ChatGPT Not Following Instructions? Check Your Prompt and Rules

Summary

• If ChatGPT ignores instructions, the cause is usually conflicting directions, missing constraints, or instructions placed where the model treats them as lower priority.
• Rewrite your prompt into a clear “role + task + constraints + output format + checks” structure and remove ambiguity.
• If Custom Instructions seem “not working,” verify you are testing in a fresh chat and that your request does not conflict with safety or policy limits.
• When ChatGPT “forgets,” it is often context-window pressure: long threads, lots of pasted text, or multiple goals competing for attention.
• Use a repeatable rules workflow: keep a stable rules block, restate critical constraints near the end, and reuse proven instruction sets consistently.

ChatGPT ignores instructions when your prompt is ambiguous, conflicting, too long, or overridden by higher-priority constraints; tightening your rules and placement usually fixes it.

Why ChatGPT “doesn’t follow instructions” (the common root causes)

When people say “ChatGPT is not following instructions,” they usually mean one of these things is happening:

• Your instructions conflict with each other. Example: “Be extremely detailed” and “Keep it under 100 words.”
• Your instructions are underspecified. Example: “Make it better” without defining what “better” means (shorter, more persuasive, more formal, more accurate, more SEO-friendly, etc.).
• Your instructions are buried or diluted. If the thread contains many messages, the model may focus on the most recent or most salient parts of the conversation.
• Your request triggers a limitation. Some requests can’t be completed as asked (for example, asking for disallowed content, or asking it to “guarantee” something it can’t know).
• You’re mixing goals. Example: “Write a funny ad, but also a legal disclaimer, but also a technical spec, but also keep it casual.”

Step 1: Diagnose the failure mode (so you fix the right thing)

Before rewriting your prompt, identify what “not following instructions” looks like in your case. Use this quick checklist:

• It ignores the format (you asked for a table, it gave paragraphs) → your format constraints are not explicit enough, or you didn’t specify what to do if it can’t comply.
• It changes tone (you asked for formal, it’s casual) → your tone instruction is vague or competing with other cues (like emojis, slang, or “make it fun”).
• It adds things you said not to add → your “don’t” constraints are too broad, or you didn’t provide an allowed alternative.
• It forgets earlier rules → the conversation is long, or you didn’t restate the critical rule near the end of your prompt.
• It refuses or hedges → your request may be outside what it can do, or it needs more context to answer safely and accurately.

Step 2: Rewrite your prompt into a “rules-first” structure

A practical way to reduce instruction-following issues is to separate rules from content. Here’s a structure you can copy and adapt:

A prompt template that improves compliance

• Role: Who the assistant should act as.
• Task: The single primary outcome.
• Inputs: The data you’re providing (and what to do if something is missing).
• Constraints: Length, tone, reading level, banned items, must-include items.
• Output format: Exact headings, bullets, table columns, JSON keys, etc.
• Quality checks: A short checklist the model should verify before finalizing.

Example: “ChatGPT is ignoring my formatting instructions”

Less reliable prompt:
“Rewrite this and make it more professional. Also put it in a table.”

More reliable prompt:
“Role: You are a professional editor.
Task: Rewrite the text below for a professional tone while preserving meaning.
Constraints: Do not add new claims. Keep it under 180 words.
Output format: Return a 2-column table with headers ‘Original’ and ‘Rewritten’. Each row should be one sentence.
Quality checks: Confirm the rewritten version is professional, under 180 words, and contains no new factual claims.
Text: [paste text]”

Step 3: Remove conflicts and hidden contradictions

Conflicting instructions are one of the fastest ways to get “ignored” rules. Fix conflicts by choosing a priority order and stating it explicitly.

Common conflicts (and how to resolve them)

• Length vs. completeness: If you need both, specify what to cut first. Example: “If you must shorten, remove examples before removing steps.”
• Creativity vs. accuracy: If accuracy matters, say so. Example: “If you’re unsure, ask clarifying questions instead of guessing.”
• Multiple audiences: Pick one primary audience and one secondary audience. Example: “Write for a non-technical manager; include a short appendix for engineers.”
• “Don’t do X” without an alternative: Replace with “Do Y instead.” Example: “Don’t be verbose” becomes “Use bullets; max 6 bullets; 1 sentence each.”

Step 4: Put critical rules in the right place (and repeat the ones that matter)

Instruction placement matters. If you have one or two rules that must be followed (like “no invented citations” or “answer in JSON”), restate them near the end of your prompt right before the input.

A simple placement pattern

• Top: Role + task + the 2–4 most important constraints.
• Middle: Supporting details, examples, edge cases.
• Bottom: Output format + “must-follow” rules + your input text/data.

This reduces the chance that the model treats your key constraints as background noise.

Step 5: If ChatGPT “forgets instructions,” reduce context pressure

“ChatGPT forgetting instructions” is frequently a context management problem: long threads, multiple pasted documents, or repeated revisions can push earlier details out of focus.

What to do when the thread gets long

• Start a fresh chat and paste a compact “rules + context” block instead of relying on earlier messages.
• Summarize the state in your own words: “Here’s what we decided so far…” then list the non-negotiables.
• Split the job into stages: outline first, then draft, then edit, each with its own constraints.
• Use a “spec”: a short requirements list that you keep stable and reuse across chats.

Example: A reusable “rules block” you can paste

Rules (must follow):
1) Ask up to 3 clarifying questions if requirements are missing.
2) Do not invent sources, quotes, or statistics.
3) Use US English.
4) Output format: [your format].
5) If a constraint conflicts with another, prioritize: accuracy > format > tone > brevity.

Step 6: When “Custom Instructions not working” is the real issue

If you’re thinking “ChatGPT custom instructions not working,” separate two possibilities:

• They are applied, but overridden by your current prompt (or by constraints the model must follow).
• They are not being applied in the way you expect due to chat/session differences, testing method, or account/platform behavior.

Practical checks

• Test in a new conversation. Old threads can carry momentum and competing context.
• Make your test unambiguous. Example: If your Custom Instructions say “Always answer in bullet points,” ask a simple question and see if it returns bullets.
• Avoid conflicting prompt text. If your message says “Write a paragraph,” that conflicts with “Always use bullets.”
• Keep Custom Instructions short and specific. Long, multi-goal instruction sets can conflict with themselves.

If you rely on Custom Instructions for critical formatting or compliance rules, consider also pasting a short rules block into the prompt for important tasks.

Step 7: Use “self-check” instructions (without making the answer worse)

A lightweight self-check can improve adherence to constraints, especially for formatting and “must include/must not include” rules.

Self-check examples you can add

• Format check: “Before finalizing, verify the output matches the required headings and contains exactly 6 bullets.”
• Constraint check: “Confirm you did not add any new factual claims beyond the provided text.”
• Scope check: “If the user asks for something outside the provided data, ask a clarifying question instead of guessing.”

Keep the checklist short. Overly long checklists can compete with the task itself.

Step 8: Handle instruction conflicts with a “priority ladder”

If you frequently see ChatGPT ignoring instructions, add a priority ladder so the model knows what to do when rules collide.

Example priority ladder:
“When constraints conflict, prioritize in this order: (1) safety/policy compliance, (2) factual accuracy, (3) required output format, (4) user’s tone preferences, (5) brevity.”

This won’t solve every issue, but it reduces unpredictable tradeoffs.

A practical workflow for keeping your best rules consistent (and reusing them)

If you’re repeatedly pasting the same “rules block” or prompt skeleton, the real fix is often consistency: save the instruction set you know works, then reuse it across chats and tasks.

Using CopyCharm to save, find, and reuse instruction sets

CopyCharm is a Windows desktop app and local-first context workbench for copied text. A simple workflow looks like this:

• Save: When you write a rules block that ChatGPT follows well (for example, a formatting spec, a “no hallucinated citations” rule, or a brand voice block), save it as a Saved Prompt in CopyCharm. If you copy a great example output or a key requirement list, you can also keep it as a clip and Favorite it for quick access.
• Find: Later, search your past clips or saved prompts in CopyCharm using a memorable phrase (like “priority ladder” or “JSON schema”).
• Reuse: Copy the saved prompt back into ChatGPT (or into another tool) and adjust only the task-specific inputs, keeping your rules stable.

When ChatGPT access matters: connector vs. manual reuse

If you want ChatGPT to pull in your saved rules without you manually pasting them each time, CopyCharm offers an authenticated ChatGPT connector backed by optional AI Access sync. After you sign in with an eligible active CopyCharm purchase, authorize the CopyCharm Desktop connection, enable and complete AI Access sync, and authorize the ChatGPT connector, ChatGPT can search and retrieve only supported Synced Data (Favorite Clips, Saved Prompts, and optional Other Clips within your selected time range). ChatGPT cannot access unsynced local CopyCharm data.

For Claude, Gemini, Cursor, email, documents, and other applications, the workflow is manual: search or retrieve the content in CopyCharm, then copy/paste it into the destination app.

Learn more about CopyCharm

Quick “fix it now” checklist (copy/paste)

• Make one primary task. If you have multiple tasks, split them into steps.
• State constraints as testable rules. Replace “keep it short” with “max 120 words.”
• Specify output format precisely. Headings, bullet count, table columns, or JSON keys.
• Resolve conflicts with a priority ladder.
• Restate the top 1–2 rules near the end.
• If the chat is long, restart with a compact spec.

Frequently Asked Questions

Back to FAQ Table of Contents

Back to FAQ Table of Contents

Back to FAQ Table of Contents

Back to FAQ Table of Contents

Back to FAQ Table of Contents

Back to FAQ Table of Contents

Back to FAQ Table of Contents

Back to FAQ Table of Contents

Related Guides

• What Is an AI Context Pack?
• How to Prepare Better Context for ChatGPT
• How to Organize Copied Text for AI Work
• Source-Labeled Context for AI: Why It Matters
• ChatGPT Context Pack for Consultants