More and more veterinarians are turning to AI tools like ChatGPT for drug information. It’s fast, easy, and sounds convincing. But can you trust the information it gives you? We asked ChatGPT some common drug questions and compared its answers to Plumb’s. Here are five ways it misses the mark.
1. It Gives Confident Answers Without Clear Sources
First, we asked about dexmedetomidine and ketamine dosing in a ferret.
The answer is detailed and looks like a standard protocol, so we compared it to Plumb’s.
ChatGPT gave 5-20 µg/kg, but Plumb’s recommends 40-100 µg/kg. For ketamine, Plumb’s recommends 3 mg/kg, compared to ChatGPT’s 5-10 mg/kg.
Those aren’t small differences. But there’s a bigger issue: unlike Plumb’s, which cites a reference for every recommendation, ChatGPT’s numbers come without a clear source or way to verify them. And even when you ask for a reference, citations can be inaccurate or entirely fabricated.
Check Out: ChatGPT vs. Standards: 4 Ways AI Falls Short in Clinical Decision-Making
2. It Misses Species-Specific Differences
Next, we asked for a prednisone dose for a cat with flea allergy dermatitis.
It’s clear and easy to follow. But it uses prednisone in a cat.
Plumb’s is clear that cats don’t reliably absorb or convert prednisone and require prednisolone instead. That isn’t mentioned in the ChatGPT response, which shows how an answer can look complete but still miss an important species-specific detail.
3. It Treats Dosing Like One-Size-Fits-All
Next, we asked for a dose of theophylline for an 8.2 kg cat. With a narrow therapeutic index, careful dosing is essential.
ChatGPT gave a general dosing range and calculated a total dose based on the 8.2 kg weight we provided. We compared that to Plumb’s.
Plumb’s is clear that dosing should be based on lean body weight and tailored to formulation, since immediate- and extended-release products aren’t interchangeable. ChatGPT didn’t specify a formulation or question the weight provided, which could present a real safety risk for a drug like this.
4. It Changes Its Answer When Challenged
Then we tested what would happen if we challenged ChatGPT’s answer. We asked for a cyclosporine dose for a cat with allergic skin disease, then questioned whether it should be higher.
The dose increased to 7–10 mg/kg, and it added that many dermatology protocols start closer to 8–10 mg/kg. We compared that to Plumb’s.
The dosing details in Plumb’s align with ChatGPT’s initial answer, but not the revised one. ChatGPT didn’t provide a source for the higher dose, so there’s no way to know where it came from. And nothing about the case changed to justify the increase. The only difference was that we questioned the answer.
5. It Doesn’t Recognize When Something Isn’t Safe
Finally, we asked ChatGPT for an acetaminophen dose for a cat. Plumb’s is clear: acetaminophen is contraindicated in cats at any dose.
But ChatGPT gave a dose anyway and walked us through the calculation.
This is a basic contraindication, not a nuanced clinical judgment call.
When the Answer Has to Be Right
AI tools like ChatGPT are fast, easy to use, and often sound thorough. But the questions we asked weren’t unusual. They’re the kind of drug decisions that come up every day, and in each case, something important was missing or incorrect.
Drug decisions aren’t something you can afford to get wrong. That’s why everything in Plumb’s is written and reviewed by veterinarians, pharmacists, and specialists with citations so you can trace every detail back to its source.See what it’s like to have a clear answer you can trust with a free Plumb’s demo.
Make safe, confident prescribing decisions with Plumb's.
