The characteristic AI failure in this territory is compression: "supported by evidence", "recommended by a guideline" and "licensed for this use" are three different states with three different professional consequences, and fluent synthesis collapses them into one confident proposition. Non-medical prescribers need the states separated as working knowledge, because off-label and unlicensed prescribing carry specific responsibilities, and an answer that hides which state you are in has hidden the responsibilities too.
The three states, defined and distinguished
Licensed use: within the product's marketing authorisation as the exact SmPC describes it, indication, population, route, the regulatory default where the manufacturer's evidence and obligations stand behind the use. Off-label use: a licensed product used outside its authorisation, different indication, age group, dose framing or route, which can be entirely appropriate and often guideline-supported, paediatric practice being the everyday example, while shifting more responsibility onto the prescriber for justifying the choice. Unlicensed use: a product without a UK marketing authorisation for any use, imports and specials among them, the furthest point from the regulatory default and the state with the strongest expectations around justification, sourcing and follow-up. The states interlock with the guideline layer in exactly the combinations the source-matrix article maps, /blog/emc-vs-nice-cks-sps-mhra-research-prescribing-sources: licensed but not recommended, recommended but off-label, research-supported but neither, and locally restricted despite national support, four configurations a complete answer names rather than smooths.
The eight things a complete answer must state
A prescribing-support answer touching this territory is complete only when it carries: licensing status, explicitly, for the exact product, since licensing lives at product level, /blog/product-specific-prescribing-active-ingredient-not-enough; the supporting guideline, named and dated, where recommendation is doing the work; evidence strength, because guideline-supported and evidence-settled are not synonyms; the availability of licensed alternatives, which off-label justification must engage with; monitoring implications of the use as proposed; consent and patient-information consequences, because patients should understand when a use sits outside the authorisation and the PIL will not describe it; documentation expectations, the recorded rationale that professional guidance expects for off-label decisions; and whether specialist initiation, shared care or advice is the appropriate route. An answer stating all eight has equipped a decision; an answer asserting a treatment while omitting them has generated homework and disguised it as a conclusion, and the honest design standard for evidence tools, ours included, is a visible licensing-status label beside the recommendation with the exact SmPC linked, so the state is read, not inferred.
The documentation and consent behaviour
The professional half, compressed to its working shape. Record the clinical reasoning: why this use, why now, why not the licensed alternative, with the sources, guideline and SmPC, cited; the record is the justification's home and the future reviewer's answer. Have the conversation the state requires: for off-label use, that the medicine is being used outside its licence, why that is recommended here, and what to expect, in proportion to the decision's weight, with the discussion documented; for unlicensed products, the fuller version. Plan the follow-up as if the responsibility were yours, because it is: monitoring, review and the criteria for stopping, per the seven-section framework at /blog/ai-medication-monitoring-plans-anp. And escalate by design where the territory warrants it, paediatric complexity, unfamiliar unlicensed products, contested evidence, using specialist and medicines-information routes as the system intends. None of this makes off-label prescribing exceptional; it makes it what it already is, ordinary practice done with the state named and the paperwork honest, which is exactly what a good AI answer should have made easier from the first sentence.
Frequently asked questions
Does guideline support make off-label use routine?
It makes it defensible and common; the documentation, consent and product-level checks remain, because national recommendation changes the justification's strength, not the use's regulatory state.
How do specials and altered formulations fit?
As unlicensed territory with pharmaceutical questions attached, stability, equivalence, sourcing, where pharmacist involvement is not optional garnish but the competent route.
Can supplementary prescribers work off-label?
Within a clinical management plan that explicitly provides for it, yes; the plan is the additional governing document, and silence in the plan is a boundary, not an ambiguity.
How common is off-label prescribing in ordinary practice?
Common enough to be unremarkable in several fields, paediatrics above all, which is exactly why the disciplines matter: routine frequency without routine documentation is how defensible practice becomes indefensible paperwork.
Should AI tools flag licensing status automatically?
They should, and the absence of a flag is never evidence of licensed status: silence in a synthesis has no regulatory meaning, and the SmPC check remains the prescriber's, whatever the interface implies.
What is the fastest way to confirm a use is off-label?
The exact SmPC's indications and population sections, two minutes against the product actually intended, which is also the check that catches the subtler case, licensed ingredient, unlicensed formulation for this route or age.