By Mario Alexandre · March 27, 2026 · 8 min read
I used to think prompt engineering was about picking the right words. Then I saw the truth: it is about structure. The words are the content. The structure is the signal. A prompt that works is built differently from one that does not.
Ask most people "what is prompt engineering?" and they will say: "It is the art of writing instructions for AI so it gives better answers." That is not wrong, but it is not the full story. It is like saying music is "sounds that sound good." True, but it leaves out the math that makes it work.
Prompt engineering is not an art. It is signal processing applied to natural language.
Your intent is a signal. It holds everything you want the AI to produce: the requirements, the context, the rules, and the format in your head. When you write a prompt, you turn that signal into words. The AI must rebuild your intent from those words.
This is exactly the problem that Claude Shannon solved in 1949:
The Nyquist-Shannon sampling theorem says a signal can be rebuilt perfectly from samples. But only if you sample fast enough: at least twice the highest frequency in the signal.
In prompt terms: an LLM can rebuild your intent perfectly. But only if your prompt covers enough dimensions of that intent. Miss a dimension and the LLM fills the gap with its own guess. Those guesses are hallucinations.
Through 275 experiments, I found exactly 6 independent dimensions that make up the "bandwidth" of LLM task specification. I call them frequency bands. Together they form the sinc-LLM framework:
These 6 bands are to prompt engineering what RGB is to color: a minimal, complete set. Every specification uses some combination of these 6 dimensions. Fewer dimensions lose information. More dimensions just add repetition.
In signal processing, aliasing happens when you sample too slowly. The result contains frequencies that were never in the original signal. They look real, but they are just artifacts of bad sampling.
AI hallucinations work the same way. When your prompt underspecifies your intent, the model fills the gaps with content that looks right but was never there. It invents facts, makes wrong assumptions, and misreads your requirements.
This is not a metaphor. The math applies directly. More bands equals more signal, which equals fewer hallucinations. That is why structured prompts beat raw prompts every time, and you can measure it.
Popular advice says: "say please," "tell the model it is an expert," "threaten consequences." None of that fixes the real problem. The model has no feelings. It does not care about social pressure. It responds to how complete your specification is.
Saying "please write a great blog post" adds zero specification. Saying "PERSONA: tech journalist at Wired, CONSTRAINTS: 800-1000 words, data-driven, no opinions without evidence, FORMAT: markdown with H2 headers" adds a lot. The second prompt works because it is more complete, not because it is nicer.
The practical tool built on this theory is sinc-LLM. You paste your raw prompt, no matter how vague or short. The tool breaks it into all 6 bands. It fills in the missing bands, turning your 12-word prompt into a complete 6-band specification.
{
"formula": "x(t) = \u03a3 x(nT) \u00b7 sinc((t - nT) / T)",
"T": "specification-axis",
"fragments": [
{"n": 0, "t": "PERSONA", "x": "Expert data scientist with 10 years ML experience"},
{"n": 1, "t": "CONTEXT", "x": "Building a recommendation engine for an e-commerce platform"},
{"n": 2, "t": "DATA", "x": "Dataset: 2M user interactions, 50K products, sparse matrix"},
{"n": 3, "t": "CONSTRAINTS", "x": "Must use collaborative filtering. Latency under 100ms. No PII in logs. Python 3.11+. Must handle cold-start users with content-based fallback"},
{"n": 4, "t": "FORMAT", "x": "Python module with type hints, docstrings, and pytest tests"},
{"n": 5, "t": "TASK", "x": "Implement the recommendation engine with train/predict/evaluate methods"}
]
}
Every band is specified. Every dimension is covered. The LLM gets a complete signal and rebuilds your intent without aliasing. That is prompt engineering: not word choice, not politeness, not tricks. Signal processing for language.
Want to go further? Read the complete 2026 guide, learn 7 techniques that work, or see 10 before-and-after examples. Just starting out? The beginner's guide walks you through it all.
// Production AI Engineering
sinc-LLM designs, audits, and stabilises production AI infrastructure: from vendor evaluation and cost accountability to incident controls and MCP architecture.
See what we do →