A clinical pharmacist at a Malaysian state hospital is preparing notes for the next drug and therapeutics committee meeting. Before opening the guidelines, she asks ChatGPT to compare the available options in a therapy class, including your molecule. The answer describes a rival's presentation in detail, lists an outdated strength for yours and cites an overseas drug information site. This is the problem with hospital pharmacy formulary AI answers: the first draft of a committee discussion can be shaped by whichever company left the clearest facts on the open web.
This article is for hospital account directors, key account managers, medical affairs and market access leaders at pharmaceutical companies serving Malaysian public, university and private hospitals. It explains why AI tools describe molecules the way they do, what pharmacists actually ask, how to audit your portfolio this week, and what you can publish within Malaysia's rules to be described accurately.
Why do AI answers favour rival molecules?
AI tools answer from what they learned in training and from pages they find and read live. Neither source includes your detailing conversations, CRM notes or the dossier you submitted for formulary listing. They describe molecules using whatever accurate, quotable text exists publicly, and that is often a rival's or a third party's.
Training data and live retrieval
A model's training data is fixed at a point in time and weighted towards content published widely, often from larger markets. That is how outdated strengths, discontinued presentations or another country's indications end up in an answer about Malaysia. Live retrieval is the second route. OpenAI's crawler documentation says OAI-SearchBot is used to surface websites in ChatGPT's search features, and Google's guidance on AI features asks site owners to make important content available in textual form on indexable pages.
Formulary decisions happen in private; AI reads in public
The Ministry of Health Medicines Formulary, often called the Blue Book, is about selecting medicines. CodeBlue reports that companies submit dossiers to the Pharmaceutical Services Programme, which assesses clinical and economic value through a formal health technology assessment process. A paper in Value in Health Regional Issues describes the Formulary Management Branch assessing safety, efficacy, effectiveness and budget impact, with outcomes going to the MOH Medicines List Review Panel.
None of that work is visible to an AI tool. The same CodeBlue report notes that being in the Blue Book is distinct from purchasing, which depends on each facility's needs. So at hospital level, pharmacists still compare options, and some will use AI to start.
Pharma companies under-publish for good reasons
Pharma teams are cautious about open web content, and rightly so. Medicine advertising in Malaysia is governed by the Medicines (Advertisement and Sale) Act 1956, and The Star reported that PhAMA's code for prescription products prohibits promotion before a product is approved. The result is that many companies publish almost nothing about their hospital portfolio in plain text. Generic manufacturers, distributors, overseas drug information sites and forums fill the space, and AI tools quote them.
What hospital pharmacists in Malaysia actually ask AI
Hospital pharmacists use AI to save time on comparison, presentation and supply questions. Most prompts are in English, the working language of Malaysian hospital pharmacy, with occasional Bahasa Malaysia when preparing notes for wider teams. These patterns are worth testing, grouped by the stage of a formulary or procurement discussion.
Preparing a class review
- "Compare the available biosimilars of [molecule] registered in Malaysia"
- "What are the differences between [brand A] and [brand B] for hospital use: presentations, storage, preparation?"
- "Summarise antimicrobial stewardship considerations for [antibiotic] in a Malaysian hospital"
Checking practical details
- "Is [molecule] available as a prefilled syringe in Malaysia?"
- "What is the stability of [product] after reconstitution?"
- "Ubat alternatif untuk [molecule] jika berlaku kekurangan bekalan di hospital"
Before the committee meeting
- "Which companies supply [molecule] to hospitals in Malaysia?"
- "Has [product] had any supply disruption in Malaysia this year?"
Notice how factual these questions are. Pharmacists are rarely asking which brand is best. They want presentation, stability, preparation, registration and supply facts, which are exactly the facts a compliant company page can state plainly.
How to check where you stand
A useful audit takes a week and needs a pharmacist or medical affairs reviewer to score accuracy. The aim is a repeatable baseline across your hospital portfolio. Our measurement approach explains why repeated sampling matters more than any single answer.
Step 1: Pick your molecules and rivals
Choose five to ten priority hospital products. For each, list the originator or biosimilar rivals and the generics pharmacists compare you with.
Step 2: Write a fixed prompt set
Write 20 to 40 prompts based on the patterns above, using generic names, brand names and the way pharmacists phrase questions. Keep the wording fixed so you can compare month to month.
Step 3: Run across assistants, several times
Run each prompt three times in ChatGPT, Gemini, Perplexity and Google's AI features. Record whether your product is mentioned, how it is described and which sources are cited. Answers vary between runs and users.
Step 4: Score accuracy with medical affairs
Have a pharmacist or medical affairs colleague mark each mention as correct, outdated, wrong or off-label. Off-label or wrong-strength descriptions are the priority, because they create safety and compliance risk as well as commercial loss.
Step 5: Map the cited sources
List every URL cited and group them: your site, rival sites, overseas drug information sites, distributors, news, forums. This shows which sources AI tools trust for your class in Malaysia and where your own information is missing.
How to fix it
The fix is to publish accurate, on-label, non-promotional facts in plain text, agreed with medical, regulatory and legal teams before anything goes live. That does not guarantee AI mentions, but it gives AI tools a correct source to quote instead of an outdated one.
Agree the rules with regulatory and medical affairs first
Start with governance, not content. Agree which page types are factual corporate information, which are disease education and which would count as promotion under the Act, Medicine Advertisements Board requirements and the PhAMA code. Write it down. Every later decision is faster once this is settled.
Publish a factual portfolio page per molecule
For each hospital product, publish a short page with the generic name, brand name, registration status in Malaysia, available strengths and presentations, and the marketing authorisation holder. Keep it neutral: no comparative claims, no efficacy language beyond what the approved labelling states. This is the page that corrects "outdated strength" errors.
Put presentation, storage and supply facts in HTML
Pharmacists ask about storage, preparation and pack details constantly. Where your regulatory team agrees it is appropriate, state these facts in HTML text rather than inside a scanned PDF. Add a supply notice page and update it whenever availability changes. A dated, accurate shortage notice is one of the most useful things a company can publish, and AI tools have little else to rely on.
Support stewardship and education without promotion
For antibiotics and other stewardship-sensitive classes, publish disease and stewardship education that links to recognised guidelines and summarises their relevance in neutral language. Keep product mentions out of education pages. Pages that read like education, not advertising, are easier to defend and easier for AI tools to quote.
Make the company entity clear and consistent
Use one spelling of your company name, Malaysian entity name and product names across your site, distributor listings, press releases and association pages. If a local distributor supplies hospitals on your behalf, say so on both sites. AI tools often confuse the originator, the licence holder and the distributor.
Earn the sources AI already reads
| Source type | Why AI trusts it | What to do |
|---|---|---|
| Official registers and MOH publications | Authoritative and specific | Check your product and company names match exactly |
| Local health news (for example CodeBlue) | Independent, dated, Malaysia-specific | Offer medical affairs spokespeople on supply, access and evidence topics |
| Professional society and conference pages | Credible to clinicians | Publish talk summaries in HTML where your codes allow |
| Distributor and wholesaler sites | Often cited for availability | Give partners accurate product and supply text to publish |
What to measure
Measure monthly on the same prompt set: mention rate, accuracy, off-label or outdated statements, and the sources cited. Report ranges across repeated runs, because a single answer is only one sample.
- Mention rate: how often your product appears when its class is discussed.
- Accuracy rate: the share of mentions with correct strengths, presentations and registration status.
- Risk flags: any off-label, outdated or safety-relevant errors, escalated to medical affairs within a week.
- Source mix: which sources are cited for your class, and whether your own pages appear.
- Change after updates: re-run prompts after a new page, supply notice or press story goes live.
Look for direction over three months rather than week-to-week swings. Share the results with medical affairs, not just marketing, so errors are corrected quickly.
Common mistakes
- Treating AI answers as a marketing problem and leaving medical and regulatory teams out of the fix.
- Assuming formulary listing or strong detailing relationships will show up in AI answers.
- Publishing promotional or comparative claims on open pages to "win" AI answers, which creates regulatory risk.
- Keeping presentation, storage and supply facts only in scanned PDFs and leave-behinds.
- Letting distributor and partner sites carry outdated product details under your brand name.
- Judging the situation from one screenshot instead of repeated runs on a fixed prompt set.
How Shubi helps
Shubi is an AI visibility company based in Malaysia. For pharmaceutical teams, we run fixed prompt sets across ChatGPT, Gemini, Perplexity and Google's AI features to show where your molecules are missing or described wrongly, and which sources the answers rely on. We then work with your medical and regulatory reviewers on the open, factual pages and third-party sources behind those answers. Start with AI Visibility Intelligence or read about our content work. Message Shubi on WhatsApp