Photo: Sanket Mishra
Hey everyone. I need to admit something. Last month, I typed a question about a medication interaction into ChatGPT at almost midnight. Not because I couldn't call my doctor. Because it was midnight, and the answer felt one search away instead of one appointment away.
If you've done the same thing, you're definitely not alone. Turns out, this is basically all of us now. It's not some fringe habit I need to feel weird about. It's apparently just what most people do now when a question won't wait until office hours. Honestly, once I started noticing it in myself, I started noticing it everywhere in conversations with friends too.
Where AI Actually Helps Me
I want to be fair here, because AI has genuinely made a few things easier, and I don't think pretending otherwise would be honest.
Before appointments, I've used it to organize my thoughts. What symptoms actually matter enough to mention. What questions I keep forgetting to ask out loud, every single time, until I'm back in the car kicking myself for not asking. It's less "diagnose me" and more "help me not waste ten minutes of my doctor's time fumbling through my notes app while they wait patiently for me to remember what I actually came in for."
It's also handy for the boring stuff. What a medication is generally used for. Whether a side effect I'm having is common enough not to panic about at eleven at night, when everything feels a little more dramatic than it probably is. That kind of low-stakes, quick-reference use feels pretty reasonable to me, and honestly, it's probably saved me a few unnecessary calls to the on-call line over the years.
There's also a quieter benefit I didn't expect. Sometimes I just need the vocabulary. A way to describe what I'm feeling so I don't stumble through an explanation in the exam room. AI is oddly good at that specific task, even when I wouldn't trust it with much else.
Why So Many of Us Turn to AI for Medicine Advice
Here's where it stopped feeling like just my weird midnight habit. SingleCare's newest survey found that 46% of Americans have used AI to answer a medication-related question. Not a small, quirky minority. Almost half of everyone you know has probably done exactly what I did that night.
And it's not just casual curiosity. Of the people who used AI this way, 49% said it actually changed how they took a medication afterward. That's a real behavior shift, not just idle Googling between episodes of something on Netflix while half-listening.
The reason most people gave was convenience, cited by 47% of respondents, more than any other factor. Fourteen percent said it was because they wanted to avoid paying for a visit, and 17% said they simply wanted answers outside office hours. Honestly, reading that list felt like reading my own midnight ChatGPT session back to me, word for word, uncomfortably accurate.
What surprised me most was how differently people use it depending on their situation. Parents with two or more kids leaned on it more than people without children at home, which tracks. When you're managing more than one person's health on top of your own, a 2 a.m. answer probably feels a lot more necessary than optional.
Where I Draw the Line
Here's my honest concern, though, and it's a real one.
A 2024 study looking at ChatGPT's medication answers found it usually didn't cite sources at all, and when it did, every single sourced response contained at least one made-up citation. Not slightly off. Fabricated, cited with total confidence, like it was reading straight from a textbook that doesn't exist.
A 2025 study went further and found AI was completely accurate on medication questions only 19% of the time. That's not a rounding error. That's a coin flip you'd lose most of the time, dressed up in a tone that never once sounds unsure of itself.
What really got me was a doctor's account inside that same SingleCare survey. One physician described a patient who stopped taking metformin entirely after asking AI about side effects, based on a fear they hadn't actually experienced any symptoms of. They'd made a real medical decision off a hypothetical one, and only reversed it after an actual conversation with their doctor talked them through the actual risk.
To be fair, most people aren't reckless about this. 68% said they verify AI answers somehow, whether that's checking other sources or actually asking a healthcare provider afterward. But that still leaves nearly a third of people who don't verify anything at all. That's the part that actually worries me, because that's exactly the group most likely to act on something confidently wrong.
My Actual Rule for Using It
So here's where I've personally landed, for whatever it's worth. AI gets to answer the questions where being wrong costs me nothing. What's this medication generally for? Is this side effect common? What should I even bring up with my doctor about this? Low stakes, easily correctable, no harm done if it's slightly off.
It does not get to answer the questions where being wrong costs me something real. Should I change something? Should I stop taking this? Is this symptom serious? Those go to an actual human who knows my actual history, every single time, no exceptions, no matter how confident the answer on my screen sounds.
Final Thoughts...
It's not a perfect system, and I'll admit the line gets blurry sometimes, especially at midnight when a real appointment feels very far away, and the AI answer is sitting right there, sounding so sure of itself. But knowing that even the most confident-sounding AI answer might be built on a nonexistent source has made me much more careful about which questions I'm actually willing to hand over.
If you're doing the midnight-search thing too, I'm not judging. I do it myself. Just maybe don't let it be the last word before you make an actual change to something that matters. Remember to be responsible about your health and always seek professional advice before acting.
