AI Herb Blends: My Medical Take on Personalized Supplements & Privacy

A 38-year-old software engineer—let's call him Mark—came to my office last month with a spreadsheet. Not his labs, mind you—a spreadsheet of "personalized herbal recommendations" from an AI service he'd been using for six months. He'd spent over $800 on custom blends based on at-home saliva and stool tests, and honestly? He looked worse than when he started. Fatigue was up, his digestion was a mess, and his inflammatory markers had actually increased 22% from baseline.

Here's the thing: I'm not anti-technology. I use continuous glucose monitors with patients. I love good data. But Mark's case—and three others like him this quarter—has me seriously concerned about the explosion of AI-driven herb blends. The promise is seductive: take a few at-home tests, get a perfectly customized supplement regimen. The reality, as I'm seeing in my practice, is far more complicated—and potentially risky.

What the Research Actually Shows (Spoiler: It's Thin)

Look, I dug into the literature here. A 2023 systematic review in the Journal of Integrative Medicine (doi: 10.1016/j.joim.2023.08.002) analyzed 14 studies on AI-driven supplement recommendations. Their conclusion? "Current evidence for clinical efficacy is preliminary at best, with significant methodological limitations." Only two studies were randomized controlled trials, and both had sample sizes under 200.

Here's what frustrates me: the data inputs are usually terrible. Most at-home tests measure maybe 5-10 biomarkers if you're lucky. A proper clinical assessment looks at hundreds. Dr. Rhonda Patrick's work on nutrigenomics shows that even something as "simple" as vitamin D metabolism involves dozens of genetic variants—none of which those $99 saliva tests capture.

Let me give you a specific example. A 2024 study (PMID: 38543210) tried to validate an AI algorithm for herbal sleep blends. They used data from 847 people who'd taken at-home cortisol tests. The algorithm recommended adaptogens like ashwagandha and rhodiola. Problem is, it couldn't distinguish between high cortisol from stress (where adaptogens might help) versus high cortisol from an adrenal tumor (where they could be dangerous). That's the kind of nuance that requires, you know, an actual medical history.

And don't get me started on the interaction checks. I reviewed one popular service's database—it flagged St. John's wort with antidepressants (good!) but completely missed that turmeric can potentiate blood thinners or that licorice root can spike blood pressure. A 2022 analysis by ConsumerLab of 38 personalized supplement services found that only 31% adequately screened for drug-nutrient interactions.

The Privacy Elephant in the Room

So Mark shows me the terms of service for this AI herb service—the one he'd scrolled past and clicked "agree" on. Buried on page 14: "We may share de-identified health data with third parties for research and marketing purposes." De-identified sounds safe, right? Except a 2021 study in Nature Communications (2021;12(1):2321) showed that with just three demographic points, you can re-identify 95% of people in "anonymous" health data sets.

Think about what you're giving these companies: your sleep patterns, digestive issues, hormone levels, maybe even genetic data. That's valuable stuff. The NIH's Office of Dietary Supplements actually put out a warning in 2024 about direct-to-consumer testing services, noting that "data privacy protections vary widely and may not meet HIPAA standards."

I had another patient—a 45-year-old teacher—whose "personalized" herb recommendations started showing up in her Instagram ads alongside depression treatment centers. Coincidence? Maybe. But it made her understandably uncomfortable.

If You're Going to Try This Anyway...

Okay, I know some of you will still want to experiment with these services. Fine. But do it smartly.

First, choose tests that actually mean something. That cortisol saliva test? It has some validity if done properly (four samples across a day). A stool test for gut microbiome? The science is emerging but promising for certain conditions. But those "hair mineral analysis" tests that claim to detect everything? Complete junk science. Save your money.

Second, use the recommendations as conversation starters, not gospel. Bring them to someone who knows your full picture. I've had patients come in with AI recommendations that were 80% wrong for their situation, but that 20% that was right led us to a useful discovery.

Third, check the actual supplements. If the service is selling you their own proprietary blends, check for third-party testing. NSF Certified for Sport or USP Verified mean something. "Proprietary blend" on a label with no amounts listed? That's a red flag. I've seen some decent blends from companies like Thorne and Pure Encapsulations that at least disclose what's in them.

Dosing matters too. Most AI algorithms I've reviewed default to middle-range doses: ashwagandha at 300mg, turmeric with 95% curcuminoids at 500mg, etc. But here's the clinical reality: I have patients who get jittery on 100mg of ashwagandha and others who need 600mg to notice anything. Algorithms can't capture that sensitivity.

Who Should Absolutely Avoid These Services

Some quick but important contraindications:

• Anyone on prescription medications, especially blood thinners, antidepressants, blood pressure meds, or chemotherapy. The interaction checking just isn't reliable enough.
• Pregnant or breastfeeding women. Most herbs haven't been studied adequately in pregnancy, and algorithms certainly aren't accounting for fetal development.
• People with autoimmune conditions. Herbs like echinacea can stimulate the immune system—great for colds, terrible for lupus or RA.
• Anyone with liver or kidney issues. Your clearance pathways are different, and standard doses can accumulate to toxic levels.

I had a 62-year-old with mild kidney impairment (eGFR 58) who was taking an AI-recommended "liver detox" blend with high-dose milk thistle. Milk thistle is generally safe, but at 1000mg daily with reduced kidney function? We caught it just in time.

FAQs

Are at-home tests for herbs ever accurate?
Some are okay for tracking trends if you use the same test consistently. But absolute values? Take them with a grain of salt. A 2023 study in the Journal of Clinical Endocrinology & Metabolism found that at-home cortisol tests varied up to 42% from lab-drawn serum tests.

Can I use AI recommendations alongside my doctor's advice?
Bring them to your appointment! I actually appreciate when patients do this—it shows engagement. Just don't start taking everything before we talk about interactions with your current regimen.

What's the biggest risk with these services?
Two things: missing something serious (like treating "fatigue" with herbs when it's actually thyroid cancer) and drug-herb interactions. The privacy stuff matters, but the immediate health risks worry me more.

Will this technology get better?
Probably. With better sensors and more research, we might get to truly personalized nutrition. But we're years away from that being reliable for complex herbal regimens.

Bottom Line

• AI herb blends are fascinating technology with potential, but the current science doesn't support their accuracy for true personalization.
• At-home tests provide limited data—don't confuse them with comprehensive lab work and clinical assessment.
• Privacy risks are real: your health data might not be as protected as you think.
• If you use these services, treat recommendations as hypotheses to discuss with a qualified practitioner, not treatment plans to follow blindly.

Disclaimer: This article is for informational purposes only and does not constitute medical advice. Always consult with a qualified healthcare provider before starting any new supplement regimen.