ChatGPT tricks most people don't know (not the usual ones)

The standard ChatGPT tips never work. "Write better prompts" - fine, but how. "Give it context" - sure, how much. "Be specific" - specific about what, exactly. These instructions describe a direction, not a technique. After a year of daily use, these are the actual features and habits that changed how useful ChatGPT is in practice.

Custom instructions are the highest-return setting that rarely gets used

Go to Settings, then Personalization, then Custom Instructions. You can tell ChatGPT things once that apply to every single conversation: your profession, your preferences, what you explicitly do not want it to do.

Current setup that works: never use bullet points unless specifically asked, never add a summary paragraph at the end of responses, write in direct prose without filler transition phrases, and assume a working knowledge of software development rather than explaining fundamentals. That configuration eliminated roughly 80% of post-editing work on written outputs. Each of those instructions addresses a specific annoying habit the model has by default. Identify your three most common edit types. Add them as instructions. Done.

Interrupt it mid-response

If ChatGPT starts going in the wrong direction, you do not have to wait for it to finish. Press Stop, then redirect immediately. The common pattern is to wait for the full response, send a correction, and waste context window on the wrong output. Interrupting mid-response and redirecting is faster and keeps the conversation tighter. This sounds obvious until you notice how rarely anyone does it.

Ask it to critique its own output

After any significant response, ask: "What is weak about this? What would you change if you were a skeptical editor?"

ChatGPT often identifies real problems it glossed over on the first pass. Not always. But often enough that it is worth doing on anything important. For writing tasks specifically, this step is more useful than asking for a rewrite - because you learn which parts actually need fixing rather than getting a different version with different problems. The self-critique surfaces reasoning the model has but does not surface unless asked.

Memory needs active management

ChatGPT has persistent memory that carries across conversations. The problem is it accumulates outdated and sometimes contradictory information over time. Go to Settings, then Personalization, then Manage Memory. Check it monthly. Delete things that are no longer true. Add things you want it to carry forward.

Memory treated as a tool you actively curate works well - it means you stop re-explaining your context at the start of every conversation. Memory left unmanaged creates confusing inconsistencies where the model acts on outdated assumptions you forgot you gave it three months ago. Both versions of this feature exist; which one you get depends entirely on whether you maintain it.

Projects are what separate casual users from people getting real value

The Projects feature lets you group conversations and give them a persistent context: specific instructions, uploaded documents, memory that applies only within that project. If you are working on something ongoing - a product launch, a long research project, a substantial piece of writing - a Project means you never re-explain the background at the start of each session.

This feature removes the most significant friction for serious work in ChatGPT. Without it, the tool becomes a series of disconnected conversations. Projects turn it into something closer to a persistent collaborator with working knowledge of your specific context. Set one up for your main current project this week and the difference will be immediate.

Temporary chat for sensitive topics

If you are working on something you do not want stored in memory or visible in your history - medical questions, personal issues, confidential client work - open a Temporary Chat from the sidebar. Nothing from that session gets saved. It exists, few users have found it, and it is worth knowing about before you need it.

The question that changes every conversation

On any complex topic - a business decision, an architectural choice, a medical situation, a negotiation - ask: "What questions should I be asking that I am not?"

ChatGPT has a broader view of most problem spaces than the person asking about them. Asking what you are missing is often more useful than asking about what you already know you want to know. This one question has changed the outcome of more conversations than any other technique. It is also the one that the standard "tips" articles never mention, because it is not a prompt template - it is a different orientation toward what the tool is for.

None of these are tricks in the clever hack sense. They are features that work as designed, used in ways that compound over time. The users getting the most from ChatGPT are not the ones who found a secret prompt formula. They are the ones who treated configuration as ongoing work rather than a one-time setup. Start with Custom Instructions, add memory management, build one Project around your current main task - and the tool you have had access to all along starts working differently.