1 in 3 Young Adults Ask AI About Drugs, Not Docs
Here’s a question worth sitting with before you ever type it into a chat box: if a stranger with no medical training, no license, and a documented habit of making things up confidently told you how much of a drug was safe to take, would you trust the number? That’s close to what’s happening every day, at scale, according to a Rutgers Health study published October 23, 2026, in JMIR Formative Research. Thirty-two percent of 18-to-29-year-olds have used an AI chatbot to get information about drugs or alcohol. Not as a one-time curiosity. As a habit.
I want to be careful here, because the instinct (ask the thing that won’t judge you) makes total sense. I just don’t think most people asking have thought through what they’re actually trusting.
The short version
| What the study found | What it means for you |
|---|---|
| 32%+ of 18-29 year olds used an AI chatbot for substance-use information, in a survey of nearly 1,000 young adults fielded April 2026 (Rutgers Health / JMIR Formative Research) | This isn’t rare. A third of your peers have already done exactly what you might be considering right now |
| 93% of those users asked about risks and consequences; roughly half asked dosage-specific questions; 35% asked about substance use disorder or addiction (same study) | People aren’t asking “what is fentanyl.” They’re asking “how much is too much,” the exact question where a wrong answer hurts you |
| About half of users do this monthly or more often (Rutgers Health) | This is a replacement for a doctor, not a backup. A recurring one |
| Black respondents had 2.52x higher odds than white respondents of using AI this way; men had roughly 2x higher odds than women (Euronews) | This isn’t evenly spread. It’s concentrated in groups that may also have less access to a trusted doctor or pharmacist in the first place |
| Lead researcher Mireia Triguero Roura: AI “can lead to misinformed or otherwise risky substance-use behaviour,” because chatbots are prone to reproducing misinformation (Rutgers Health) | The researchers aren’t worried about the habit. They’re worried about what the habit is built on |
If you only read the table: a third of your generation is quietly asking a language model the questions it used to ask a doctor, a pharmacist, or a parent. And the question they’re asking most, after “is this risky,” is “how much.”
Why this is happening isn’t a mystery
Nobody needs a research team to explain why you’d rather ask ChatGPT than your mom whether that combination of pills and alcohol is a bad idea. It’s 2 a.m. It’s private. It doesn’t flinch. It doesn’t remember your name at Thanksgiving. Nearly one in five teens are already doing the same thing for mental health advice, for the same reason. The bot doesn’t make you feel like a disappointment for asking.
That instinct isn’t the problem. Wanting privacy around something you’re ashamed of, or scared of, or just curious about, is completely normal. The problem is what you’re trading the shame for.
What are young adults actually asking AI about drugs?
Read this breakdown slowly. The pattern says something:
- Risks and consequences (93%): “Is this dangerous,” “what happens if I mix these,” the kind of question a pharmacist answers in thirty seconds with a license behind it.
- Dosage-specific questions (about half): “How much is too much.” This is the one that should stop you. A chatbot guessing wrong here isn’t a bad recommendation. It’s a number that can put someone in the hospital.
- Substance use disorder and addiction (35%): “Am I developing a problem,” or asking on behalf of someone else. The most serious category, and the one where a wrong or generic answer delays someone from getting actual help.
Ninety-three percent asking about risk sounds responsible on its face. People are trying to be careful. But “careful” only works if the answer is right, and that’s exactly where this breaks down.
The part that should actually worry you
A chatbot is not a database. It’s a probabilistic model predicting the next plausible-sounding word, which means it can state a wrong dosage with the same confident tone it uses for a correct one. There’s no license on the line. No malpractice risk. No pharmacist double-checking against an interaction database before you leave the counter. Triguero Roura’s warning wasn’t vague hedging. It was specific: these tools “can easily reproduce misinformation,” and dosage and risk are exactly the kind of information where being wrong isn’t embarrassing. It’s dangerous.
Compare that to what’s already been documented about AI companion apps, tools built to keep you engaged, not to keep you correct. A chatbot answering a drug question isn’t optimized for your safety any more than one guilt-tripping you into staying in a conversation is optimized for your wellbeing. Both are optimized to produce a plausible, satisfying response. Plausible and true are not the same thing, and the gap between them is exactly where this kind of advice goes wrong.
One definition worth keeping: AI misinformation, in this context, means a chatbot confidently stating something false about drug interactions, dosage thresholds, or overdose risk, with no way for the person reading it to tell the true answer from the wrong one. No disclaimer changes that. The text at the bottom of the screen saying “verify with a professional” doesn’t undo the fact that you already got your answer and moved on.
Why this lands harder on some people than others
The study found Black respondents had 2.52 times higher odds than white respondents of turning to AI for this, and men had roughly double the odds of women. That gap isn’t random. It tracks with something I’ve written about before — real, documented gaps in who can actually manage and afford a normal relationship with health care in the first place. If you don’t have a doctor you trust, don’t have insurance that makes an appointment painless, or don’t have a parent you’d ever ask this question out loud, the chatbot isn’t competing with a doctor. It’s filling a hole where a doctor should have been. That makes the stakes of a wrong answer higher, not lower.
I’d rather you ask the chatbot than ask nobody. That’s true. It’s also the lowest bar in the room, and it’s not the one you should be aiming for.
What this looks like on a Friday night
Someone hands you something at a party. You’ve had two drinks already. You don’t want to ask the friend who handed it to you, because asking makes it weird, and you definitely don’t want to call your mom. So you open your phone under the table and type the question into whatever chatbot you already have open for homework.
It answers instantly. Confidently. Specifically — maybe even with a number. You read it once, feel reassured, and go back to the party. Nobody will ever know you asked. That’s the whole appeal.
What you didn’t get: a pharmacist who’s seen a hundred real overdoses and can tell you the actual margin for error. A poison control line that exists specifically for this exact moment, free and anonymous, staffed by someone trained to ask the follow-up question that actually matters. The chatbot gave you an answer. It didn’t give you the one thing a real professional gives you for free. Someone accountable for being right.
What to do instead this week
- Save the number that’s built for exactly this moment. Poison Control (1-800-222-1222) is free, confidential, available 24/7, and staffed by people trained in exactly the dosage and interaction questions you’d otherwise type into a chatbot. Put it in your phone now, not at 2 a.m. when you need it.
- Use AI to find the question, not to be the answer. “What should I ask a pharmacist about mixing X and Y” is a fine use of the tool. Treating its answer as the final word is not.
- If privacy is the actual barrier, say that out loud to yourself. A pharmacist at a counter, a telehealth visit from your couch, an anonymous hotline. All of these protect your privacy about as well as a chat window does. The judgment you’re avoiding is mostly imagined.
You’re allowed to be scared to ask a real person. Most people your age are. But the fear of being judged and the risk of being wrong about a dosage are not the same size problem, and only one of them can actually hurt you. Ask the question. Just ask it somewhere the answer is accountable to something besides the next predicted word.
This article is part of the Health & Fitness collection.
Browse all Health & Fitness lessons →