2Site / Wiki / Understanding AI

Prompts that work: patterns, not magic

Understanding AI·3 min read·Updated: Aug 4, 2026
The short version
  • Good prompts consist of four building blocks: role and context, a concrete task, examples and the desired format.
  • Iteration is normal: sharpening the result almost always beats rolling the dice on a brand new prompt.
  • The most common mistakes are vague instructions, too many tasks at once and a missing target format.

Affects you if you use AI tools in daily work and the results are often too generic, too smooth or simply unusable.

01Not talent, a pattern

Two people give the same AI the same task and get completely different results. The difference is almost never the tool, it is the prompt. The AI knows neither your shop nor your audience nor your tone. Everything it does not know, it fills with average. That is exactly why bad AI copy all sounds the same. This applies to product texts just as much as to mails, category pages or replies to reviews.

02The four building blocks

You do not need to memorise long templates. Four questions are enough, and you simply answer them right in the prompt: Who writes here, and for whom? What exactly should be produced? What does good sound like? What should the result look like?

  • Role and context: who this is written for and in which situation. Example: you write for a small online shop for camping gear.
  • A concrete task: what exactly should be produced, with every fact that must go in.
  • Examples: one or two of your own texts whose tone should be matched.
  • Format: length, structure, address. For example: 80 words, two paragraphs, informal address.

03Before and after: the product description

  • Before: "Write a product description for an enamel mug." Result: interchangeable marketing speak with invented details.
  • After: "You write for a small online shop for camping gear, the audience is hikers. Write a product description for an enamel mug, 350 ml, 98 grams, scratch-proof, dishwasher-safe. Sober and warm, no superlatives. 80 words, informal address."
  • The second prompt is longer, but the result is close on the first try and only needs polish.

The difference is not a secret technique, it is information. The second prompt answers in advance the questions the AI would otherwise have filled with average.

04Sharpen instead of rerolling

Hardly any prompt lands on the first try, and that is not a flaw in the system, it is the normal way of working. Reply to the result with a targeted correction: shorter, no superlatives, make the second paragraph more concrete. The AI keeps the context and improves precisely. Only when the direction is fundamentally wrong does a restart with a better first prompt pay off. My experience: rounds two and three are the real lever, not the perfect first shot.

05Typical mistakes

  • Too vague: "make it better" gives the AI nothing to work with.
  • Everything at once: product copy, SEO title and social post in one prompt. Better one after the other.
  • No target format: without a guideline the AI decides length and tone, and usually differently than you would.
  • No context: the AI does not know your shop, you have to supply it every time or save it as a template.
  • Taking results unchecked: facts, measurements and prices need your review, the AI will invent some if needed.

Collect prompts that work in a shared note, ideally with one good example result. A small library builds up on the side, and nobody on the team starts from zero.

What you can do now
  • Take a task you have often and build the prompt properly once, along the four building blocks.
  • Give the AI one or two of your best texts as examples, the tone improves immediately.
  • Set the target format before you hit send: length, structure, address.
  • Correct within the conversation instead of starting over with every attempt.
  • Check facts, measurements and prices in every result before it goes into the shop.

Wondering what this means for your project?

Let's talk