Most people write prompts like they’re texting a search engine. One line, vague noun, hit enter. Then they’re surprised when the output is bland or wrong. The fix isn’t to learn some secret prompt-engineering framework — it’s to stop leaving out information a reasonably smart person would need to help you.

Why Your Prompts Aren’t Working

ChatGPT isn’t psychic. It generates the most statistically plausible continuation of whatever you gave it. Give it something thin, it fills in the blanks with the average — the most generic version of what you asked for. That’s not a flaw. It’s just how the thing works.

“Write me a cover letter” gets you a cover letter that could belong to literally anyone. “Write me a cover letter for a mid-level UX designer role at a fintech startup, emphasizing my background in B2B SaaS and my habit of running user interviews before touching Figma” gets you something you might actually send.

The gap between those two prompts is specificity, not complexity.

The Four Things Worth Adding to Almost Any Prompt

1. Context about who you are or what you’re doing

Not a full biography. Just enough to matter. Are you a teacher building a quiz? A freelancer writing a proposal? Someone trying to explain a medical test result to an elderly parent? That context shifts the output in ways that matter — vocabulary, depth, assumed knowledge, tone.

One sentence of context does more work than most people realize.

2. What format you actually want

ChatGPT defaults to whatever structure it thinks fits. Sometimes that’s fine. Often it gives you five paragraphs when you needed bullet points, or a bulleted list when you needed flowing prose for a client email. Say what you want. “Give me three short paragraphs, no headers” is a completely valid instruction.

3. What to avoid

This one gets skipped constantly. If you’re writing something that shouldn’t sound corporate, say “avoid corporate jargon.” If you’re summarizing and don’t want the model inventing details, say “only use information I provide, don’t fill in gaps.” Negative constraints are just as useful as positive instructions. Most people never use them.

4. An example of what good looks like

If you have a sample — an email in the right tone, a paragraph in the right style — paste it in. “Match the tone of this” is one of the most effective things you can do and it requires zero technical knowledge. The model is very good at pattern-matching when you give it something to match.

A Quick Before/After

Weak prompt: “Summarize this article.”

Better prompt: “Summarize this article in 3-4 sentences for someone who has no background in economics. Skip the jargon, focus on the practical impact. Paste article: [text]”

The second one takes fifteen extra seconds to write. The output gap is not small.

When to Use the System Prompt (or “Custom Instructions”)

If you use ChatGPT regularly for the same kinds of tasks, stop re-explaining yourself every session. There’s a “Custom Instructions” feature under your account settings where you can set persistent context — your profession, preferred output format, writing style. Set it once and most of your prompts get shorter overnight.

I use mine to tell ChatGPT I prefer short sentences, British English spelling, and that I’m usually writing for non-technical readers. That alone cuts probably a third of the editing I’d otherwise do.

The Iteration Habit Most People Skip

Most people try one prompt, don’t love the result, and conclude the tool isn’t very good. More useful: treat the first output as a draft and have a short conversation about it. “Make this shorter.” “The second paragraph is too formal, rewrite it in a more direct tone.” “Cut the last two bullet points.”

ChatGPT remembers the conversation context. You don’t need to start over every time something’s slightly off.

This is the bigger habit shift — moving from “one prompt, one output” to treating it as a back-and-forth. The second version is almost always better than the first, and the third is usually good enough to use.

What Prompt Frameworks Are Actually For

You’ll see things like RISEN or CO-STAR or a dozen other acronyms floating around. They’re not useless — they’re checklists that remind you to include role, context, tone, format. But most people don’t need a framework. They need to stop submitting one-sentence prompts and expecting detailed, tailored output.

If a checklist helps you slow down for ten seconds before submitting, use one. If it feels like bureaucracy for a task that should take thirty seconds, skip it.

The One Mistake That Actually Costs People

Being vague about the goal because they’re not sure what they want yet. This is understandable, but it makes the model’s job nearly impossible. If you’re genuinely not sure, use the first prompt to explore — but say that. “I’m trying to figure out the best way to approach X, can you lay out a few different angles before we go deeper?” That’s a legitimate prompt that gets a useful output.

The failure mode is submitting a vague prompt as if it were a clear one, then being frustrated when the output doesn’t match the specific vision you hadn’t actually articulated yet.

One last thing: the quality ceiling varies by task. For creative writing and brainstorming, better prompts make a large difference. For factual lookups or math, less so — and those are exactly the cases where you should be double-checking the output anyway, regardless of how well you wrote the prompt.