Prompting Is the New Literacy — And Most People Are Writing Badly

Ten years ago, "typing fast" was a paid skill. You put "60 WPM" on your resume. Then everyone learned to type and the skill vanished below the line of what anyone paid for. Prompting is on the same arc — just compressed into eighteen months instead of fifty years.

In 2026, "I can use ChatGPT" is the new "I can send an email." Nobody is impressed. But most people are still writing prompts the way they write Google searches — three keywords and a question mark. And then they wonder why the output is mediocre.

Google is a search box. AI is a colleague.

That framing shift is everything. A search box rewards keywords. A colleague rewards context. If you walked up to a senior engineer and said "auth flow?" they would ask fifteen follow-up questions before writing a single line. A good prompt front-loads those fifteen questions.

The people who get "bad output from AI" are almost always giving it 10% of the context they would give a new hire. Then they are surprised when the result reads like something written by someone who just joined the company today and missed the kickoff.

The five-line structure

This is the skeleton I use for almost every serious prompt. It takes 45 seconds longer than "just ask the thing" and produces output that is usually 3x better.

• Who is the output for (audience, level, context)
• What is the deliverable (format, length, tone)
• What constraints matter (style, things to avoid, non-negotiables)
• Relevant context (code, past decisions, background the model cannot guess)
• One example of "this is the shape of the answer I want"

You will notice none of these are clever tricks. They are the same things a good brief contains. Prompting well is mostly "writing a clear brief," which is a skill that has always been valuable — just now the audience is a model instead of a human.

Context is the cheat code

The single largest delta between amateur and professional prompting is how much context gets included. Amateurs write "refactor this function." Professionals paste the function, the file it lives in, the test that is breaking, a description of the system it is part of, and a sentence about the style the codebase uses.

Both will get code back. One will get code that fits. The other will get code that looks plausible and breaks in the third edge case.

The three prompts that keep showing up in my work

• The "explain this to me like I need to defend it in a meeting" prompt — forces the model to produce the reasoning, not just the answer
• The "steelman the opposite" prompt — surfaces what you are missing before someone else does
• The "act as a skeptical reviewer" prompt — turns the model into a critic of its own first draft, which usually improves the second draft

Why most prompt guides age badly

Specific magic-word tricks ("act as an expert", "take a deep breath") come and go with each model version. Context, audience, format, constraints, examples — that will still work in 2030. Learn the skills that outlast the model.

Prompting is not a trick. It is a briefing skill — and briefing has always been the senior skill.

Once you realise prompting is just writing a clear brief, two things happen. You get dramatically better output, and you stop looking for the next prompt-hack Twitter thread. Both are improvements.