The AI Learning Hub Journal

Prompting Is a Real Skill

Anatomy of a prompt that worksWEAK PROMPTwrite about climate changeWHAT YOU GET BACKA generic 600-word essayNo angle you can actually useWrong length for the taskReads like every other answerYou end up rewriting it anywayVague in, vague outthe model had to guess what you meantSTRONG PROMPT — six parts, each doing a jobROLE + CONTEXTYou are helping me revise for GCSE geography.TASKExplain why sea levels are rising.FORMATFive bullet points, then one summary line.LENGTHUnder 150 words in total.AUDIENCEWrite for a 15-year-old, no science jargon.EXAMPLEMatch this style: ice melts, water expands, coasts shrink.WHAT YOU GET BACKFive tight bullets plus a one-line summaryUnder 150 words, in plain languagePitched at a 15-year-old, no jargonUsable straight away — maybe one tweakEvery part you leave out is a decision you have handed to the model
Six named parts turn a wish into a brief — the model cannot guess the constraints you never wrote down

Four Things That Fix Most Bad Outputs

People who get impressive results from AI are rarely using secret phrases. They are supplying more of what the model needs. Context: who you are, what this is for, what constraints apply. Examples: one or two samples of what good looks like. Format: length, structure, and what to leave out. Iteration: reacting to the first attempt instead of accepting it. "Explain photosynthesis" and "I'm in year 11 revising for a biology exam, explain the light-dependent reactions in about 200 words, then give me three exam-style questions with answers" pull from the same model and produce completely different value.

  • Context: your level, your subject, the purpose, the audience
  • Examples: showing one good answer beats describing what good means
  • Format: say the length, structure and tone you want, and what to omit
  • Iteration: your second message is where most of the quality actually comes from

Iteration Is the Part Most People Skip

Treating the first response as the final answer is the single most common mistake. A better loop: read it, identify precisely what is wrong, and say that. "Too general — use the example I gave." "Cut it in half and drop the introduction." "You explained what it is; I asked why it happens." Each round narrows the target. It is closer to editing than to searching. If three attempts have not got you there, the prompt is probably not the problem — either the task is underspecified, or you have not decided what you actually want, or this is not a job the model can do.

  • Say what specifically is wrong rather than starting over from scratch
  • Give the fix, not just the complaint: "use shorter sentences" beats "this is bad"
  • If a chat has drifted, restart with a better first message instead of piling on corrections
  • Three failed attempts is a signal to rethink the task, not to try harder

Roles, Steps and Thinking Out Loud

A few techniques genuinely work because of how generation is conditioned on the context. Telling the model to work through a problem step by step often improves the answer, since the intermediate steps become part of what it conditions on. Assigning a perspective — "explain this the way a physics teacher would to someone who just got it wrong" — shifts the vocabulary and level. Asking it to critique its own output and then rewrite catches real problems. And asking it to give you three quite different versions, rather than one, is a fast way to find the direction you actually wanted.

  • Ask for reasoning steps on anything with a chain of logic — and read the steps, not just the answer
  • Assigning a perspective changes register and depth more reliably than asking for "a better answer"
  • "Now critique that and rewrite the weakest part" is a cheap quality upgrade
  • Request several distinct options when you are still exploring what you want

Prompting Is Really Just Thinking Clearly

Here is the part that transfers far beyond AI. To write a good prompt you must decide what you want, who it is for, what a good result looks like, and what would make it wrong. That is specification — the same skill behind a solid essay plan, a clear message to a teacher, or a brief for a group project. This is why "prompt engineering" as a job title is fading while the underlying ability keeps mattering: the interfaces get easier, but knowing exactly what you are asking for never stops being the hard part. If you cannot describe the outcome, no tool can produce it.

  • Vague prompts come from vague thinking, and no tool fixes that for you
  • The clarity you build here shows up in essays, applications and job interviews
  • Interfaces will keep improving; the ability to specify a good outcome will not become obsolete
  • If you cannot explain what "good" looks like, that is your first task, not the model's

Prefer slides, quizzes, and saved progress? Read this lesson in the library — free, no sign-up.