5 ChatGPT Codes That Change How It Answers

These five codes are short labels you type before or after a ChatGPT prompt. ChatGPT reads each one as an instruction about how to answer. They are a prompting habit people share, so they work best when you also say what the code means. Every prompt below does that for you. Copy one and replace the details in square brackets.

How to use the codes

1. Open a new ChatGPT conversation.
2. Pick the code that fits what you want and copy its prompt from this guide.
3. Fill in the square brackets with your own details, then send it.
4. If you want ChatGPT to keep using a code for the whole chat, send the setup prompt at the end of this guide first.

1. /redteam

Use it when you want your idea stress tested before you share it. ChatGPT looks for the weak spots and the ways it could fail.

/redteam means: critique this as harshly as you can, like a sceptical expert whose job is to find problems. I'm planning to [your idea]. My goal is [goal] and my constraints are [constraints]. List every weak spot and wrong assumption, ranked from most to least serious. For each one, suggest how I could fix it or test it cheaply.

2. /eli10

Short for explain like I'm 10. Use it for any topic that feels too technical.

/eli10 means: explain this like I'm 10 years old, using short sentences and one simple example. Explain [topic]. After the explanation, give me one sentence I could use to explain it to a friend.

3. /socrates

Named after the Socratic method, where a teacher guides you with questions. ChatGPT asks you one question at a time until you reach the answer yourself. Good for decisions and for learning something properly.

/socrates means: guide me to the answer with questions. Help me work out [your question or decision]. Ask me one question at a time and wait for my reply before asking the next. When I reach a conclusion, sum up what I decided and why.

4. /ooda

The OODA loop is a decision cycle first used by military pilots: observe, orient, decide and act. Use it when a problem feels messy and you need a clear next step.

/ooda means: work through my problem using the OODA loop. My problem is [problem]. Here is the context: [context]. Observe: list the facts I've given you and what's missing. Orient: explain what those facts mean for me. Decide: recommend one option and say why. Act: give me the first three concrete steps to take this week.

5. /alt3

Use it when the first answer feels too safe. You get three different takes on the same request so you can pick the best one or combine them.

/alt3 means: give me three clearly different versions of the answer, each with a different approach or tone, and give each one a letter. [Your request]. After the three versions, tell me which one you'd pick for my situation and why.

Two bonus codes

/meta

Asks ChatGPT to look at your question before it answers, so it can point out anything unclear or missing.

/meta means: before answering, tell me what you think I'm really asking and what information is missing. Then answer. My question is [question].

/truth

Asks ChatGPT to be upfront about what it doesn't know. It can still make mistakes, so check anything important yourself.

/truth means: say clearly when you are unsure, and never make up facts or sources. If you don't know, say so. My question is [question].

Use the codes for a whole chat

Send this once at the start of a conversation. After that you can type just the code, like /eli10 followed by your topic, and ChatGPT will know what it means.

For the rest of this chat, treat these codes as instructions. /redteam: critique my idea as harshly as you can and rank the weak spots. /eli10: explain like I'm 10 with one simple example. /socrates: guide me to the answer with one question at a time. /ooda: work through my problem as observe, orient, decide, act and end with next steps. /alt3: give me three clearly different versions, each with a letter. Reply with OK when you're ready.

Keep in mind

These codes are a prompting habit, and the instruction you give with them does the work. ChatGPT can still get things wrong with any code, so double check facts and advice before you rely on them.