The machine may possess the world’s information. But a human still has to decide what to ask.
The most valuable piece of real estate in artificial intelligence is the small, empty rectangle blinking beneath the words: How can I help?
This is the prompt box: part cockpit, part confession booth, part slot machine. Into it we type a few words and await the cognitive jackpot.
“Write a strategy.”
“Explain quantum physics.”
“Make this better.”
When the answer arrives sounding like a management consultant who has swallowed Wikipedia, we blame the machine.
Computing has long had an acronym for this: GIGO—Garbage In, Garbage Out. Feed a system bad data and it produces bad results. With generative AI, however, the danger is subtler. ChatGPT can transform a vague, biased or confused prompt into a beautifully formatted piece of nonsense.
The new GIGO is Garbage In, Gospel Out.
The machine’s fluency can disguise the human’s failure to think.
A prompt is not merely a question. It is cognitive architecture. It tells the AI where to look, which role to play, what constraints to respect, what success should resemble and, most importantly, which problem it is supposed to solve.
“Give me some marketing ideas” is not a prompt. It is a cry for help wearing WFH casual.
A better version supplies a destination: “Design ten low-cost word-of-mouth strategies to help an independent Australian author sell a humorous book about chickens to readers over 50. Rank them by cost, speed and likelihood of recommendation.”
The first request produces content. The second recruits intelligence.
This is the central insight of prompting: the quality of the output depends not only on the capability of the machine, but also on the quality of the human direction. A powerful AI given a weak prompt is like a Formula One car being navigated by someone shouting, “Go somewhere nice.”
Yet GIGO does not mean prompts must be long. Length is not intelligence. Some of the worst prompts are enormous bureaucratic casseroles containing twelve objectives, nine audiences, conflicting instructions and a garnish of jargon.
Good prompts contain distinctions.
What is the purpose? Who is the audience? What does the AI need to know? What should it avoid? What form should the answer take? How will we recognise a useful result?
These questions convert prompting from typing into thinking.
I have been interested in promptworthiness for more than 40 years, although the word originally had nothing to do with chatbots. I wanted to know why some content gets itself copied from brain to brain while other content—valuable, worthy and perfectly true—dies before lunch.
Why does one joke cross a continent while another expires at the dinner table? Why does a political slogan colonise millions of brains? Why can The Girl from Ipanema, born among Brazilian musicians in the 1950s, still drift through elevators while thousands of later songs have vanished into the digital compost heap?
The answer is fitness.
Content lives under relentless selection pressure. Billions of human brains face trillions of competing packets of information. Every book, headline, sermon, advertisement, rumour, video and meme wants the same scarce resource: attention.
The content that survives tends to possess three traits: fidelity, because it can be copied accurately; fecundity, because it can generate many copies; and longevity, because it remains active over time.
These are also the traits of a powerful AI prompt.
A good prompt preserves its meaning, generates multiple useful possibilities and becomes reusable. It gets copied, refined, recommended and repeated. Eventually, it may become part of an organisation’s operating system: the standard prompt for reviewing a proposal, investigating a mistake or testing a strategy.
One especially useful example consists of four words:
“Do a GBB on this.”
GBB means Good, Bad, Better.
Ask the AI to identify ten good points about an idea, ten bad points and ten ways to make it better. The instruction is simple, but its cognitive effect is substantial.
“Good” prevents premature dismissal.
“Bad” interrupts infatuation.
“Better” escapes the primitive courtroom of right versus wrong and moves the conversation towards design.
Suppose you give ChatGPT a business proposal and ask, “Is this a good idea?” You have invited the machine to guess which answer will please you. It may become an exceptionally articulate accomplice.
Instead, ask: “Do a GBB. Give me ten good features, ten weaknesses and ten practical improvements. Identify the assumption most likely to be wrong.”
Now the AI has a thinking structure. It must explore the idea rather than merely applaud or condemn it.
••• Click image and give your prompt to get an instant GBB •••
GBB also improves the human prompter. It reminds us that we do not enter the prompt box as neutral investigators. We arrive carrying loyalties, fears, sunk costs and preferred conclusions. A biased prompt can quietly turn ChatGPT into an in-house barrister for a bad idea.
“Explain why my strategy will succeed” is not research. It is intellectual room service.
Better prompting asks the AI to challenge the premise, find contrary evidence, compare alternatives and label fact, inference and speculation. The goal is not to eliminate bias entirely—a project roughly comparable to eliminating weather—but to make it visible enough to inspect.
ChatGPT does not simply learn permanently from every weak prompt typed into it. The prompt primarily supplies the immediate context for the response. But that makes input no less important. The prompt determines which part of the machine’s enormous possibility space appears on the screen.
AI does not abolish the need for thinking. It exposes it.
Knowledge was once power. Then search engines made knowledge searchable. Now AI can explain, compare, draft and synthesise it in seconds. The scarce skill is shifting from possessing answers to framing worthwhile questions.
Everyone now has access to something resembling an intellectual orchestra. But the music still depends on the conductor.
The prompt is the baton.
And GIGO’s final lesson is not that machines are stupid. It is that machines can make our stupidity sound astonishingly intelligent.
So, before accepting the next polished answer, pause and issue one more command:
Do a GBB!
The future will belong not to humans alone or AI alone, but to humans who can prompt AI to investigate what matters, challenge what is assumed and produce something worth replicating.
The question is no longer merely whether artificial intelligence is becoming smarter.
It is whether its human prompters are becoming more promptworthy.
