Heidi's impact report makes a strong case for AI-assisted documentation — demonstrating reduced documentation burden, improved clinician experience, and measurable capacity release across multiple NHS settings. It also reinforces a point that pharmacists know well: medicines details are safety-critical, and the quality of medication documentation directly affects patient safety.
Why AI Scribes Can Improve Documentation Quality
The baseline is not good. Heidi's report notes that documentation quality can be poor in critical areas such as drug allergy documentation. This is consistent with wider evidence: allergy recording in NHS records is frequently incomplete, inconsistent, or inaccurate — creating ongoing prescribing risk for every clinician who subsequently treats the patient.
AI scribes can improve this by generating structured, consistent documentation from the consultation — capturing medication discussions, allergy conversations, and prescribing decisions in a standardised format. When the AI generates an allergy field from a conversation where the patient clearly states "I'm allergic to penicillin — I came out in a rash," that structured capture may be more complete and more consistently formatted than the rushed note a GP types between patients.
Why Medicines Are a Special Risk Category
Medicines documentation is different from other clinical documentation because errors have immediate, direct, and potentially dangerous consequences.
Drug names. Heidi's report acknowledges that speech-recognition models may struggle with complex drug names. "Amlodipine" and "amitriptyline" sound similar in fast clinical speech. "Metformin" and "methotrexate" could be confused in a noisy environment. A wrong drug name in the record is not a documentation error — it is a prescribing safety hazard that could persist through every future prescription.
Allergies. An incorrectly recorded allergy may block appropriate prescribing (false positive) or fail to generate a warning when a contraindicated drug is prescribed (false negative). Both are clinically dangerous. Allergy documentation must distinguish between true allergy (anaphylaxis, angioedema, Type I hypersensitivity) and intolerance (GI side effects, mild rash) — because the clinical implications and prescribing restrictions are fundamentally different.
Doses and frequencies. "Take two twice a day" and "take one twice a day" differ by a factor of two. The AI must accurately capture the dose, frequency, duration, and route as discussed — and the clinician must verify before the prescription is generated.
Monitoring. If the consultation included a discussion about monitoring requirements ("we'll check your LFTs in two weeks"), the documentation should reflect this as a committed action — not merely a conversational mention. The difference matters when the monitoring is not done and the patient experiences a preventable adverse event.
Interactions and contraindications. The AI scribe captures the consultation as it occurred. It does not independently verify whether the prescribed medication interacts with the patient's existing drugs, is contraindicated in their clinical context, or requires dose adjustment for renal or hepatic impairment. That verification is a separate clinical knowledge task — not a documentation task.
Heidi's Human-in-the-Loop Model
Heidi's report emphasises that its technology is a human-in-the-loop system: the AI generates a draft, and the clinician reviews, edits, and approves before the note is published to the medical record. The report explicitly notes that LLM summarisation can hallucinate, making clinician review essential.
This is not a flaw — it is a safety feature. The human review step is the point at which medicines errors can be caught: wrong drug names corrected, allergy records verified, doses checked, monitoring commitments confirmed. The review is where clinical expertise adds value that AI cannot provide autonomously — contextual judgment about what is correct, what is safe, and what is clinically appropriate for this specific patient.
Why Pharmacists Are Central to the Next Phase
Pharmacists bring medicines expertise that GPs, nurses, and other clinicians may not have in the same depth. Medication review, interaction checking, formulary alignment, monitoring protocols, deprescribing, counselling, and safe switching are core pharmacist competencies.
As AI scribes become embedded in primary care workflows, the pharmacist role in reviewing medicines-related documentation becomes more — not less — important. A practice pharmacist who reviews AI-generated consultation notes for medicines accuracy, allergy completeness, and monitoring appropriateness adds a safety layer that the AI itself cannot provide.
How iatroX Supports Medicine-Related Clinical Questioning
iatroX complements the documentation workflow by giving clinicians and pharmacists a place to ask guideline-grounded questions, check clinical reasoning, and practise exam-style medicines scenarios.
For prescribing verification: "What does NICE recommend as first-line for this condition?" For medicines safety: "What monitoring is required for this medication?" For interaction checking context: "Is this combination clinically significant?" For exam preparation: GPhC CRA Q-banks and clinical calculations.
AI can help write the note. Pharmacist-grade review helps make sure the medicines story inside the note is safe.