OpenAI's Healthcare Public Data plugin includes three medicines-related sources, and treating them as three routes to the same underlying medicines knowledge misses the actual design: each answers a genuinely different question, and understanding the distinction is essential before trusting any medicines answer these connectors help generate.
The three questions, separated
What is the product, and what does its US label say? DailyMed's territory: official US medication labelling, active and inactive ingredients, dosing and warnings, packaging and identifiers, product-specific information drawn from the regulated labelling record. How should the medicine be represented consistently across systems? RxNorm's territory: normalised medicine terminology, connecting ingredients, strengths, dose forms and branded products for interoperability purposes, genuinely useful for EHR systems and medicine-list reconciliation needing a shared identifier standard, and explicitly not a source that itself determines appropriate prescribing. What public safety or regulatory information has been reported? openFDA's territory: recalls, regulatory information and adverse-event reporting datasets, useful for signal exploration across a population, and explicitly not a source that establishes causation or event frequency for any individual case.
DailyMed, in detail
Official US medication labelling covering active and inactive ingredients, dosing guidance, warnings, packaging and identifiers, and product-specific information tied to the exact authorised product rather than the active ingredient generically. This is the connector closest to what a UK clinician would recognise as SmPC-equivalent content, and it describes US-authorised products specifically, the jurisdiction gap this cluster's dedicated DailyMed-versus-emc analysis treats in full.
RxNorm, in detail
A normalised terminology system connecting ingredients, strengths, dose forms and branded products, built to let different systems refer to the same medicine consistently, the interoperability layer that makes medicine-list reconciliation across different records and systems technically possible. OpenAI's own implementation guidance explicitly warns against using RxNorm to infer dose or interchangeability, a warning worth taking at face value: RxNorm standardises what a medicine is called and how its variants relate to each other, it does not encode clinical judgement about appropriate dosing or whether two products are safely interchangeable for a given patient.
openFDA, in detail
Recalls, regulatory information and adverse-event reporting datasets, a genuinely useful resource for exploring safety signals across a population and tracking regulatory actions. OpenAI's own guidance explicitly warns against treating openFDA adverse-event reports as causal incidence data, a warning that reflects a structural property of spontaneous adverse-event reporting systems generally: a report captures that an event was associated with a medicine in someone's clinical judgement, not that the medicine caused it, and the reporting rate reflects reporting behaviour as much as true event frequency.
Why treating these as interchangeable is the specific error to avoid
A clinician or an AI system blending DailyMed's authorised-label content, RxNorm's terminology matching and openFDA's safety-signal data into one undifferentiated "medicines information" answer risks exactly the confusion each source's own scope was built to prevent: an interchangeability inference RxNorm was never designed to support, a causal safety claim openFDA's reporting structure cannot establish, or a treatment-pathway claim DailyMed's product-specific labelling does not address. None of this is a criticism of any individual source, each performs its actual function well, it is a caution about the synthesis layer sitting on top of all three, which needs to preserve rather than blur these boundaries.
The UK parallel
DailyMed's closest UK equivalent is the relevant SmPC on emc, covered in full in this cluster's dedicated comparison. openFDA's closest UK equivalent is MHRA Drug Safety Updates and Yellow Card reporting, running on the UK's own regulatory timeline and structure. And RxNorm's closest UK equivalent is dm+d, the UK's own medicine identification and terminology standard, built around UK-authorised products rather than the US catalogue RxNorm standardises. None of these three US sources substitutes for its UK counterpart on any question where UK licensing, UK safety communications or UK medicine identification specifically matter.
Frequently asked questions
Which of the three is most directly useful for a UK clinician working with ChatGPT for Clinicians?
None directly, since all three describe US-specific products, terminology or regulatory data; a UK clinician using this connector stack should treat any medicines answer it produces as US-context information requiring UK-specific verification before it informs a UK prescribing decision.
Does RxNorm's interoperability function have any UK relevance at all?
The underlying need it serves, consistent medicine identification across systems, is genuinely universal; the specific standard itself is US-built, with dm+d serving the equivalent function for UK systems and UK-authorised products.
Why does OpenAI issue explicit warnings for two of these three sources but not DailyMed?
Plausibly because RxNorm's terminology-matching function and openFDA's adverse-event reporting are the two sources whose outputs most tempt an inference the underlying data does not actually support, dose or interchangeability from a terminology match, causation from an association report, while DailyMed's labelling content is more directly what it appears to be, though the same general verification discipline should still apply to any medicines answer regardless of source.