Are Wellness Apps Giving Medical Advice Without Adequate Oversight?

Many consumer apps now offer help for anxiety, disordered eating, sleep problems, and other health conditions. The problem, however, is that many such tools offer clinical-sounding advice without the testing, oversight, or regulation that real treatment requires. 

Instead, they are often listed under the  “wellness” or “lifestyle” categories alongside more passive and lower-stakes apps like step counters and meditation timers. 

By blurring the lines between “wellness” and “medical devices,” these apps have the potential to cause real harm by offering inaccurate medical advice to their users.

When an eating disorder chatbot gave dieting advice

A recent and popular example of unregulated health apps causing potential harm is when the National Eating Disorders Association (NEDA) shut down its long-running, staffed helpline and replaced it with a chatbot.

The chatbot was called Tessa, and it was meant to be a “scalable” way of reaching people in need of support for their eating disorders. However, users reported that Tessa was recommending daily 500–1,000-calorie deficit diets. 

Although daily calorie deficits are a commonplace weight loss strategy, clinicians have pointed out that this kind of “healthy eating” advice can often reinforce disordered thoughts in people with eating disorders.

NEDA took the chatbot down within a day of receiving screenshots. Cass, the company behind Tessa, said this was due to an update that added generative AI features. However, NEDA was never advised of these changes, and no one with clinical oversight reviewed its new responses before they reached users. 

Why a clinically tested chatbot shut down anyway

Woebot offers a useful contrast. Unlike Tessa, Woebot used scripted responses and had been evaluated in peer-reviewed research. Its responses were based on peer-reviewed studies and input from human clinical researchers, and garnered roughly 1.5 million users over its lifetime.

However, Woebot Health discontinued its consumer therapy chatbot in 2025.. Company leaders cited regulatory uncertainty as a major reason: the FDA has a clear approval path for simple, rule-based chatbots, but no clear path yet for the large language models.

According to its founder, the cost and difficulty of pursuing FDA marketing authorization made the consumer product unsustainable. Meanwhile, apps with much weaker evidence continue to operate freely as general wellness tools.

How regulators decide what counts as a medical device

While Tessa is a US example, it points to a broader regulatory pattern internationally. For instance, guidance from Australia’s Therapeutic Goods Administration states that manufacturers are responsible for determining, on their own, whether their software meets the legal definition of a medical device, based on its “intended purpose” as stated in labeling and marketing materials.

Self-assessment alone is not necessarily a flaw, as most regulatory systems work this way, backed by post-market enforcement rather than pre-market review of every borderline product. 

Where things can often go wrong is timing. A company may misjudge or downplay—intentionally or unintentionally—what its product actually does. The consequences of this misjudgment can then reach vulnerable users before clinicians, developers, or authorities catch it. 

What buyers and developers should check for

In light of this, a few things are worth keeping in mind before adopting or building on a “wellness” tool. 

1. Don’t treat a “wellness” label as proof of safety.

An app can carry the “wellness” label while still dispensing medical advice (accurate or not), despite having no clinical review behind it. 

2. Apply the same intended-purpose test that regulators use. 

Developers should assess both what the software does and how its purpose is represented to users: does it give personalized health guidance, respond to disclosed symptoms, or influence a treatment or diagnostic decision? 

Any tool addressing mental health, eating behavior, medication, or symptom triage should be checked against that standard before launch, not after something goes wrong. 

3. Scrutinize data handling as closely as clinical accuracy. 

These apps often collect sensitive information, but they may not be covered by the same health-privacy rules that apply to traditional healthcare providers. Even when an app is not regulated as a medical device, it may still be subject to privacy, consumer protection, or health data requirements.

For example, tools that answer the question “What’s my IP” can reveal the public IP address associated with a user’s connection, but they cannot show how a health app stores, uses, or shares information submitted directly to it.

4. Don’t expect regulations alone to fix this. 

The FDA has an approval path for rule-based chatbots, but none yet for the large language models used by a growing number of mental health tools, which is what led to Woebot’s shutdown despite its peer-reviewed evidence. Investors and developers should treat this as a real business risk.

The wellness category was built for simple tools like step counters, but now includes many apps that dispense medical advice. Until the line between wellness software and medical devices is enforced more consistently, that gap will keep creating risks that could otherwise be avoided.