5HT #117: the bull case for OpenEvidence, the Oura of air, and 3 more

The standard take on vertical AI companies is that they’ll eventually get streamrolled by OpenAI and Anthropic, that they’re thin wrappers waiting for foundation models to catch up.

Well, OpenEvidence is everyone’s favorite vertical AI co to pile on in healthcare right now. OpenAI launched ChatGPT for Clinicians in April, aimed straight at them. Anthropic just planted its flag with a major Optum deal. In Venrock’s recent healthcare report, only 14% picked the standalone/IPO outcome, meaning 86% don’t think OpenEvidence survives on its own.

Some smart money disagrees. OpenEvidence raised $700M in under a year from Sequoia, GV, Nvidia, Thrive, and the Mayo Clinic, ran its valuation from $1B to $12B in twelve months, and just reportedly fielded offers at $20B it can’t even be bothered to take—all while revenue doubled to $300Mish annualized in seven months.

Plus, I disagree. 😆 Maybe this is just me smoking my “builder with a unique growth-first perspective” hopium. Or maybe it’s because CEO Daniel Nadler and I overlapped in the same house in college, and I’ve been a long-time fan. Either way, I’m a huge believer. Here’s my case.

Quick catch-up for context: OpenEvidence is a free (!) AI medical search and clinical decision-support platform. An NPI number gates access—meaning every user is a real, licensed clinician—and the business largely runs on pharma advertising. 

So doctors just, like, sign up (no hospital IT department, no procurement cycle, no 18-month enterprise sale). And boy have they. In April alone, OpenEvidence handled 27 million (!) clinical consultations and was used by roughly two-thirds of American doctors (!!). As Nadler put it, “We did the hardest thing in the history of American health care. We got the majority of American doctors to all voluntarily adopt a single technology platform.”

The bear case deserves to be heard, of course. 5HT+ community member Azella P. says her research team evaluated OpenEvidence and found ChatGPT gave reasonably similar answers that are "not worth paying extra for." Her read: clinicians mostly use it for fast retrieval. A doctor mid-shift needs an answer in 15-20 seconds, and a quick synthesis of existing papers does the job. 

Buuuut that's exactly the use case frontier models are best at, right? So every model release “shrinks the delta.” Meanwhile, in theory a health system/any tech founder could theoretically take an open medical model, bolt a RAG pipeline over PubMed on top, and rebuild the core text features in months for a fraction of the cost. In her view, the real moat is the business model—the licensing deals—not the AI. And publisher licensing deals obviously aren’t exclusive (kind of defeats the point of licensing).

Also, critics say it can flub rare conditions and edge cases, its performance hasn't really been peer-reviewed (though they’re working on it), it's pharma-funded (👀), and EHRs like Epic are bundling copilots directly into the screens doctors already live in… while OpenEvidence makes them switch tabs. Azella's prediction: it gets acquired, because it doesn't last alone.

My bull case starts with the thing this gets backwards. “Frontier models improving” isn't a threat to OpenEvidence. It's more like a subsidy! Vertical apps sit on top of foundation models, so when the base gets smarter, OpenEvidence gets smarter too. 

To me, the real fight is everything stacked on top: the licensed corpus—full-text deals with NEJM, JAMA, and 400+ Wiley journals including the Cochrane database—plus the specialty-tuned weights, the nuanced interpretation of evidence, and the deep understanding of what a clinician actually needs at 2am in the ED. Nadler calls the AI layer "search glue," a beautifully unsexy way of saying the model doesn't need to make things up. And what space does that matter more than in health?

To me, EvidenceGrade is the sharpest proof. Launched this month, it grades the evidence behind every answer using the same framework behind WHO and Cochrane guidelines. Why does that matter? Well, because the real failure mode of AI in medicine isn't just hallucination—it's flattening. A landmark RCT and a 40-person observational study in a completely different population get smushed into the same confident paragraph. The difference between "probably" and "we have three convincing trials." Plus, it’s exactly what its users need.

Could OpenAI build this? Technically, sure. Will they? It requires practicing physicians on staff, publisher relationships, and product decisions optimized for millions of clinicians instead of billions of consumers. 

A general model will never lose sleep over study design, BUT a doctor can get sued over it. That gap—the features a generalist refuses to care about—is the entire vertical AI thesis.

Plus, the prize for winning doctors is enormous. There's a whole hall of fame of generational businesses built on little more than physician attention, including Doximity (used by 85%+ of U.S. physicians, ~$645M in annual revenue, publicly traded at $4Bish) and M3 (a Japanese giant that reaches 90%+ of Japan's physicians, worth $7B+ even now). And these are pre-AI-era companies, so they’re more like a floor than a ceiling.

And no, I don't think pharma funding is inherently misaligned. A relevant drug ad shown to the right specialist is closer to the Instagram ad you actually clicked (and the stuff you happily bought) than to spam. The economics tell you how valuable this attention is: pharma pays $70 to $1,000+ CPMs to reach verified physicians—versus $5-15 on social—because one prescriber's attention moves millions in downstream decisions. Attention + trust + relevance is just one of the best business models on the planet.

What’s next? To me, their obvious next strategic focus should be to make the OpenEvidence brand THE trust signal of medicine:

✅ Get into every EHR that doesn't want to build this itself. They just deployed across NewYork-Presbyterian, Columbia, and Weill Cornell. Now run the table.
✅ Make it dead simple for any health app to embed OpenEvidence instead of building their own “clinical copilot,” maybe even absorb clinical liability in exchange for placement (!).
✅ Crazier idea: Become the doctor's identity layer. "Sign in with OpenEvidence"—your NPI-verified credentials and settings, portable across every platform. Think Google/Meta login, but for medicine.

Bottom line: When regulated stakes, a proprietary corpus, and verifiable professional identity stack up, super-serving one customer isn't a wrapper. It's the whole moat. So the questions for every consumer health builder are: What does the general model refuse to care about for your customer? And can you make caring about it a daily habit?

Ah, peptides, everyone’s favorite topic these days. Venrock’s recent report called them “the supplement aisle with better PR,” which honestly is just 🧑‍🍳💋. (My take on them is here.)

Anyway, whatever you think of them, they’re about to have a bigggg week. Starting today, the FDA is convening a two-day meeting with outside experts to recommend whether compounding pharmacies should be allowed to continue making certain peptides. As of this afternoon, BPC-157 has already officially been voted "in."

The panel will review evidence for several other popular peptides 💁‍♀️, helping shape what access looks like going forward.

Worth saying there has been criticism about who’s on the panel, pointing to financial ties some doctors and pharmacists have to peptides. Nonetheless, I expect that some will get the okay (like supplements, I broadly do think buyers should be able to decide for themselves). And my hope is this increases the funding and pressure for better research. Peptides are clearlyyy not going anywhere, and I want more concrete evidence on which, if any, work.

Didya know the fastest-growing occupation in America isn't an AI engineer, but a nurse practitioner? Last week, Corner Health raised $32.5M to turn as many of them as possible into small business owners, and I think it’s brilliant.

Fun fact: In most states (27 of 50), NPs have full practice authority (diagnose, prescribe, run their own clinic, no MD required). In a country with an accelerating lack of primary care physicians (projected shortfall of 70K by 2038!) and 78% of NPs wanting to start a business, this is a huge win-win for everybody.

Also, the care holds up: a randomized trial in JAMA found no difference in patient outcomes between NP- and physician-led primary care. More ownership for NPs, more primary care for everyone. 👍

One of the most common questions we get from Healthyish Content clients is: How long does SEO/GEO take to work?

The honest answer: Longer than most other marketing channels, but probably less than you think. This isn’t a “hack” because authority takes consistent publishing, credible backlinks, and trustworthy content, especially in a YMYL (Your Money or Your Life) category like health.

The good news is in the current SEO/GEO landscape: I’ve never seen rankings move faster 👍, traffic compound sooner 👍, and conversions rise earlier 👍. Why? Because the rise of AI slop has quietly lowered the bar for quality. 

That’s one of many reasons there’s never been a better time to invest in best-answer-style content.

Timelines vary depending on your domain authority (recommend doing a quick, free check using popular tools Ahrefs or Semrush), but for sites in the 25–50 range that already have a baseline of trust, 3–6 months is where things get exciting, 6–12 months is where the curve starts bending upward, and 12+ months is the leap.

For sites below 25 or above 70, we break down those timelines here, too—along with why GEO moves veryyy differently from traditional SEO’s long, steady climb.

Bottom line: While AI is changing the game, there’s never been a better opportunity to play.

We’ve quantified sleep, glucose, HRV, and VO2 max—basically every input except the one you use 20,000–30,000 times a day: air.

My sense is there’s rising anxiety around air quality, but not much guidance on what to do about it at home. Sure, you can buy an air purifier—we have this Coway one in literally every high-traffic room—but that doesn’t give you much signal.

Enter the Aranet4 and Airthings, two (of many) monitors in a space without IMO best-in-class, consumer-friendly solutions. 👀

I put both in my office about six weeks ago, and so far, I’m into it. 👍

When CO₂ climbs past 1,500 ppm, I get an alert, and I open a window. (For context, outdoor air is typically around 425–430 ppm. Some research links elevated indoor levels with worse cognitive performance, though the evidence is mixed.) What a gloriously simple productivity hack.

The monitors also track more than CO2. Airthings measures temperature, radon, and particulate matter. Cooking makes particulate levels spike. A smoky day turns the display red—which feels especially relevant right now, as friends in Michigan are barely leaving their home because of smoke from the Canadian wildfires. 🙁

The builder takeaway: Air quality anxiety is rising faster than the tools.

If you’re looking for a business idea, I’d recommend building an air quality monitor for the nursery. We’re about to have a third kid, and I’d pay just about anything to know if there’s something in the air that could affect my baby—and I know I’m not alone. Talk about super-serving a customer!

But the biggest opportunity in the space is a wearable that connects air quality to how you feel and integrates with the rest of your health data. A friend of mine is actually building one like this, and I’m pumped about it.

We’re in the Fitbit era of air quality, but I’d bet someone will build the Oura of air next.

Most clicked last week: My favorite buckwheat pillow 🥹. If you get one, hit reply and let me know what you think!

👋 Who are you again? I’m Derek Flanzraich, and I’m in the business of healthyish. I founded two venture-backed startups–Greatist (👍) and Ness (👎)—and helped dozens of brands you know grow including Ro, GoodRx, Midi, BetterHelp, and LMNT. I now run Healthyish Content, plus a few other healthyish things (like Fixie Dust, launching this fall).

Every Thursday, I share what’s hot, what’s not, and what’s next in consumer health. (Disclosure: I invest in and advise companies in this space—when I cover one, I'll say so.) I sometimes add affiliate links, but they influence my picks zero.

See something smart, strange, or seriously overhyped? Hit reply, I respond to every email.