Lesson 1: What AI Can and Cannot Do in Pharmacy

Before you start

This course gives general information, not clinical advice. Your local policy, approved references (such as the BNF, NICE and NHS resources) and your professional judgement always come first. If anything here conflicts with your employer's policy, follow the policy.

In this first level you will:

1. Learn how a general AI assistant produces text.
2. Sort pharmacy tasks into "suitable for AI support" and "never for general AI".
3. Look at dated evidence of how pharmacy teams use AI now.

Task 1: Understand how an AI assistant writes

ChatGPT, Gemini, Claude and Microsoft 365 Copilot are built on large language models (LLMs). In plain terms:

1. The model was trained on a very large amount of text.
2. From that text it learned patterns: which words tend to follow which, in what style.
3. When you type a prompt, it predicts a likely next piece of text, then the next, and builds a reply one piece at a time.
4. It aims for text that sounds right. Being true is not built in.

An analogy: think of a trainee who has read thousands of patient leaflets and emails but has never opened the BNF. Their writing is fluent and well laid out. Ask them a dose and they will give you a confident number in the right format, and it may be wrong.

That is why AI output is fluent but not reliably accurate. OpenAI's own help pages say ChatGPT can give incorrect or misleading answers, can sound confident when it is wrong, and can make up quotes, studies and references. This is often called a hallucination. Models also have a training cut-off, so they may not know about recent changes unless they are using a search tool.

Do: Next time you read an AI answer, ask "How would I know if this were wrong?" If you cannot check it against an approved source, you cannot rely on it.

Task 2: Sort your tasks into two groups

A general AI assistant is not a clinical reference. Use it where a mistake is easy to spot and cheap to fix. Keep it away from anything where a mistake could harm a patient.

Suitable for AI support (you check every word):

• Admin: drafting emails, meeting notes, rota announcements, letters to suppliers
• Drafting: making an approved leaflet easier to read, or changing its tone or format
• Learning: explaining a topic for CPD revision, writing practice questions, summarising your own notes
• Workflow: turning a process into a checklist, tidying up a standard operating procedure draft so a colleague can review it

Never for general AI:

• Working out or checking doses
• Checking interactions or contraindications
• Deciding whether to supply, refer or treat
• Any other clinical decision about a particular patient

For these, use approved references, the clinical systems your workplace has approved, and your own judgement. Later levels cover tools, data protection and professional rules in detail.

Do: Write down five tasks from your last shift. Put each one in a group. If you are unsure about a task, it goes in "never" until your policy says otherwise.

Task 3: Know what the profession is actually doing

Always check how old evidence is and where it comes from. Three examples:

• January 2025, UK: The Royal Pharmaceutical Society (now the Royal College of Pharmacy) published a policy supporting the "responsible and effective use" of AI in pharmacy. It said AI should complement and augment people, not replace them, and that pharmacists must protect patient data.
• March 2026, UK: A Royal College of Pharmacy survey of 141 pharmacists (16–30 March 2026, reported in July 2026) found that 63.8% used AI at work. Of those users, 85.6% used it to streamline admin and 62.2% for professional development. 74% of all respondents had concerns about AI in practice. Among those with concerns, the most common (92%) was the risk of errors and hallucinations. Note that most respondents worked in hospital (57%) and only 2.1% in community pharmacy.
• April 2026, UK: On 20 April 2026 the General Pharmaceutical Council (GPhC) published a position statement on AI. It says pharmacy professionals stay accountable for their practice when AI is used. Level 4 covers this in detail.

For wider context, a US survey of 194 pharmacy preceptors (February–April 2024, published 2025) found that 30.4% had used an AI chatbot, and 75.3% said they were unlikely to make patient care decisions based on chatbot information.

The pattern is the same each time. Teams use AI mostly for admin and learning, and their main worry is accuracy.

Do: When someone quotes an AI statistic to you, ask three questions: Who ran the survey? When? How many people, and in which settings?

Task 4: Apply the rule of thumb

Before you use AI for any task, go through these steps:

1. Name the task. Is it admin, drafting, learning or clinical?
2. Check the group. If it is clinical, stop and use approved references.
3. Plan your check. Decide which approved source you will check the output against.
4. Keep ownership. You are responsible for anything you send or use, whoever drafted it.

Key takeaways

• This course is general information, not clinical advice. Local policy, approved references and professional judgement come first.
• LLMs predict likely text. That makes them fluent, but not reliably accurate, and they can invent facts and references.
• Suitable uses: admin, drafting and learning, with a human checking every word.
• Never use general AI for dosing, interactions or clinical decisions.
• Dated evidence (RPS policy January 2025; RCPharm survey March 2026; GPhC statement April 2026) shows teams using AI mainly for admin and learning, with accuracy as their main concern.