The AI Learning Hub Journal

Four Reasons an Answer Is Bad

Four causes, four signatures, four different fixesrewriting the prompt or hitting regenerate are guesses — fifteen seconds of diagnosis beats three more attemptsTHE ANSWER DISAPPOINTSbefore retrying, ask: which of the four?AMBIGUOUSTHE SIGNATUREa competent answer to aquestion you did not ask —consistent, well made,aimed elsewhereTHE FIXstate the reading youmeant: "our cost, notthe customer's"MISSING CONTEXTTHE SIGNATUREgeneric filler — orconfident claims aboutyour situation thatare not trueTHE FIXsupply the fact: "teamof four, not 400 —redo on that basis"TASK TOO BIGTHE SIGNATUREeverything covered,nothing developed — atable of contents thatgot expandedTHE FIXshrink and chain:"just the first section,in depth"SIMPLY WRONGTHE SIGNATUREreads perfectly, isfalse — the only onethat style cannotrevealTHE FIXverify externally orchange approach —rewording will not fix itONLY ONE OF THE FOUR IS A WORDING PROBLEMmissing context needs facts, an oversized task needs splitting, and a plain error needs checking — not better phrasingDIAGNOSE FIRST — THE FOUR CAUSES NEED FOUR DIFFERENT FIXESa fourth rewrite of the prompt is the signature of a skipped diagnosis
A bad answer has one of four causes — and only the first is fixed by rewording, so diagnose before you retry.

Diagnose Before You Retry

When an answer disappoints, most people immediately rewrite the prompt or hit regenerate. Both are guesses. There are really only four causes, and they have different fixes. The request was ambiguous, so it answered a reasonable but different question. The context was missing, so it filled the gap with a plausible default. The task was too big, so it produced something shallow across everything instead of solid on one thing. Or it is simply wrong — the reasoning broke or the fact is not true. Spending fifteen seconds identifying which one you are looking at is worth more than three more attempts, because the four fixes are genuinely different and only one of them is "rewrite the prompt".

  • Ambiguous request → a reasonable answer to a different question
  • Missing context → a plausible default filled in where your facts should have been
  • Task too big → thin coverage of everything instead of depth on one thing
  • Simply wrong → the fix is verification and a different approach, not better phrasing

Telling Them Apart

Each cause has a signature. Ambiguity looks like a competent answer to a question you did not ask — the give-away is that it is internally consistent and well-executed, just aimed elsewhere. Missing context looks like generic filler where specifics should be, or confident statements about your situation that are not true. Oversized tasks look uniformly shallow: every section is one paragraph, nothing goes deep, and it reads like a table of contents that got expanded. Plain error looks fine and is not — which is why it is the only one you cannot detect by reading style, and why it needs checking rather than rereading. Learn the four signatures and diagnosis takes seconds.

  • Ambiguity: well-made answer to the wrong question
  • Missing context: generic filler or invented specifics about your situation
  • Too big: everything covered, nothing developed
  • Plain error: reads perfectly, is false — only checking catches it

The Fix Follows the Diagnosis

For ambiguity: state the interpretation you wanted, explicitly. "I meant the cost to us, not the cost to the customer." For missing context: supply the fact, do not ask it to guess better. "We are a team of four, not a company of four hundred — redo it on that basis." For an oversized task: cut it down and chain the parts. "Just the first section, in depth." For plain error: this is the one where prompt changes will not save you. Verify externally, ask for reasoning so you can see where it breaks, or accept that this is not a task the model can do reliably and use something else. Rewriting the prompt for the fourth time is a symptom of skipping the diagnosis.

  • Ambiguity → state the interpretation you meant, in one sentence
  • Missing context → supply the fact and say "redo it on that basis"
  • Too big → shrink the scope and chain the steps
  • Plain error → verify, look at the reasoning, or change approach entirely

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