Stop Piling Everything Into One Mega-Prompt
What it is
The habit of cramming a dozen unrelated instructions into a single prompt — tone, format, length, edge cases, examples — and expecting ChatGPT to honour all of them at once.
Why it works
It doesn't, reliably. Past a handful of instructions the model starts silently dropping some. Splitting the work into a short chain of focused prompts makes each step verifiable and keeps a failure in one step from poisoning the rest.
When to use it
Recognise the anti-pattern whenever your prompt has more than about three distinct requirements, or when output quality is inexplicably uneven — that unevenness is usually overload.
When not to use it
Genuinely simple asks don't need splitting; a chain of trivial steps is its own kind of waste.
Prompt
Instead of one giant prompt, I'll go in steps.
Step 1: <first single job>. Do only this and stop. I'll confirm before we move on.Example
A prompt that asked for research, an outline, a draft, SEO tags and a social post at once produced a mediocre everything. Split into five steps, each came back sharp — and only step 3 needed a reroll.
Advanced version
When you must keep it in one prompt, number the requirements and end with 'confirm you addressed each numbered item' — the checklist forces the model to account for every one.
Common mistakes
- Blaming the model for ignoring instruction #9 of 12.
- Rerolling the whole mega-prompt to fix one part, wasting the parts that were fine.
- Never noticing the pattern, because each output is only a little worse than it should be.