You set “Always answer in bullet points, never use marketing fluff, write at a 10th-grade level” in Custom Instructions weeks ago, and for a while ChatGPT obeyed. Now you ask a question and get back a flowery three-paragraph essay, no bullets in sight. Toggling the Custom Instructions panel off and on does not help. The instructions are still saved, the model just is not following them. Custom Instructions live in a specific scope and compete with chat-level prompts, Project instructions, and the model’s own defaults. When they get ignored, the cause is almost always scope, overflow, or model switch, not the panel itself being broken.
Common causes
Ordered by hit rate, highest first.
1. Project instructions or chat-level system prompt overrode them
If your chat is inside a Project, the Project’s custom instructions take precedence. The model is following those instead of your account-level Custom Instructions.
How to judge: Look at the sidebar, is this chat inside a Project? If yes, open the Project, then Configure, and read the Project instructions. They are likely contradicting yours.
2. Context window pressure pushed instructions out of the active window
Custom Instructions are prepended to context but on a very long chat with many attachments, the model can effectively de-prioritize them as the live conversation dominates attention.
How to judge: Did the chat start obeying instructions and only stop after 30+ turns? That is the signature of context pressure.
3. The model was silently switched mid-conversation
When you hit a usage cap, Plus accounts get rerouted to a smaller model that handles long system prompts less strictly. The fallback model follows instructions more loosely.
How to judge: Check the model picker. If it shows a fallback label like “GPT-5 mini” instead of GPT-5, that is the cause.
4. Custom Instructions are too long or contradictory
Instructions over 1500 characters, or with contradictions like “be concise” and “always include 3 examples”, confuse the model. It picks one or follows neither consistently.
How to judge: Open Personalization, then Custom Instructions, and read both fields. If either is over 500 words or contains opposing rules, that is the issue.
5. The instructions toggle is actually off
Sometimes the toggle gets flipped off by accident, especially when a Project is opened or after a settings sync. Visually the fields are populated but the toggle below is off.
How to judge: Personalization, then Custom Instructions, confirm “Enable for new chats” is on.
6. The model interprets the instruction differently than you expect
A rule like “use British English” is followed; a rule like “be more direct” is fuzzier and the model often regresses to its default tone. Vague instructions get ignored under load.
How to judge: Replace your soft rule with a hard, testable one (“end every reply with the word DONE”) and see if it sticks. If hard rules stick and soft do not, the instructions just need rewriting.
Before you start
- Have your current Custom Instructions text in front of you so you can edit.
- Decide whether to keep account-level instructions, Project-level, or both with clear roles.
- Pick one or two rules that matter most, so you can test compliance after the fix.
Information to collect
- Exact text of both Custom Instructions fields (About You, How to respond).
- Whether the current chat is inside a Project, and the Project instructions if so.
- The model shown in the picker for the failing chat.
- Approximate turn count and attachment count in the chat.
- Whether you set instructions in the web UI or the mobile app.
- Plan tier and any recent cap warnings.
Step-by-step fix
Step 1: Confirm Custom Instructions are enabled
Settings, then Personalization, then Custom Instructions. Both fields should have text and the “Enable for new chats” toggle should be on. Save. Toggling off and on triggers a fresh apply.
Step 2: Test with a brand new chat
Start a fresh chat outside any Project, ask a question that should clearly trigger an instruction (for example, “explain photosynthesis”, expecting bullets if you specified bullets). Fresh chats are the cleanest test of account-level instructions.
Step 3: Audit Project instructions for contradictions
If the failing chat is inside a Project, open the Project, then Configure. Read the Project instructions. If they contradict your Custom Instructions, either align them or move the chat out of the Project.
Step 4: Shorten and sharpen the instructions
Cut Custom Instructions to under 500 words total. Replace vague rules with concrete, testable ones:
Bad: Be more concise and direct.
Good: Maximum 5 bullet points per reply. No introductory sentence.Save and test in a new chat.
Step 5: Pin your model choice
In the model picker, pick GPT-5 explicitly instead of leaving “Auto”. This reduces silent switches that water down instruction following.
Step 6: Reapply instructions inline for stubborn chats
In a long chat where instructions have drifted, send a reminder:
Reminder: from now on in this conversation, follow my account
Custom Instructions exactly. In particular: max 5 bullets, no
introductory sentence, 10th-grade reading level.This re-anchors the model without restarting the chat.
Step 7: For team-wide consistency, move rules into a Project
If multiple chats need the same behavior, create a Project, put the rules into the Project’s instructions, and run those chats inside the Project. Project instructions are usually obeyed more reliably than account-level ones for chats inside that Project.
Verify
- New chat outside any Project follows the rules from turn 1.
- New chat inside a Project follows the merged Project plus account rules.
- After 20 turns and one attachment, instructions are still followed.
- Switching from desktop to mobile and back keeps behavior consistent.
Long-term prevention
- Keep account-level Custom Instructions short, focused on tone and format only.
- Put workflow-specific rules into Projects, not at the account level.
- Use hard, testable rules (“never use the word ‘leverage’”) instead of fuzzy ones.
- Audit instructions monthly, remove rules you no longer care about.
- Pin the model in the picker for important sessions, do not rely on Auto.
Common pitfalls
- Setting 1000 words of instructions and expecting all of them followed equally.
- Mixing contradicting rules (“be concise” and “always give 3 detailed examples”).
- Assuming Project instructions and account instructions stack additively, they can conflict.
- Editing Custom Instructions in mobile and expecting instant sync on desktop.
- Blaming the model when the real culprit is a Project instruction you forgot about.
FAQ
- What is the priority order? Inline prompt in this turn > Project instructions > account Custom Instructions > model defaults.
- How long can Custom Instructions be? Each field caps around 1500 characters, but shorter is followed more reliably.
- Do Custom Instructions apply to API calls? No, the API uses its own system prompt that you control in code.
- Does Memory replace Custom Instructions? No, Memory holds facts about you across chats. Custom Instructions set rules. Both coexist.
- Why do new chats obey but old chats ignore? Old chats accumulated context that crowds out the prepended instructions over time.
- Can I have different Custom Instructions per workspace? Account-level applies everywhere; for per-workspace rules use Projects.