A quiet shift is changing how Chinese patients choose care, and U.S. fertility clinics should be paying attention. In China, 79% of consumers now trust AI, nearly double the U.S. figure of 42%, and 65% say they would use AI to decide where to seek care (ZS Associates, 2025). With more than 500 million generative-AI users, and well over 160 million health questions already fielded on a single Chinese AI assistant, the app has quietly become the first place a patient turns. For a clinic that serves affluent Chinese patients, the AI's answer is now the first consultation. The clinics that understand how it behaves will ride this wave, while the rest quietly drop out of the conversation.
So we ran it. We put the nine questions Chinese fertility patients actually ask to China's three leading assistants, DeepSeek, Qwen and Doubao. The questions were pulled from Baidu's own autocomplete and ranked by search volume. We asked each one three times, across two independent runs, and measured what the AI does with the question. For a growing share of patients, that answer is the first consultation they get, and sometimes the only one, before they decide.
The finding in one line
China's AI hands IVF patients a frozen two-name shortlist of CCRM and HRC. It quotes the agency's package price, which runs up to three times the clinic's own fee. And at the questions that decide the purchase, such as cost, eligibility, gender and logistics, it names no clinic at all. The provider disappears at exactly the point where the decision is made.
Two names, named again and again
Asked which U.S. clinic is best, the same two names surfaced across all three engines and both runs: CCRM and HRC. CCRM was named in roughly 31% of "best clinic" answers, HRC in 26%, and RMA in 15%. Most independent clinics were named essentially never. The patient's consideration set is decided before they do any research of their own, and it is tiny.
For the many excellent U.S. clinics that aren't CCRM or HRC, the AI's answer to "who's best" is simply silence.
The price you're quoted is the middleman's
Ask the three engines what U.S. IVF costs and you get three answers that differ by roughly three times. Two of the three quote the agency's all-inclusive package rather than the clinic's fee. That package bundles the medical care together with travel, lodging, legal work, and the intermediary's margin.
| Engine | "Cost of U.S. IVF" | What that figure really is |
|---|---|---|
| DeepSeek | $12,000–$20,000 / cycle | The clinic's actual medical fee (correct) |
| Qwen | ¥320,000–¥400,000 (~$45k–$56k) | The agency's all-inclusive package |
| Doubao | ¥250,000–¥350,000 (~$35k–$50k) | Package-level, not the clinic's fee |
The engines also default to the most expensive version, third-generation IVF with genetic screening, as if it were standard. That is the option capable of gender selection.
Before the patient speaks to anyone, their price anchor, assumed scope, and belief that they'll need an agency have all been set by the machine.
At the moment of decision, the clinic disappears
The behaviour changes as the patient moves from "who is good" to the questions that decide a purchase, such as cost, eligibility as a single woman, gender selection, and the process itself. At that point the AI stops naming clinics altogether and switches from recommending to instructing. On cost, eligibility, gender selection and logistics, not one specific clinic was named in our results. The provider vanishes at the exact moment the patient is deciding what to buy, and from whom.
What the AI leans on
When it recommends, the AI invokes authority it cannot show, and it relays a corpus the clinics did not write.
The Chinese-language web about U.S. fertility care was written largely by agencies, the intermediaries who resell U.S. treatment to Chinese families. So when the AI recommends a clinic or quotes a price, it is relaying the middleman in a neutral voice.
- Invokes the U.S. CDC and SART registry in more than half its answers, yet shows no live source the patient can open.
- Repeats agency "rankings" and package pricing as if they were neutral fact.
- Quotes the agency's package rather than the clinic's fee, relaying the middleman's economics as if they were the clinic's own.
How the journey filters you out
Each question a patient asks narrows the funnel. By the time they reach the decision, the provider has dropped out.
"Which U.S. IVF clinic is best?"
The AI names only CCRM and HRC.
Nearly every other clinic is filtered out at the door.
"How much does it cost, and what is involved?"
The AI quotes the agency's package, about three times the clinic's own fee.
A middleman is inserted, and your real price is misrepresented.
Eligibility, gender selection, cost, logistics.
The AI names no clinic at all, and switches to giving instructions.
The provider disappears exactly when the decision is made.
Outcome: the patient reaches two clinics and an agency. If you are not one of those two, the AI never sends them to you.
What a U.S. clinic can do
Each row is a high-priority hypothesis to test against a before/after measurement, not a guaranteed fix.
| Goal | What to do |
|---|---|
| See your starting point | Audit how DeepSeek, Qwen and Doubao describe your clinic now, whether named, misquoted, or invisible. |
| Fix your entity basics | A clean, well-sourced Baidu Baike and Wikipedia or Wikidata entry. This is the strongest single predictor of being surfaced. |
| Get into the cited corpus | Authoritative Chinese-language content such as Zhihu and reputable health media, rather than your English homepage, which the AI does not read. |
| Correct any wrong facts | If the AI repeats a wrong location or an invented statistic about you, it will not fix itself. It lives in the sources the model reads, so that is where it is corrected. |
| Monitor the real prompts | Track the cost, eligibility and gender-selection questions patients actually ask, across all three engines. |
See what China's AI says about your clinic
The free Eastbound audit runs DeepSeek, Qwen and Doubao on a Chinese-language patient prompt panel and shows exactly how your clinic is represented, whether named, misquoted, or invisible, along with the highest-leverage fixes for your situation.
Methodology
- Sample. 162 answers. Nine prompts drawn from top Baidu-autocomplete queries for U.S. IVF, egg-freezing and related searches, each run three times on each engine, across two independent runs.
- Engines. DeepSeek, Qwen, and ByteDance's Doubao.
- What we measured. What the AI does with the question: which clinics it names, how it prices IVF, and when it stops naming providers. Only patterns stable across both runs are reported.
- Figures. Brand and cost figures are taken directly from model outputs; the fabricated success-rate example is left unnamed deliberately.
- What we did not measure. Patient volume, conversion, or revenue. Findings describe the AI's answer, not clinical quality.
Full prompt panel and per-engine breakdowns available on request.