Why AI gives you generic answers
Generic questions produce generic answers. The fix is not a better prompt — it is giving the AI enough context to reason, rather than guess.
The AI is not holding back. It is working with what you gave it. Most people ask questions — short, clean, context-free — and then receive answers that are equally short, clean, and useless. The answers are generic because the question was generic. Add the real situation, and the answer changes.
This is not a flaw in the model. It is the natural consequence of asking something without explaining why you are asking.
Questions versus jobs
There is a difference between asking a question and giving a job. "How do I improve my LinkedIn profile?" is a question. "I am a 40-year-old project manager switching into data roles after 15 years in construction, and my current profile makes me look like I am going backwards — what should I change?" is a job.
Both arrive in the same text box. The AI processes both using the same model. But the second one gives it something to reason with: a specific situation, a specific goal, a specific constraint. The first gives it nothing except a category of problem that could apply to any of hundreds of millions of people. The output from the first prompt is written for all of them.
The answer you wanted existed. The model just had no way to produce it without the context that would have made your situation distinct.
Why the prompt engineering era failed
For a few years, the internet was full of acronyms telling you how to write better prompts. RACE. CRISPE. A dozen others. Each one had a framework that asked you to fill in structured fields before sending anything to the AI.
The frameworks were not wrong, exactly. They were just cumbersome and unnecessary. The underlying insight — that AI output improves when you give it more context — was right. The method — memorising an acronym and filling in fields — was not how people actually think or talk.
Prompt engineering as a skill set also had a fundamental problem: it was optimising the wrong variable. The goal was to write a perfect single prompt and get a perfect single answer. But that is not how good thinking works, and it is not how good conversation works either. The better approach is to start messy and iterate.
Do not craft the perfect prompt. Start with the problem, give AI the messy truth.
Chapter 2.2 · ~1303s
The marble analogy
In the course, the way iterative prompting gets explained is through a marble analogy: your first response from the AI is a block of marble. It is your job to chisel away at it — responding to the output, redirecting, refining — until you have the thing you actually needed. Trying to get a perfect statue on the first send is not just hard; it takes longer than the alternative.
The analogy is useful because it reframes the first prompt. It does not have to be good. It just has to get the marble in the room.
What you cannot skip is giving the AI enough to work with before the chisel work begins. A blank block of marble with no shape specified gives you back a blank-looking output. The context you provide on the first send determines the quality of the material you are chiselling.
What context the AI actually needs
More real context gives the AI better reasoning, better plans, and better results. The problem is knowing what counts as useful context.
It is not a resume. It is not a detailed backstory. It is the things a colleague would need to know before they could help you effectively: what is happening, what is making it difficult, what you are actually trying to achieve, and what is off the table. That is roughly it.
When people give the AI a question instead of a situation, they leave all of that out — because questions do not naturally include it. The question is the end point. The context is everything that explains why you are standing at that end point, and context is what the AI uses to decide what kind of answer will actually help.
The SPILL method
The course names this approach SPILL. The name is also the instruction: stop engineering and spill what is on your mind, the way you would to a therapist, or to a colleague you trust. You do not need to remember what the letters stand for. The point is to lower the barrier to giving context rather than raise it.
SPILL works because it is asking you to do what you would naturally do before asking a real person for help. You would explain what has been going on. You would say what is making it hard. You would say what you need to happen. You would say what cannot change. Then you would stop talking and let them respond.
That sequence — or the absence of it — is the entire reason most AI answers feel generic. The course is where you see the method applied on screen, across real problems, so you can develop a feel for when you have given enough and when you are still holding back.
What usually goes wrong
The most common failure is treating the AI like a search engine. A search engine is designed for short queries. The AI is designed for conversation. When you write to it the way you would write a Google search, it answers the way Google would — by returning the general-purpose version of whatever you asked.
The second failure is giving context, getting a better response, and then not iterating. The first response is not the final one. It is the starting point. The moment you get something close but not quite right, the best move is to say exactly what is wrong with it — not to rewrite the whole prompt from scratch. Small corrections to a partially good answer are faster and more accurate than starting over.
The output scales with what you put in
There is no trick to this. The AI has a vast amount of reasoning capability available. It can apply that capability to a three-word query or to a paragraph of actual situation. What it applies in both cases is the same reasoning; what changes is how precisely it can direct that reasoning at your specific problem.
The people who complain that AI answers are generic are usually sending generic questions. The fix is immediate and costs nothing.
Drawn from chapters 2.2, 2.3 of AI Magic 2033.
