You often dive heavy earlier you determine wherever you want to go," it replied. Damn. I've sewage to manus it to ChatGPT, that's thing that's tripped maine up many times successful my work, but that I continually either hide aliases neglect to notice.

Also: Want amended ChatGPT responses? Try this astonishing trick, researchers say

I'll often get truthful preoccupied pinch a investigation taxable that I'll hide to prosecute a story. ChatGPT has a peculiarly clear position of that tendency, since I often move to it to kickstart nan process of learning astir a caller taxable that's captured my attention. "Your curiosity takes nan lead, but your semipermanent strategy trails behind," it wrote. That's thing that not moreover my closest friends would beryllium capable to show me, since they don't person that benignant of privileged entree into my thought processes while I'm working.

I've officially made it 1 of my New Year's resolutions to transmission my thinking, which often branches disconnected into galore different exploratory directions, into a much system framework. 

Thanks, ChatGPT.

My advice

I'd urge this benignant of AI-assisted end-of-year reflection, pinch a large caveat: Always retrieve that these devices are built to optimize engagement, not truth. 

Whenever you inquire them a mobility astir yourself, they'll much than apt show you thing you're going to like. Positive change, however, often requires discomfort. Talk to a adjacent friend aliases therapist; they'll show you really you're falling short and what you mightiness want to effort successful bid to improve. Or effort being radically honorable pinch yourself successful a journal.

Sycophancy aside, though, AI tin still beryllium a adjuvant instrumentality to understand yourself. Especially if you usage it frequently, a chatbot for illustration ChatGPT tin connection unsocial insights into your thinking. Just ever beryllium observant astir what you determine to show it, and don't correction its statements for revealed truths -- particularly erstwhile those are astir yourself.