A paediatric specialist clinic in Malaysia came to us with a familiar problem. Their old WordPress site was beautiful in design, slow in production, and invisible to the AI engines their younger patients had started using to find healthcare providers.
The site failed Core Web Vitals on every key page. New patients were bouncing on mobile before the hero image had finished loading. And when a parent asked ChatGPT or Perplexity for a specialist in their area, the clinic did not get a mention.
The brief
The clinic gave us three goals.
- Make the site genuinely fast for real patients on real Malaysian networks.
- Get the brand visible inside AI search answers for high-intent local queries.
- Keep the YMYL trust signals (credentials, certifications, testimonials) front and centre, health buyers verify before they book.
The challenge
YMYL (Your Money or Your Life) sites have a higher bar than most. Google’s quality guidelines weight expertise, authority, and trustworthiness signals more aggressively for health content. AI engines do the same. A fast site alone is not enough. A trusted slow site is also not enough. You need both.
The legacy site had three blockers:
- A 4MB hero image served from a non-CDN origin, killing LCP on 4G connections.
- No structured data to help search engines understand who, where, and what services were offered.
- Trust signals scattered across pages instead of consolidated near booking moments.
What we did
1. Rebuilt the site, mobile-first
We migrated from the page-builder stack to a static-rendered architecture with edge caching. Every image was compressed to WebP under 200KB, dimensioned, and lazy-loaded except the hero. Webfonts were trimmed from six weights to two with font-display: swap. Third-party scripts were deferred or interaction-loaded.
Result on the new build: Lighthouse mobile in the 95-99 range, and field LCP under 2.0 seconds at the 75th percentile.
2. Schema-first content structure
Every key page got the right schema markup: MedicalOrganization with credentials and area served, MedicalProcedure schemas for each treatment, FAQPage for common patient questions, and Person schemas for the practitioners with their qualifications.
We rewrote the FAQ content using buyer-actual phrasing, not “What is the treatment?” but “Is it safe for my child?”. The kind of question a worried parent actually types.
3. AI prompt mapping
We mapped 22 buyer prompts across patient stages, from broad symptom questions all the way through high-intent booking enquiries in the clinic’s local area. Each high-intent prompt got a dedicated page or section that answered it directly, in quotable language.
4. Trust stack consolidation
Practitioner credentials, association memberships, before/after photos, and patient testimonials were grouped into a single trust section that appears just before every booking CTA. Schema-tagged so AI engines can pull credentials when asked about practitioner authority.
Results
- Lighthouse mobile lifted from sub-50 to 95+ across all key pages.
- Field LCP at the 75th percentile dropped to under 2.0 seconds.
- First AI citation appeared on Perplexity at day 47. By day 90, the brand was being mentioned on 3 of 5 AI engines for high-intent local queries.
- Mobile bounce rate fell 38% over the first quarter.
What we learned
YMYL sites compound when speed, trust, and structured data are treated as one project, not three. The clinic could have spent the same budget on ads alone and seen short-term lifts. Spending it on the foundation now means every future ad rial dollar lands on a site that converts and a brand that AI engines already mention.
For health businesses in Malaysia: schema markup is no longer optional, and “we will do SEO later” is no longer an answer. Buyers ask AI first.