How a Family Health Manager Turns Rare Doctor Visits into Everyday Kindness

The Unlikely Hero: A Health App for the Whole Family

When Ant Group renamed its health AI from AQ to Ant Afu in late 2025, the company boasted over 15 million monthly active users. A month later, that number jumped to 30 million, with daily health questions topping 10 million. By early 2026, they claimed 100 million total users. But third-party data told a different story: only about 29 million monthly actives, each using the app 17.8 times a month, for an average of 13.2 minutes per session.

Those numbers can be spun two ways. Optimists see an explosion in health AI. Skeptics see a classic case of download-and-ditch, fueled by red envelopes and promotions. But both narratives miss something deeper. The real question isn't how many people open the app—it's whether a single health consultation can grow into a long-term, trusting relationship.

After digging through team interviews, product updates, user feedback, and public data, I think Ant Afu's most valuable unit isn't the individual patient. It's the "family health manager"—the person who shoulders health responsibilities for parents, kids, spouse, and self. This role turns scattered, low-frequency medical needs into a continuous family duty. The product's success hinges not on daily chat counts, but on whether it can safely move someone to a clear next step, and whether that next step gets better because of the data it already has.

Beyond the Chatbot: The Four-Layer Architecture

On the surface, Ant Afu looks like a medical chatbot. You describe symptoms, upload a lab report, or snap a photo of a pill bottle. The AI asks follow-up questions, then offers explanations and advice. Most reviews focus on accuracy: Is it as good as DeepSeek or Doubao?

But Ant Afu isn't satisfied with being a Q&A tool. Since its launch, the product has evolved through four layers:

• Information layer: Understands text, voice, dialects, and images—reports, prescriptions, medicine boxes.
• Understanding layer: Uses proactive questioning, health records, and family member profiles to fill gaps and explain risks.
• Action layer: Pushes toward real doctor consultations, appointment booking, insurance payments, or virtual companionship.
• Continuity layer: Stores reports, device data, diet, goals, and outcomes for future reference.

General-purpose AI can match a single medical Q&A, but it lacks the stable family context and the ability to hand off to real-world services. The strategic map is impressive, but it doesn't mean users complete their tasks. If most conversations end with a text answer, the product remains just a medical dictionary.

The Real User: The Family Health Manager

In a 2025 interview, Zhang Junjie, head of Ant's health division, revealed a surprise: early adopters weren't young tech-savvy users, but 40-to-60-year-olds. They were dealing with their own health issues—abnormal checkups, sleep problems, chronic conditions—while also managing their parents' and children's health.

For a healthy young person, serious medical care is rare. An annual physical, the occasional cold—that doesn't justify a standalone app. But health tasks aren't evenly distributed across a family. Usually, one person takes on the role of information gatherer, risk assessor, appointment maker, companion, and medication reminder. They may not be the patient, but they're the organizer.

This "family health manager" handles at least four kinds of tasks: monitoring chronic conditions, interpreting test results, coordinating care for multiple generations, and making decisions about when to seek help. Each event alone is low-frequency, but together they create a steady stream of needs.

That's why Ant Afu added large fonts, dialect support, voice input, and photo recognition. For a young person comfortable with prompts, a blank chat box works. But for someone trying to ask about their father's symptoms without knowing medical jargon, lowering the barrier to entry is everything.

However, not all 40-to-60-year-olds manage a whole family. Some focus on their own chronic issues. Others might be young adults tracking weight loss or sleep. So "family health manager" is a hypothesis to test, not a label for everyone. Ant should verify: How many accounts have two or more family members? How often are those profiles reused? What's the difference in long-term engagement between multi-member and single-member accounts?

The Core Journey: Reaching a Safe Next State

Health users don't just want knowledge. They want to know, "What should I do now?" That could mean: keep observing at home, get more information, stop taking a certain medication, book a routine appointment, consult a real doctor soon, or go to the ER immediately. A good health AI shouldn't push transactions or pretend to diagnose. It should guide people to a clear, safe next step.

Ant Afu's ideal journey has six steps:

1. Report interpretation as the first value moment.
2. Proactive questioning to gather context.
3. Risk-stratified advice with specific triggers and urgency.
4. Connection to real services—appointments, doctors, insurance.
5. Follow-up and data capture.
6. Learning for the next interaction.

Report interpretation is a perfect entry point because it lowers three costs: understanding medical jargon, filtering abnormal results, and deciding what to do next. If the user confirms the interpretation and saves it to their profile, the product gains its first piece of long-term context.

But accurate text recognition isn't the same as correct medical interpretation. Reference ranges vary by lab, and the same value means different things for different ages, sexes, and medical histories. The AI might read every word correctly and still give misleading advice.

Proactive questioning is better than a blank chat box for most people, but it can create a false sense of completeness. When the interface shows a progress bar, users may think the system has collected everything it needs. Medical completeness isn't about filling a form; it depends on the specific condition and risk. The product should clearly state what it knows, what it's missing, and how confident it is.

"See a doctor" isn't a failure—vague referrals are. A good recommendation includes why, when, which department, what red flags to watch for, and what to prepare. "Consult promptly" is just a disclaimer.

Does the Feature Add Value? Five Checks

Ant Afu is packed with features: AI diagnosis, report reading, skin photo analysis, specialist avatars, health records, goals, reminders, device integration, appointments, consultations, insurance, and virtual companionship. Listing them in homepage order doesn't make a product. Each feature must pass five tests:

• Does it reduce real cost?
• Does it improve decision-making?
• Does it move the user to the next step?
• How does it handle failure?
• What new risks or responsibilities does it create?

Memory Must Be Accurate

Ant Afu aims to personalize advice using health records and conversation memory. The more you use it, the better it knows you—and the harder it is to switch. That's a logical moat, but it's also fragile. Two detailed App Store reviews complained about lost history and inconsistent metric descriptions. Those aren't proof of widespread amnesia, but they highlight a critical risk: in single Q&A, limited context is tolerable; in long-term management, a wrong name, date, or value can cascade into harmful advice.

Medical memory shouldn't be about quietly storing everything. It needs five attributes: clear subject (who does this belong to?), timestamp (is this current?), source (report, device, self-report, or AI inference?), error correction (can users edit mistakes?), and control (can users choose what's used, delete, or export?). When memory informs advice, viewing it isn't just a privacy setting—it's a core feature.

The Smart Scale: An Activation Experiment

In June 2026, Ant Afu launched a "Lose 100 Million Jin" weight-loss campaign, offering cheap smart scales, daily check-ins, and AI suggestions. A 21-day challenge followed. Then Ant invested in Boohee, a food database, to power an "AI photo diet" feature that estimates calories. This isn't just a gimmick. The scale forces a key activation loop: download app → bind device → get first measurement → receive AI explanation → set goal → remeasure.

Weight is ideal for this because it changes visibly, has a short feedback cycle, and is easy to understand. But 21 days of check-ins don't equal a habit. The real test is whether users still measure at 30, 60, and 90 days after the incentive ends, and whether weight changes come from healthy behavior, not water loss.

Commercial Neutrality Must Be Explained

Ant Afu has said health Q&A results are free of ads and commercial influence. Yet in June 2026, it launched a health insurance agent and partnered with Pacific Health Insurance. These aren't contradictory—the Q&A layer stays pure, while services are separate. But it raises a governance question: when the same product understands a user's anxiety and can recommend doctors, drugs, and insurance, how do you separate content judgment from commercial conversion?

Users need to know what's health advice, what's a product, why it's recommended, whether the platform earns money, and if there's a non-commercial alternative. The more detailed your health data, the more precise the targeting—and the stronger the trust firewall must be.

Sticky Isn't in the Chat Box

When discussing retention, people often lump together different mechanisms: events (like red packets), family aggregation, data continuity, behavioral habits, and service closure. Each works differently. Events drive downloads, family creates recurring needs, data adds switching costs, behavior builds habit, and services close the loop. They can reinforce each other or fail independently.

A more sensible metric structure for Ant Afu would track: activation (first meaningful interaction), retention (return for a second health event), and outcome (did the user reach a safe next step?). Different tasks need different windows. Weight management can be weekly; chronic disease is monthly or quarterly; a physical exam may only recur yearly. Using daily active users for all health tasks forces meaningless reminders.

Health products shouldn't aim to trap users in the app. If someone's problem is solved and they don't open it for a while, that's success. High frequency driven by anxiety is a red flag. So the north star should be: the number of family health tasks that reach a clear, safe next state within a reasonable time—whether that's booking an appointment or correctly deciding to watch and wait.

Scale Isn't Proof: Five Things We Still Don't Know

Ant Afu has proven it can attract users and answer questions. But the public evidence doesn't yet show:

• Reliable long-term memory: No data on record completeness, reuse rates, contradiction rates, or error correction.
• Clear responsibility for AI vs. human: Users must know whether they're talking to an AI, a doctor-trained avatar, a human doctor, or a formal telemedicine service.
• Real service fulfillment: Hospital counts don't guarantee booking success, wait times, or whether AI info reaches the doctor.
• Commercial trust: How will insurance and drug recommendations be disclosed to avoid eroding trust?
• Actual health outcomes: We see usage stats, not goal attainment, adherence, or reduced anxiety.

Ant Afu has built a path from Q&A to family health management. It hasn't yet proven the path leads to better health. That's okay—it's early. But the difference between a tool and a relationship is whether the next interaction is better because of the last one.

Kindness as Product Design

At its core, Ant Afu is an exercise in kindness. It's about meeting people where they are—a worried parent, a middle-aged adult juggling generations—and gently guiding them to the right next step. Kindness means remembering what matters, not just what's convenient. It means knowing when to say, "You're okay to watch this," and when to say, "Please go to the ER now." It means being transparent about what the AI knows and doesn't, and never pretending to be more than it is.

The real test isn't how many people download the app. It's whether, a year from now, a family health manager can say, "Ant Afu helped me catch my dad's blood pressure before it became a stroke." That's the kind of kindness that scales.