Fitbit Air gets the most important thing right about AI health: the safest coach is often the one that tells you to do less. That’s my thesis, and it matters because Google Health Coach sounds less like a magic wellness oracle and more like a cautious restraint system, at least in the testing described by Victoria Song at The Verge. In a category stuffed with overconfident “personalized” advice, that restraint is the product.
The $99 Fitbit Air is not interesting because it invents a new health sensor fantasy. It’s interesting because Google appears to understand that AI health advice should be boring before it becomes ambitious. Hydrate. Avoid heat. Skip strength training. Walk a little. Ask whether the calves feel strained. That’s not flashy. Good. Health AI should earn trust through caution, not performance theater.
Fitbit Air makes the right bet: cautious AI beats wellness bravado
The Verge’s review describes Google Fitbit Air as an extremely light, comfortable tracker with strong battery life, quick charging, and the usual Fitbit core: step count, resting heart rate, sleep, heart rate variability, blood oxygen, readiness, sleep stages, and cardio load. In roughly a month of testing, Song says she charged it only three times. During WWDC, a 45-minute charge took the device from 20 percent to 85 percent.
That hardware story is tidy. The software story is messier, and more important.
"The AI coach, used properly, can be useful"
That line from The Verge’s review card is the whole fight. Google Health Coach is not automatically useful. It becomes useful when the user feeds it context, corrects it, checks it, and treats it as a coaching layer rather than a doctor. That is a narrow but defensible lane.
The strongest counterargument is obvious: if an AI health coach needs hours of handholding, it’s not really smart. Song says she spent five to six hours telling the coach what she needed, detailing three-, six-, nine-, and 12-month goals, and explaining about 10 years of medical context. That’s a lot of unpaid setup labor for a product sold as personalized.
Still, Fitbit Air deserves credit for avoiding the worse mistake. It doesn’t appear to push heroic optimization when the data says the body is strained. It tells the user to pull back.
Google Health Coach connects sleep, HRV, heat, and workout readiness
The strongest example in The Verge’s testing is not a fancy insight. It’s a refusal. Song’s sleep was weak, readiness score unimpressive, heart rate variability below baseline, and environment hot and humid, with temperatures above 90 degrees Fahrenheit. The coach advised skipping planned strength workouts and focusing on hydration, staying out of the heat, and getting some steps.
That’s how consumer health AI should behave. A body doesn’t operate from one metric. A coach that sees bad sleep, low recovery, heat stress, and planned exertion should not pretend the answer is a motivational push notification.
Here’s the useful contrast:
| Health app behavior | Bad version | Better Fitbit Air version |
|---|---|---|
| Readiness | Shows a score and leaves the user guessing | Explains the day’s limits and suggests changes |
| AI coaching | Sounds confident with thin context | Asks follow-up questions and defers to clinicians |
| Workout advice | Pushes consistency regardless of state | Recommends lowering intensity when recovery looks poor |
| Personalization | Uses friendly wording as a substitute for evidence | Improves when given goals, records, and history |
The value is not that Gemini can chat. The value is that Fitbit Air can take existing signals and convert them into a decision at the exact moment a user might make a bad one. That’s the difference between data decoration and actual coaching.
Fitbit Air exposes the weakness of single-score wellness apps
Wearables love clean scores because clean scores sell clarity. Recovery, stress, sleep, exertion, and illness rarely behave that neatly. A readiness number can be useful, but only if the system explains what pushed it down and what the user should do next.
That’s where Google Health has a real opening. The Verge says the rebranded Google Health app puts the Gemini-powered chatbot front and center, gives morning summaries of sleep and readiness, answers questions about health, interprets trends, and can suggest fitness plan changes. Song used it to create a travel-friendly workout routine during business trips while dealing with medication side effects, with a less aggressive step goal and body weight strength movements.
The danger is that an AI layer can make fuzzy advice sound sharper than it is. The defense is transparency. If the advice is “skip strength training,” the user should be able to see whether heat, HRV, poor sleep, cardio load, or medical context drove that recommendation.
This is the same product discipline XOOMAR readers should demand from AI elsewhere: less theatrical output, more user control. For adjacent coverage on that broader AI product tension, see our pieces on AI Slop Forces Patreon into a Fight for Creator Power and Quiet Android 17 Pixel 9 Pro Test Rewires Daily Work.
The AI health dumpster fire started with overconfidence
Song’s broader critique lands because it matches the category’s worst habit: wellness AI often implies sophistication while hiding behind nonmedical disclaimers. It speaks personally, then retreats legally. That can be dangerous when users are tired, anxious, injured, or eager to believe an app understands them.
Fitbit Air’s better move is conversational humility. In The Verge’s example, the coach did not simply issue a verdict. It asked whether Song’s calves felt strained and how she felt about the assessment. Those questions matter. Good health AI should push users back toward their own bodies, not replace self-awareness with a polished automated answer.
The tool still stumbles. Song says that even after extensive setup, the coach reverted to older data during a later check-in and had to be reminded of prior conversations. Some parts of the app reflected her updated 5,000 daily step goal, while others still showed 10,000. Asking the AI to update it did not stick.
That is not a small flaw. A health coach that forgets context can become worse than dumb. It can become persuasive and stale at the same time.
Fitbit Air still asks for trust users can’t fully audit
The strongest case against Google’s approach is trust. Even cautious AI health coaching can cause trouble if users don’t understand how recommendations are generated. False reassurance is a risk. False alarms are a risk. So is the quiet pressure to treat a wearable’s advice as medical truth because it sounds calm and specific.
Google has added some guardrails, according to The Verge. The coach defers to healthcare professionals and does not give diagnoses. It provides sources for health facts, many of which Song found were clinical studies or reputable sources. Nearly 500,000 people beta-tested it since October 2025, and Google said it had over a million points of feedback before shipping an improved version last month.
Privacy is part of the trust test. Uploading medical records requires identity verification through CLEAR, and permissions must be renewed periodically. The Verge also notes that health data use for training Google’s AI is turned off by default and must be opted in to. Fitbit acquisition terms require health data to be stored separately and not used for Google’s targeted ad business.
Those are meaningful protections. They are not enough by themselves. Trust requires the user to understand the recommendation, not just accept the terms.
Google should make Fitbit Air’s coaching explain itself before it becomes normal
The next version of Fitbit Air’s health coaching should make its reasoning visible. Not vaguely. Plainly.
Users should be able to tap a recommendation and see:
- Primary driver: Poor sleep, low HRV, heat exposure, cardio load, medication context, or another factor.
- Confidence level: Whether the coach is making a strong call or a cautious suggestion.
- Data source: Wearable data, user-entered history, uploaded records, or general health guidance.
- Limits: What the coach cannot assess and when to contact a clinician.
- Control: Easy opt-outs from coaching, records upload, and AI training.
Google Health Coach will also reach beyond Fitbit Air. The Verge says Pixel Watches get it too, and Google hopes to expand it to third-party wearables. That makes explainability more urgent, not less. A small coaching quirk on one tracker becomes a bigger behavioral force when the same interface follows users across more devices.
What would prove my optimism wrong? If average users get generic advice unless they spend hours feeding the system medical context. If the app keeps forgetting goals. If Google turns conservative coaching into a subscription hook instead of a safety principle.
For now, Fitbit Air points in the right direction. The best AI health tracker won’t be the one that talks the most. It’ll be the one that knows when a quiet warning is enough.
The Bottom Line
- Fitbit Air suggests AI health tools may be safer when they encourage restraint rather than constant optimization.
- At $99, the device could make cautious AI coaching more accessible to mainstream wearable users.
- The review highlights that AI health advice still depends on user context, correction, and healthy skepticism.
Originally published on XOOMAR. For more news and analysis, visit XOOMAR.
