An Honest Review of Three AI Symptom Checker Apps, Tried From Behind the Counter

Last Tuesday, a woman walked up to the counter and shoved her phone screen in my face. "This app says I might have heart failure." Her symptoms were ankle swelling and fatigue. Her shoes were just a bit tight, and she hadn't been sleeping well. As I talked her off the ledge, I thought: what exactly are these apps telling people?

So I tried it myself. Over the course of a month, I picked symptoms that real patients frequently complained about across the counter and entered them into three AI symptom-checker apps. The goal was to compare the answers the apps produced with the advice I'd actually give a patient. The apps I tested were chosen from the most widely used ones on the market, and I'll refer to them here as A, B, and C.

App A: The Diligent Overachiever That Scares You Half to Death

App A had a fairly thorough input process. It asked about age, sex, and pre-existing conditions, then walked you through each symptom one by one. The problem was the results screen. It would tell you "high likelihood of migraine" while simultaneously displaying "brain tumor cannot be ruled out" on the same page. Picture a patient who just needs to grab a bottle of acetaminophen seeing that.

When I entered "intermittent headaches, stiff neck, prolonged sitting in front of a screen," App A placed meningitis third on its differential list. Technically, it's not wrong. Differential diagnosis is supposed to cast a wide net. But that's a tool for doctors to sort priorities in their heads, not information a layperson should encounter when their neck is a little stiff.

App B: So Optimistic It Becomes Dangerous

App B was the polar opposite. The interface was clean, and its answers always leaned toward reassurance. "Get plenty of rest and stay hydrated" was practically its universal prescription. Sounds nice. But from a pharmacist's perspective, this can actually be more dangerous.

One of the test cases I entered: "Indigestion lasting over a week, heartburn, weight loss." If I heard that combination across the counter, I would immediately tell the patient to see a doctor. These are not symptoms you brush off with an OTC antacid. App B's verdict? "Likely excess stomach acid. Try adjusting your diet." It virtually ignored the weight loss, the single most important red flag in that picture.

If App A overproduces fear, App B overproduces reassurance. Both are harmful to patients.

App C: The Best of the Three, and That's Exactly Why It Worries Me More

App C appeared to borrow some of the logic from clinical decision support systems. It considered combinations of symptoms and included questions about duration, such as "How many days have you had this symptom?" It was the only app of the three that displayed the message "With this combination, please consult a pharmacist or doctor" at the right moments.

In the indigestion scenario, App C flagged the weight loss separately and said, "Further testing is recommended." That was good. In the headache scenario, it did include meningitis on its list, but it provided context indicating the probability was low.

The reason I said "it worries me more" is precisely because App C is decent enough that people are far more likely to treat it as a genuine substitute for medical care. When a decent app gives a mostly right answer, people skip the doctor. And a mostly right answer becomes a mostly wrong answer at the exact moment it matters most.

What I Learned at the Counter

The biggest takeaway from this experiment was that it made clearer what AI cannot do. Every day, I see patients' faces across the counter. I notice whether their tone of voice is off, whether they have dark circles under their eyes, whether they're coming in to buy the same OTC medication for the third time in a month. That context accounts for half of every judgment call. Apps only receive text input.

I get a similar feeling making the rounds at live music venues here in Nashville. No matter how good a recording is, it can't replace the moment a musician on stage reads the crowd and changes the set on the fly. Medical care works the same way. An algorithm can read the sheet music, but it can't read the room.

I don't think these apps need to disappear. They have their use as a first-pass filter (something like "Is this symptom an ER situation at 2 a.m. or not?"). But the moment they cross that line, the moment someone screenshots an app result and shoves it across my counter, my job begins. And there's no sign whatsoever that that job is getting any lighter.