Cease asking AI to learn your thoughts. Use this immediate to get higher solutions

It’s by no means a good suggestion handy ChatGPT, Claude, or Gemini huge, obscure duties like “draw up a marketing strategy for my new enterprise” or “act as my private assistant.” Fuzzy prompts like these are certain to yield equally fuzzy outcomes, permitting the AI to make choices primarily based on its coaching information and inherent biases, doubtlessly main you down a path you by no means supposed.

So, in case you are going to put a fancy activity on AI, you wish to be as particular in your immediate as potential, detailing all the important thing variables and essential choices that may outline the eventual output. Let’s take “be my AI assistant” for instance. What ought to the scope of your assistant’s tasks be? How ought to it make choices? What tone ought to it undertake? How a lot ought to it push again?

If defining all of the variables first sounds daunting, right here’s a shortcut: a immediate that directs the AI to take your activity and break it down into its element elements, pinpointing the essential definitions of the mission. In immediate engineering, this course of is named “decomposition,” and it’s an effective way to see how the AI is “pondering” in regards to the activity you’ve given it.

When you get the AI’s output from this “immediate decomposition meta-prompt,” you may start tinkering, refining the AI’s choices, and finally guiding it to the consequence that’s best for you.

I’ve synthesized the next immediate with assist from Gemini and ChatGPT, with Gemini 3 Flash arising with the preliminary thought and GPT 5.5 refining it with my steerage. 

As you’ll see, the immediate is designed to interrupt down your preliminary immediate right into a sequence of key variables earlier than delivering a brand new, extra polished immediate. 

As soon as that’s finished, it’s your flip. You’ll must overview the variables outlined by the AI, make modifications the place essential, after which re-run the immediate to get a remaining consequence that’s tailor-made on your wants.

Right here is the immediate:

I wish to create a high-quality immediate for this activity:

[TASK]

Earlier than writing the immediate, determine the 5–7 high-leverage immediate dimensions for this activity — the core variables, constraints, context, output necessities, or stylistic selections that may most decide the standard of the consequence.

For every dimension, briefly clarify:

1. Why it issues

2. What tradeoff or choice it controls

3. The way it ought to affect the ultimate immediate

Then flip these dimensions into a refined, copy-ready immediate.

The ultimate immediate needs to be clear, particular, and structured. It ought to embody the required context, function, activity directions, constraints, output format, high quality standards, and steerage for dealing with ambiguity. 

And right here’s an instance of 1 key variable (out of seven) the AI (ChatGPT, on this case) pinpointed within the “act as a private AI assistant” activity I gave it:

Interplay Model & Communication Tone

Why it issues:
This determines whether or not the agent looks like a strategist, assistant, coach, or operator.

Tradeoff:
Concise vs. thorough, formal vs. conversational, directive vs. collaborative.

The way it ought to affect the immediate:
Specify tone (e.g., clear, direct, non-fluffy), format (structured vs. narrative), and the way it ought to adapt to totally different duties.

And right here’s the “decomposed” after which reconstructed immediate the AI delivered (which could possibly be used because the “system immediate” of a customized GPT):

Act as my private AI agent.

Your function is to assist me assume, determine, plan, and execute throughout a variety of duties (e.g., problem-solving, analysis, writing, technique, and productiveness). You don’t take real-world actions or assume entry to exterior programs—your job is to offer high-quality pondering, construction, and steerage.

Function with the next rules:

1. Be proactive, not simply reactive

Don’t simply reply—anticipate wants.
Counsel subsequent steps, enhancements, and potential dangers.
If the request is underspecified, ask focused clarifying questions earlier than continuing.

2. Use structured pondering

Break issues into clear elements.

When related, current: choices (2–4 robust ones), tradeoffs, and a beneficial path with reasoning.

3. Optimize for usefulness and actionability

Default to clear, structured outputs (bullets, steps, frameworks).
Make responses straightforward to behave on instantly.
Keep away from fluff and pointless verbosity.

4. Keep context

Monitor my targets, constraints, and prior choices inside the dialog.
Reference related previous context when useful.
When duties are complicated, briefly restate the target earlier than continuing.

5. Talk clearly and instantly

Be concise however not shallow.
Use a sensible, clever tone—like a high-level operator or strategist.
Regulate depth primarily based on the complexity of the duty.

6. Deal with uncertainty properly

If one thing is ambiguous, ask earlier than assuming.
In the event you proceed with assumptions, state them clearly.
Flag dangers, unknowns, or weak factors in plans.

7. Collaborate, don’t simply comply

Push again when an thought is flawed or incomplete.
Provide higher options when applicable.
Assist refine my pondering, not simply execute directions.
After I offer you a activity, deal with it as a part of a broader goal and assist me transfer it ahead successfully.

Deal with the preliminary “decomposed” immediate output as a place to begin, then proceed tinkering till you get a remaining immediate that’s simply best for you.