Generative AI Practitioner

Domains and learning objectives

A domain’s weighting is its share of the discipline, and the share of the examination paper drawn from that domain. The study material and the examination questions are written to the objectives.

1. Domain 1

   Generative Model Foundations

   18%

   • 1.1Distinguish the pretraining, instruction-tuning and inference stages of a large language model lifecycle and what each contributes to behaviour
   • 1.2Interpret the effect of sampling controls such as temperature, top-p and maximum output length on variability and reliability
   • 1.3Identify how tokenisation and context window limits affect cost, truncation and the handling of long or non-English text
   • 1.4Distinguish how diffusion models turn noise into images from how autoregressive models generate text token by token
   • 1.5Select an appropriate model class or capability tier for a stated task, balancing quality, speed, cost and access constraints
   • 1.6Recognise hallucination, knowledge cut-off effects and arithmetic weakness as predictable consequences of how models are built

2. Domain 2

   Prompt Design and Iteration

   22%

   • 2.1Select prompt components (role, context, task, constraints, output format, examples) appropriate to a stated goal
   • 2.2Distinguish zero-shot, few-shot and reasoning-eliciting prompting and identify when each improves results
   • 2.3Evaluate a weak prompt against a target outcome and select the revision most likely to improve the output
   • 2.4Interpret an unsatisfactory response to diagnose whether the fault lies in the instructions, the supplied context or the task framing
   • 2.5Apply output-format specifications, including tables, structured lists and JSON-like templates, to obtain reusable results
   • 2.6Construct multi-step prompt chains that decompose a large task into ordered, checkable stages

3. Domain 3

   Retrieval and Grounded Workflows

   15%

   • 3.1Distinguish tasks that need retrieval or grounding from tasks the model can answer reliably from its training alone
   • 3.2Interpret the stages of a retrieval-augmented pipeline: splitting documents into chunks, embedding, similarity search and prompt assembly
   • 3.3Evaluate how chunk size, retrieval quality and stale sources affect the accuracy of grounded answers
   • 3.4Select an appropriate grounding method (file upload, long-context prompting, web-connected search, retrieval pipeline) for a stated information need
   • 3.5Identify citation and traceability practices that let a reader verify a grounded answer against its sources

4. Domain 4

   Multimodal Creation and Editing

   10%

   • 4.1Select image-generation prompt elements, including subject, style, composition, lighting and negative guidance, for a stated creative brief
   • 4.2Distinguish the practical capabilities of text-to-image, image-to-image, vision-input, speech-to-text and text-to-speech tools
   • 4.3Evaluate generated media for common artefacts and defects such as distorted text, anatomical errors and inconsistent lighting
   • 4.4Interpret disclosure, provenance and licensing considerations that apply before publishing AI-generated media

5. Domain 5

   Output Evaluation and Quality Control

   15%

   • 5.1Evaluate an AI output against explicit criteria such as factual accuracy, completeness, relevance, tone and formatting
   • 5.2Distinguish types of error, including fabricated facts, false citations, subtle omissions and confident wrongness, and select a suitable verification method for each
   • 5.3Construct a simple rubric and small test set for a generative AI task that a team performs repeatedly
   • 5.4Calculate basic quality measures, such as pass rates and error rates, from a structured spot-check recorded in a spreadsheet
   • 5.5Interpret the limits of using one AI model to review another model's output

6. Domain 6

   Safe Use, Data Care and Work Integration

   20%

   • 6.1Identify categories of information, including personal data, client confidences and commercial secrets, that must not be entered into tools without appropriate safeguards
   • 6.2Interpret core data protection principles, such as purpose limitation, minimisation and individual rights, as they apply to workplace generative AI use
   • 6.3Distinguish consumer and enterprise tool arrangements, including training-on-your-data settings, retention and administrative controls
   • 6.4Evaluate a proposed workplace use case for risk level and select proportionate human oversight and disclosure
   • 6.5Recognise prompt injection, jailbreaking and data leakage risks in connected or automated workflows
   • 6.6Select integration patterns that embed generative AI into everyday work, such as drafting, summarising, meeting notes and spreadsheet analysis