TL;DR
- The Setup: Nine prompt frameworks are in circulation, and each one has a following that treats the letters as the skill.
- The Problem: The letters are a memory aid for four jobs you have to do anyway. Reciting them without doing those jobs produces confident, useless output.
- The Cost: Framework hopping feels like progress while the actual gap stays untouched, and that gap is the constraint only you could have supplied.
- The Fix: Pick one framework and stop shopping. Spend the saved attention on the two things no acronym can give you, which are real context and a standard for what a good answer looks like.
- The Payoff: A plain prompt carrying your actual numbers beats a clever prompt carrying none. I have tested this enough times to stop being diplomatic about it.
I have read more prompt frameworks than I will admit in public. Nine are in circulation right now, each with a following that swears the letters are the difference between a usable answer and a wasted afternoon. I have run most of them, and I have watched capable marketers memorize three acronyms and still produce worse output than a colleague writing plain sentences with better inputs.
That gap is the entire article. The frameworks are not the skill. They are a checklist for four things every serious request does, and once you can name those four from memory, the letters stop mattering. What keeps mattering is whether you brought the constraint that made the request specific.
Nine Frameworks, Four Jobs
Put the acronyms side by side and the overlap is almost funny. One leads with action, purpose, expectation. Another with role, action, context, expectation. A third adds a scenario, a fourth an example, a fifth swaps the order and calls it new. Strip the letters and you have the same four questions in different jackets.
Acronyms are how humans hold a sequence in working memory, and a sequence is better than a blank box. The problem starts when the acronym becomes the thing you optimize. Someone asks why the output is bland, and the room debates the five letter version against the four letter version. Nobody asks what numbers went into the prompt.
The four jobs are stable enough to say out loud. Tell it who it is answering as and why that person cares. Name the task with a verb. Hand over the situation, including the part that makes it difficult. Describe what a good answer would let you do. Order is a preference. Any of the nine gets you there if you fill in the blanks instead of typing the acronym and hoping.
The Letter You Skip Is the One That Costs You
Everyone has a favorite letter to skip, and it is almost always the same one. People love specifying role because it feels like craft. They tolerate task because the task is why they opened the window. They skip outcome, because describing what good looks like requires having decided what you want, and deciding is the part that hurts.
Skip the outcome and you get an answer that is technically responsive and practically useless. Ask for a LinkedIn post without saying whether success means replies from prospects or saves from peers and you will get the average of both, which is a post that does neither. Ask for a positioning statement without saying whether it has to survive a sales call with a skeptical CFO and you get language that reads fine and collapses on the first follow-up.
The outcome line does two jobs at once. It constrains the shape of the answer, and it tells you whether you were ready to ask. If you cannot finish the sentence about what a good answer would let you do, you are not blocked on the model. You are blocked on a decision you have been avoiding, and no framework fixes that.
A Framework Cannot Know Your Numbers
The second thing no acronym can give you is the raw material. If the answer needs a fact about my business, the fact goes in the prompt. Not a description of the fact. The fact.
So instead of asking for a case study about a demand generation win, I paste the actual sequence, the actual reply rate, the actual thing the buyer said on the call. Instead of asking for a cold email that stands out, I paste two emails that got replies last quarter and one that got ignored, and I say which is which. The model is very good at pattern matching across examples you supply and very bad at guessing what your market looks like.
This is the honest limit of prompt engineering. Output can only be as specific as input. A framework organizes what you know and cannot manufacture what you left out. When an output is bad now, my first question is not which template was used. It is what the person failed to paste.
The same logic runs up a level, and it is why I stopped treating prompts as a personal productivity trick. Once you have repeated context worth reusing, it belongs somewhere a system can read it, which is the whole argument in the four stages of building an AI operating system. Prompts are the smallest unit. Context is the asset.
What I Do Instead of Framework Shopping
I keep one framework. When the output is bad, I run a fixed four question check before I touch the wording.
Did I name the reader? Not a demographic. The specific person whose objection I need to handle. A post written for a marketing director worried about headcount reads differently from one written for a founder who thinks marketing is a cost center, and the difference shows in the first two lines.
Did I paste the evidence? Numbers, quotes, the last three things a customer actually said. If the prompt has no evidence in it, the answer will have no evidence in it, and I will spend more time fixing that than I saved by typing faster.
Did I define done? Word count is not done. Done is a sentence like “a reader who has never heard of us can repeat the argument to a colleague.” That sentence changes what the model writes more than any instruction about tone.
Did I ask for the argument, or did I ask for the artifact? This is the one that separates a draft I can use from a draft I have to rewrite. Asking for a landing page gets me a landing page shape with filler in it. Asking for the argument that page has to make, and the objection it has to survive, gets me something I can build from.
That connects to a mistake I see in content teams, treating the format as the deliverable. A collateral shaped object called content is the same error as an acronym called a prompt. Both substitute a container for a decision. I wrote about that failure in stop publishing collateral and calling it content, and the fix is identical. Decide what it has to accomplish, then let the format serve that.
Nine prompt frameworks will not fix a prompt that has no numbers in it. The acronym was never the variable. What you pasted was the variable, and what you decided a good answer looks like was the variable. Everything else is spelling.
Open the last prompt you were disappointed by and add two lines. One line of real evidence from your own pipeline, and one describing what a good answer would let you do next. Rerun it unchanged. If the output does not improve, the problem was never the framework.
What I Actually Think
Framework collections are useful as a vocabulary and dangerous as a hobby. Learning nine is fine. Rotating through them looking for the one that finally works is a way to spend a quarter feeling productive while the real constraint sits untouched.
What changed my output more than any template was writing down, once, the four things every request needs and then refusing to send a prompt that skipped one. That is a boring habit. It also produces drafts I can ship instead of drafts I rewrite, and it survives every model change, every new interface, and every acronym some account will publish next month.
Your thinking is the input that matters. The model is a compiler for it. Give it something worth compiling.
Build a prompt practice that holds up →
