Home › Guides

Why Is ChatGPT (or Cursor, Claude, Grok) Ignoring My Prompt? 7 Causes and Fixes

You wrote clear instructions. The AI did something else. This is almost never random: it is nearly always one of these seven causes.

1. The key instruction is buried

Models weight the start and end of a prompt most. A rule hidden in paragraph four of a wall of context gets skimmed. Fix: put the task and hard constraints first, context after.

2. Two instructions conflict

"Be thorough" plus "keep it short" forces the model to pick one. Fix: resolve it yourself: "Max 150 words; cover only the top 3 issues."

3. The goal is not checkable

"Make it better" or "clean this up" has no finish line, so the model invents one. Fix: define done: "Done = all functions have type hints and no function exceeds 40 lines."

4. Too many tasks in one prompt

Refactor, add tests, update docs, and fix the bug, all at once, means something gets dropped. Fix: one task per prompt, or a numbered list with "do step 1 only, then stop."

5. Negative instructions only

"Don't use jargon" tells the model what to avoid, not what to do. Fix: pair it with a positive: "Write for a 12-year-old; use everyday words."

6. No example of the output

If format matters, show it. One short example beats three sentences describing a format.

7. The chat is too long

In long sessions, especially in Cursor or ChatGPT, early instructions fade. Fix: start a new chat with a single consolidated prompt, or restate the rules.

Before and after

Before: write a blog post about our app, make it engaging, not too long, SEO friendly, mention features

After: Write a 600-word blog post announcing [App] to indie developers. Goal: get signups. Structure: hook, problem, 3 features (A, B, C) with one sentence each, CTA. Include the phrase "free invoice tool" in the title and first paragraph. Tone: casual, no buzzwords.

Let the bot do it

Paste your prompt into /fix in the free Prompt Medic Discord and get a diagnosis plus a rewrite in seconds. 3 free fixes a day.

Add to Discord

FAQ

Why does ChatGPT ignore some of my instructions?

Usually because the instruction is buried in a long prompt, conflicts with another instruction, is phrased vaguely, or is one of too many tasks in one request. Put the most important rule first, remove conflicts, and split big asks into steps.

Does yelling in ALL CAPS help?

Rarely. A specific, testable instruction ("respond in under 100 words") works far better than emphasis ("BE CONCISE!!!").

Why does the AI forget instructions later in a chat?

Long conversations push early instructions out of focus. Restate key constraints, or start a fresh chat with a consolidated prompt.