What makes a good prompt?
One example per building block – to remember it
Try it yourself: weak vs. good
> Write something about discounts.
Result: vague, generic, barely usable
Why the same question gets different quality answers
A language model – as shown in "How does a language model actually 'think'?" – predicts token by token what's likely to come next. A vague prompt leaves many plausible continuations open; a precise prompt narrows the space of sensible answers sharply. A good prompt means giving the model enough context that the most likely answer is also the one you actually want.
The four building blocks of a good prompt
Context & role: Who's asking, in what situation, for whom? Giving a role and starting point cuts off many wrong interpretations from the outset. Concrete goal & format: "Summarize this" leaves everything open. Explicitly naming length, structure and audience decides whether the result is usable right away or needs rework first. One task at a time: Bundling five different requests into one prompt raises the risk that individual parts get answered worse or incompletely. Separate, focused prompts more reliably deliver complete results. Iterate instead of one-shot: The first answer is a starting point, not a final result. Targeted refinement ("shorter", "more concrete", "for a different audience") almost always beats a perfect prompt on the first try.
Why this matters for you as a decision-maker
These four building blocks also serve as a yardstick for evaluating AI tools and vendors: a tool that only delivers good results with carefully constructed prompts either needs trained users or – better – built-in prompt templates for recurring tasks. That's exactly what Beyond Prompt builds into automations, instead of hoping for a perfect prompt from users every time.
Key takeaways
- ✓Context, role and desired format usually decide answer quality more than the question itself.
- ✓A concrete output format (length, structure) only happens if you explicitly ask for it.
- ✓Bundling multiple tasks into one prompt raises the risk of incomplete or weaker answers.
- ✓Prompting is a dialogue: the first answer is a starting point to refine, not a final result.
- ✓These principles apply equally to any language model – ChatGPT, Claude, Gemini.
→ Which chat assistant fits you? AI chat assistants compared
Want to improve your own prompts?