Prompts Still Matter.
The Formula Course Still Doesn't.
The prompt-engineering gurus have one genuinely good point: how you word things still changes what the AI gives you. The research agrees with them. That same research is exactly why their formula course is the wrong thing to buy.
First, let me give them their best point
I'm not going to pretend prompting doesn't matter — that would be easy to knock down, and it's not true. It does. Studies still find large language models are sensitive to how a prompt is worded and formatted: in some tests, changing only the formatting moved accuracy by as much as 76 points. Wording has real effect. The gurus are right about that, and I'll grant it fully.
So if you came expecting "prompting is dead," it isn't. Good. Now watch where they lose the argument.
The leap they make
They go from "wording matters" straight to "so memorise my 50 formulas, my RTF, my RACE, my 100 magic prompts." That's the sleight of hand — and their own evidence doesn't get them there.
Because when researchers ask what actually moves the result, the answer isn't a magic word. It's clarity and context. The recurring finding: most prompt failures come from ambiguity, not from the model's limits. Ambiguity is not fixed by typing "act as a world-class consultant." It's fixed by saying what you actually want. The sensitivity they cite is sensitivity to clear structure — an argument for thinking clearly, not for buying a vocabulary.
What's actually obsolete — and what isn't
Here's the clean line, and notice I'm not claiming prompting is over:
Still real — but not their product:
- Clarity and context. Knowing what you want, giving the background the model can't guess, and iterating until it's right. Free, common sense, and you already do it with people every day.
- Context engineering — for people building AI products: agents, retrieval, tool use, systems. Genuine engineering, a real discipline — but learned by building, not a course of magic words for a business owner who just wants better answers.
The guru's whole trick is blurring those two — waving at the hard, real builder's craft to sell you a memorisation course that is neither clarity nor engineering.
"But the framework helps beginners structure their thoughts"
The strongest defence they have, so let me take it head-on. Sure — a template can be training wheels that force you to be specific. Fine. But then the value was the clarity it forced out of you, not the acronym. Keep the clarity; drop the scaffold the moment you can ride without it — which is about a week. Nobody should pay course money to memorise training wheels they'll outgrow before the refund window closes.
What I see running this for real
We run AI on real work daily. When it goes wrong, it is almost never a missing magic word — it's an instruction that was ambiguous or incomplete: I left out context, or I hadn't decided what "good" looked like before I asked. The fix is never "find the better formula." It's always "say it properly." That's the entire skill now, and it matches exactly what the research keeps finding.
The bottom line
The gurus are right that prompts matter. The honest conclusion from that fact is to get clear — not to get a formula. One of those is free and rising in importance as the models get better at plain language. The other is a worksheet priced like a secret. The same studies they quote to prove prompting matters are the studies that say you don't need their course to do it.
Published by IMA AI — June 2026. On prompt sensitivity and the "ambiguity, not the model" finding: research on prompt-formatting effects on LLM performance (arXiv, 2024) and current prompt-engineering surveys.