What kind of problems can a computer solve by creating something from scratch, rather than just following a fixed set of instructions? This is the core question that generative AI answers, and it is a central topic for the Generative AI Leader exam. Understanding generative AI's capabilities helps leaders identify where it can create real business value, moving beyond simple automation to imaginative problem-solving.
Jump to a section
A simple way to picture Generative AI Capabilities and Use Cases
Have you ever stared into your fridge, full of ingredients, and had no idea what to cook? You could follow a recipe for a specific dish, like a traditional software program that does one thing. But what if you could tell someone, 'I'm craving something warm, savoury, with a hint of spice, and I'm out of chicken' — and they invent a brand new, perfect meal for you on the spot, using what you have?
That's the difference between traditional software and generative AI. A traditional program is like a vending machine: you press B3, and you always get the same bag of crisps. Generative AI is like having a personal chef who learns your tastes. You give the chef a 'prompt' — a description of what you want — and they use their knowledge of thousands of recipes, ingredients, and cooking techniques to create something entirely new. They can write a poem in the style of Shakespeare, design a logo for a dog bakery, or draft a friendly email. They aren't just retrieving a pre-made meal; they are generating a new one based on patterns they learned from studying millions of existing meals. This ability to create novel, human-like content — text, images, code, music — is the core capability of generative AI.
Generative AI is a type of artificial intelligence (AI) that creates new content. This is different from traditional AI, which typically analyses data to make a prediction or classification. Traditional AI might look at millions of cat photos and learn to identify whether a new photo contains a cat or a dog. Generative AI, on the other hand, learns from data to produce similar but original examples. It might look at millions of cat photos and then generate a picture of a cat that has never existed.
The magic behind this comes from a type of computer model called a Large Language Model (LLM) or a foundational model. Think of an LLM as a massive, complex brain that has been trained on a huge portion of the public internet. It has 'read' billions of sentences, paragraphs, articles, books, and code repositories. By doing this, it has learned patterns: how words relate to each other, how grammar works, how stories are structured, and even how to imitate different styles of writing.
The process begins with a 'prompt'. A prompt is simply the instruction or question you give to the model. For example, a prompt could be: 'Write a short poem about a cloud that dreams of becoming a thunderstorm.' The LLM doesn't 'think' like a human. It uses the patterns it learned during training to predict the most likely next word, one word at a time. It starts with the prompt and then guesses the most probable first word of the poem. Then it uses that word and the prompt to guess the second word, and so on. This process is called 'autoregressive generation' (self-predicting). The result is a coherent, grammatically correct, and often very creative piece of text that is entirely new.
Generative AI capabilities span several different content types. The most common are:
Text generation: Writing emails, articles, summaries, marketing copy, code, and even poetry. This is the cornerstone of chatbots.
Image generation: Creating realistic images or art from a text description (e.g., 'a photorealistic image of a cat wearing a top hat, sitting in a library').
Code generation: Writing software code in various programming languages based on a description of what the program should do.
Audio generation: Synthesising human-like speech, creating music, or generating sound effects.
Video generation: Creating short video clips from text prompts or existing images.
Why does this matter? Before generative AI, if you wanted a computer to generate a personalised email for a customer, you had to write a special software program for that specific task. This was expensive, time-consuming, and brittle. Small changes required reprogramming. Generative AI replaces this with a general-purpose tool that can adapt to almost any task you describe in plain language. It democratises creativity. A non-programmer can now 'create' a software prototype or a marketing image just by describing what they want. For a Generative AI Leader, this shift is crucial: the value is not in the AI itself, but in how you use it to solve specific business problems, improve customer experiences, and accelerate innovation.
Define the Task Type
Identify whether the business problem requires creating new content (generative) or analysing existing data (predictive). A task like 'write a product description' is generative; 'predict which product will sell best' is not. This classification is the first step in choosing the right technology.
Choose the Content Modality
Determine what kind of output is needed: text, image, code, audio, or video. This dictates which specific generative AI model (e.g., a language model for text, an image generation model for visuals) is appropriate. For the exam, text and image generation are the most commonly tested modalities.
Craft the Prompt
Write a clear, specific instruction for the AI. A good prompt includes context (e.g., 'you are a marketing expert'), the desired output format (e.g., 'a 100-word email'), and constraints (e.g., 'use a friendly tone'). This step is where non-technical leaders have the most impact, as prompt quality directly affects output quality.
Ground and Fine-Tune the Model
For business-critical tasks, the general model needs to be connected to real, current data (grounding) or specially trained on your own documents (fine-tuning). This prevents hallucinations and ensures the output is based on your specific facts, not just internet patterns.
Review, Edit, and Deploy
The AI generates a first draft. A human must review it for accuracy, tone, and appropriateness before using it in the real world. This 'human in the loop' step is essential to mitigate risks like bias, hallucination, and copyright infringement. After approval, the output can be deployed in emails, apps, or reports.
Consider Sarah, the marketing director for a midsize company that sells organic gardening supplies. She has a big product launch coming up for a new line of compostable plant pots. A month ago, she would have needed to coordinate a long, rigid process: brief a copywriter for the email campaign, brief a designer for the social media graphics, and wait for the IT team to write a script to generate personalised discount codes. This would take weeks.
With generative AI tools, Sarah's workflow looks completely different. She logs into a tool powered by a large language model and starts the process:
Step 1: Brainstorming taglines. She prompts the AI: 'Generate 10 short, punchy taglines for a new compostable plant pot that is eco-friendly and made in the UK.' The AI produces ideas in seconds, including one she loves: 'Grow green, leave no trace.'
Step 2: Creating marketing copy. She takes her favourite tagline and asks the AI: 'Write a 300-word blog post explaining why plastic plant pots are harmful and why our new compostable pot is the solution. Use a friendly, informative tone.' The AI drafts a post that she edits in 15 minutes instead of writing from scratch for three hours.
Step 3: Generating visual assets. She uses an image generation tool with the prompt: 'A photorealistic image of a plant sprouting from a brown, textured pot on a wooden table, with a soft green background. The pot should have a subtle, embossed logo.' In a minute, she has several images to choose from for her social media campaign.
Step 4: Personalising customer outreach. She copies her customer list into a spreadsheet and uses an AI tool to generate a personalised email for each customer segment, recommending products based on their previous purchases. The AI writes a friendly email offering a discount on the new pots.
Step 5: A/B testing subject lines. She asks the AI: 'Generate five different email subject lines for the new product launch, each with a different style: one urgent, one curious, one benefit-driven, one social-proof-based, and one humorous.' She picks the best two to test with her audience.
What Sarah does in a single afternoon would have taken her team a week. She is not replacing her team; she is augmenting their skills. The copywriter now focuses on strategy and final polish instead of the first draft. The designer creates custom assets instead of stock images. The IT team is freed up for more complex projects because Sarah doesn't need a software script for the personalised emails. The Generative AI leader's job is to recognise opportunities like this: finding high-impact tasks where generative AI can reduce friction, accelerate output, or create entirely new capabilities.
The Generative AI Leader exam objective 1.3 is about 'Identifying common use cases for generative AI'. The exam does not expect you to build a model or write code. It tests your ability to differentiate between problems that generative AI solves well and problems it does not. The exam questions are scenario-based. You will read a short story about a company or a department facing a challenge, and you must select which generative AI capability is the best fit.
The most frequent question patterns are:
'Which of the following is the best use case for a generative AI model?' The correct answer is always a task involving creating new, original content (text, image, code, audio). The traps will be tasks involving data analysis, classification, or simple automation that does not require content generation. For example, 'Creating a customer support ticket' is a database entry, not content generation. 'Drafting a reply to a customer support ticket' is a good use case.
'A retail company wants to generate unique product descriptions for thousands of items. Which technology is most suitable?' The answer is a large language model.
'A bank needs to identify fraudulent transactions from a dataset. Which AI approach is best?' This is a traditional prediction or classification problem, not a generative one. The trap is that beginners see 'AI' and assume generative AI is the answer. It is not.
Key topics and definitions you must memorise:
A 'prompt' is the input you give to a generative AI model.
'Fine-tuning' is the process of taking a pre-trained general model and training it further on a specialised dataset (e.g., your company's emails) so it becomes an expert in that domain.
'Grounding' refers to connecting the model to real, up-to-date facts (like your company's product catalogue) to prevent it from making things up (hallucinations).
'Hallucination' is when a generative AI model confidently produces factually incorrect information. This is a major limitation you must understand.
Use cases are commonly categorised by the type of content: text summarisation, content creation (writing), code generation, image generation, dialogue (chatbots), and translation.
Common traps set by the exam include: confusing generative AI with predictive AI (e.g., 'generating a sales forecast' is predictive, not generative); thinking generative AI can perfectly reason about facts (it cannot — it predicts words, not truth); and assuming generative AI replaces human judgement entirely (the exam emphasises the 'human in the loop' — AI augments, not replaces). The correct answer pattern for use-case questions will always describe a task where the primary output is new, original content for a human to review, not a numerical prediction or a simple data transformation.
Generative AI creates new content (text, images, code, audio) by predicting the next most likely word or pixel, not by retrieving a pre-existing answer.
A large language model (LLM) is the foundational technology behind most text-based generative AI, trained on vast amounts of public text data.
The exam distinguishes generative AI use cases (creating content) from predictive AI use cases (analysing data to make forecasts or classifications).
A prompt is the plain-language instruction you give to a generative AI model, which determines the quality and relevance of its output.
Hallucination is a key limitation where the model produces confident-sounding but factually incorrect information, requiring human oversight (a 'human in the loop').
Fine-tuning adapts a general-purpose model to a specific business domain using your own data, making it more accurate and relevant.
Common generative AI use cases include drafting emails, summarising documents, creating marketing copy, generating images, writing code, and building customer-facing chatbots.
Generative AI does not 'understand' truth; it predicts patterns, so it must be used with caution for tasks requiring factual accuracy.
These come up on the exam all the time. Here's how to tell them apart.
Generative AI
Creates new, original content (text, images, code).
Used for open-ended tasks like drafting, designing, and composing.
Output is probabilistic and can vary each time you run it.
Predictive AI
Analyses existing data to make forecasts or classifications.
Used for closed-ended tasks like identifying fraud or forecasting sales.
Output is determinate (a yes/no or a number) based on the same input.
Large Language Model (LLM)
Learns patterns from vast amounts of data and adapts to new tasks via prompts.
Flexible; one model can write a poem, a legal contract, or code.
Can hallucinate and produce incorrect or biased outputs.
Traditional Software
Follows hard-coded instructions written by a programmer.
Rigid; each program does one specific, pre-defined task.
Produces predictable, consistent output every time (if no bugs).
Prompt Engineering
No additional training required; works with the pre-trained model.
Cheaper and faster to implement for one-off or simple tasks.
Less customised; the model retains its general knowledge base.
Fine-Tuning
Requires additional training on a specialised dataset.
More expensive and time-consuming but yields a domain expert model.
Highly customised; the model's focus shifts towards the new data domain.
Human-in-the-Loop (HITL)
A human reviews and edits the AI's output before it is used.
Reduces the risk of hallucinations, bias, and errors reaching customers.
Ideal for high-stakes applications like legal, medical, or financial advice.
Fully Automated AI
The AI system operates without human intervention after deployment.
Risks propagating errors or biases at scale without human oversight.
Suitable for low-risk tasks like generating initial creative ideas or drafts.
Mistake
Generative AI and predictive AI are the same thing.
Correct
Predictive AI analyses data to forecast outcomes (e.g., will this customer churn?), while generative AI creates new content (e.g., write a message to retain this customer).
Both are called 'AI', so beginners lump them together. The exam explicitly separates them.
Mistake
Generative AI models 'understand' and 'know' facts like a human does.
Correct
They understand patterns in language, not actual truth. They predict the statistically most plausible next word, which can lead to factually wrong but grammatically correct answers (hallucinations).
The output is so fluent and convincing that it feels like understanding. Beginners project human consciousness onto it.
Mistake
You need a computer science degree to use generative AI tools effectively.
Correct
Generative AI tools are designed to be used with plain language prompts. Leaders and non-technical professionals are the primary users in many business scenarios.
Past experience with complex, code-based software creates a mental barrier. The exam's focus on use cases assumes no coding skills.
Mistake
Generative AI can replace creative professionals entirely.
Correct
It is an augmentation tool that automates the first rough draft, but human oversight, editing, strategic direction, and quality control remain critical.
Headlines about 'AI replacing jobs' cause fear. The exam's correct perspective is 'AI assists humans, not replaces them'.
Mistake
Once trained, a generative AI model is static and perfect.
Correct
Models have a knowledge cutoff date (they only know things up to when they were trained) and can become outdated. They also have biases from their training data and require careful monitoring, grounding, and fine-tuning.
Beginners assume software is 'finished'. Generative AI models require ongoing management, which is a key leadership responsibility.
Reveal each answer, then mark whether you got it right. Score 60%+ to unlock the next chapter.
Regular AI (often called predictive or discriminative AI) analyses data to make decisions or classifications, like identifying spam. Generative AI creates new data that looks like the training data, like writing a new poem or drawing a picture of a cat that never existed.
No. It does not have consciousness, feelings, or understanding. It is a statistical pattern-matching machine that predicts the most likely next word or pixel based on the patterns it learned from its training data.
A large language model is a type of AI model trained on huge amounts of text data (essentially a large part of the internet). It can understand and generate human-like text by predicting the next word in a sequence. Examples include GPT-4 and PaLM.
A prompt is the text instruction you give to a generative AI model to tell it what you want. A good prompt is specific and includes context, format, and tone. For example, 'Write a professional thank-you email to a client for a recent meeting' is a prompt.
Hallucination is when the model generates information that sounds plausible and confident but is factually incorrect. It can invent facts, citations, or even entire events. This happens because the model prioritises linguistic fluency over truth.
It is more likely to change and augment your job than replace it entirely. It automates the creation of first drafts and repetitive content tasks, allowing you to focus on higher-level strategy, editing, and quality control. Leaders who learn to use it effectively will have an advantage.
You've finished Generative AI Capabilities and Use Cases. Continue through the Generative AI Leader study guide to build a complete picture of the exam.
Done with this chapter?