ChatGPT “secret codes” are usually shorthand instructions such as ELI5, TLDR, REDTEAM or /human. They tell the AI what kind of response you want. Treat them as prompting shortcuts rather than switches that unlock a hidden capability. For more predictable results, explain what the shortcut means and include the context for your task.
You have probably seen the posts.
Type one word. Get a better answer. Turn ChatGPT into your business coach. Make it stop agreeing with everything you say.
I understand the appeal. Nobody wants to write a page of instructions every time they need help.
But the useful part is being clear about what you need. A simpler explanation. A proper critique. A shorter answer. A plan you could actually follow.
That is something you can ask for directly.
Here are ten shortcuts, what they are useful for, and the fuller instructions I would put behind them. The examples are written for business owners, creators and people building with AI.
Do ChatGPT secret codes actually work?
They can help steer an answer when the model understands what you mean. A familiar abbreviation such as TLDR gives it a clue about the format you want. A less familiar label such as TRUTHMODE leaves more room for interpretation unless you define it.
A slash or capital letters don’t make an instruction more authoritative. And asking for honesty cannot guarantee that an answer is correct.
OpenAI’s prompt engineering guidance describes using instructions, relevant context and examples to guide responses. That is the foundation behind the approach here. These particular shortcuts are practical prompt suggestions, not an official OpenAI command list or a benchmarked set of results.
I would use a short label to remember the instruction. Then I would tell AI what I expect it to do.
Which ChatGPT prompt shortcut should you use?
| Shortcut | Use it when you need | What to specify |
|---|---|---|
| ELI5 or ELI10 | A simpler explanation | Your starting knowledge and a concrete example |
| TLDR | A useful summary | Length, audience and what must be preserved |
| REDTEAM | A challenge to your plan | Assumptions, failure points and ways to test them |
| TRUTHMODE | A more critical assessment | Evidence, uncertainty and balanced feedback |
| /human | More natural writing | A writing sample and specific style preferences |
| 80/20 | Help choosing priorities | Your goal, constraints and available time |
| SOCRATES | Help learning through questions | One question at a time and feedback on your answers |
| LINDYMODE | An emphasis on established approaches | Evidence, current relevance and limitations |
| ASK FIRST | Fewer guesses about your situation | The missing details that would change the answer |
| SCOPE CUT | A manageable first version | One user, one outcome and what can wait |
The last two are labels you can define for yourself. You can invent your own shortcuts whenever you find an instruction worth reusing.
1. How do you use ELI5 or ELI10 to explain something simply?
ELI5 means “explain like I’m five.” ELI10 asks for an explanation suitable for someone around ten. In practice, you can be more specific: explain this to an intelligent beginner who doesn’t know the terminology yet.
That avoids getting an answer that is simple but patronising.
Copy this:
Explain this in everyday language for a beginner. Define unfamiliar terms, use one concrete example and keep the important caveats. Don’t assume technical knowledge: [topic].
For app building, you might ask:
Explain what a database does using a simple booking app as the example. What information gets saved, and what happens when someone changes a booking?
Now the explanation has something to attach to. You can connect the concept to the thing you want to build.
If software terminology is stopping you from starting, my Building With AI guide explains the wider process in plain language.
2. How do you use TLDR without losing the important details?
TLDR means “too long; didn’t read.” Use it to request a summary, but tell AI what the summary needs to help you do.
“Summarise this” could produce an overview that leaves out the one decision you needed to make.
Copy this:
Summarise the text below in five bullets. Preserve the main decision, supporting evidence, important limitations and any next actions. Don’t add information that isn’t in the text.
For meeting notes, change the format:
Summarise these notes as decisions, open questions and next actions. Include an owner or deadline only when the notes actually name one.
This is useful when the output will become something other people rely on. A neat summary with invented deadlines is still wrong.
3. How do you use REDTEAM to find weaknesses in an idea?
Use REDTEAM when you want AI to look for ways a plan could fail. Give it the actual plan, including the constraints and evidence you already have.
Copy this:
Critically examine this plan. Identify three plausible failure points, the assumption behind each one and a small test that could investigate it. Separate evidence from speculation. Don’t invent objections just to fill the list.
Suppose your idea is an app for independent tutors. You have spoken to tutors who lose track of rescheduled lessons, but you don’t yet know whether they want another tool.
A useful critique would investigate adoption: will they enter the same information twice? Does their existing calendar already solve enough of the problem? Would a simple reminder template be sufficient?
Those questions can lead to an experiment. “The market is competitive” doesn’t give you much to do next.
AI can help you identify what to investigate. To find out whether people care, use the app idea validation process and collect real evidence.
4. Can TRUTHMODE make ChatGPT stop agreeing with you?
You can ask for a more critical response. But TRUTHMODE doesn’t guarantee truth, and being harsh isn’t the same as being accurate.
I would ask for a balanced assessment with reasons.
Copy this:
Assess this idea without assuming it is good. Explain what is promising, what is weak and what we don’t know yet. Challenge unsupported assumptions, including mine. Say what evidence would change your assessment.
The distinction matters when you are emotionally attached to an idea. You don’t need an AI cheerleader, but you also don’t need a dramatic takedown written for entertainment.
You need help seeing the decision clearly.
If the response makes a factual claim that matters, check it. Asking for confidence levels or caveats can make uncertainty more visible, but it doesn’t independently verify the answer.
5. Does /human make ChatGPT writing sound more natural?
It can suggest the style you want, but “human” is vague. Humans write everything from tax notices to group-chat messages.
Give AI a sample of your writing and describe what you want preserved.
Copy this:
Rewrite the draft below using my writing sample as a style reference. Keep my meaning. Use plain language, varied sentence lengths and concrete wording. Remove clichés and unnecessary repetition. Don’t add personal stories, results or claims. Use no em dashes.
Then paste your sample and the draft in clearly labelled sections.
Read the result aloud. Would you say it? Does it still sound like you when it disagrees, explains something or asks the reader to act?
I care much more about those questions than whether a draft has been labelled “humanised.”
This connects to what I wrote in Stop adding roles to your AI prompts: spend your attention on the instructions that actually describe the work you need.
6. How do you use an 80/20 prompt to prioritise?
Use 80/20 as a request to focus on the actions likely to matter most. Don’t assume there is a measurable 80/20 split in every situation.
AI needs your constraints to make the prioritisation useful.
Copy this:
My goal is [goal]. I have [time, budget and resources]. From the options below, recommend the two actions most likely to move me toward that goal. Explain the tradeoffs, identify assumptions and tell me what to postpone.
For example:
I have four hours this week to investigate an app idea. I haven’t spoken to potential users. Should I build a landing page, interview people, choose a name or make a prototype? Help me choose based on what I need to learn first.
The best answer depends on what you already know. A prioritisation prompt should make that visible rather than produce another impressive list.
7. How do you use SOCRATES to learn through questions?
Define SOCRATES as a request for guided questions. Specify one question at a time, otherwise you may get a worksheet instead of a conversation.
Copy this:
Help me understand [topic] by asking one question at a time. Start by checking what I already know. After I answer, explain any misconception briefly and ask the next question. If I get stuck, give a hint or a concrete example.
You can use this when learning a technical concept, preparing a presentation or working through a product decision.
For example, have AI ask you about the user, trigger and desired outcome behind an app idea. Your answers may expose how much of the idea is still a feature list without a clear problem.
The value is in doing the thinking yourself with support. If you just need a direct answer, ask for one.
8. What does LINDYMODE mean in a ChatGPT prompt?
In these shortcut lists, LINDYMODE is used to request established approaches rather than whatever is currently attracting attention. Define the criteria more carefully if you want a useful answer.
Old doesn’t automatically mean good. New doesn’t automatically mean unreliable.
Copy this:
For this problem, start with established approaches that have a documented track record. Explain why each may fit my situation, where it falls short and whether current conditions change the recommendation. Link sources for factual claims where possible.
This can help when you are drowning in new frameworks and tools. It is especially useful for asking whether a familiar process could solve the problem before you add complexity.
For software features, prices or current product capabilities, still request up-to-date verification. An established idea and a current implementation are different things.
9. How do you make ChatGPT ask questions before answering?
Tell it to ask only for missing information that would materially change the answer.
Otherwise, you can end up answering a long questionnaire when you wanted help with one decision.
Copy this:
Before recommending an approach, ask up to three questions whose answers would materially change your recommendation. If you already have enough information, proceed and state any assumptions.
This is useful for questions like “Which tool should I use to build my app?” The answer should depend on what you want to build, where people will use it and how much control you need.
For a starting comparison, see the best vibe coding tools for building apps.
10. How do you use AI to make a first version smaller?
Define a shortcut such as SCOPE CUT to help you remove optional work while preserving the useful result.
This is particularly relevant when AI keeps suggesting features you never asked for.
Copy this:
Help me define a first version for [specific user] who wants to [outcome]. List the steps needed to complete that task. Separate essential features from things that can wait. Include what must work reliably. Don’t add extra audiences, integrations or outcomes without explaining why they are necessary.
Then ask:
If we had half the available time, how could we narrow the promise while still delivering a complete, useful result?
That is the direction I want for a first app. A smaller promise I can actually deliver.
Ship ugly and fast, but useful. For the product framework behind that approach, read Simple, Lovable, Complete and what features belong in your first app version.
Can you create your own ChatGPT shortcuts?
Yes. Define a label and the behaviour you want it to represent in the conversation.
For example:
In this conversation, when I write SCOPE CUT, review the current plan for one specific user and outcome. Identify what is essential, what can wait and what would break the core task if removed. Don’t build anything unless I ask.
Then use the label with the relevant plan.
Keep the full instruction somewhere you can reuse. Don’t assume a label defined in one conversation will automatically mean the same thing in a different conversation or tool.
You also don’t need a label at all. “Please shorten this to three bullets and keep the main limitation” is already a perfectly usable instruction.
What is the simplest way to get better answers from ChatGPT?
Give it the situation, the task, the constraints and the output you need.
Here is a reusable structure:
Context: Here is what I am working on and what I already know.
Task: Here is what I want you to help me do.
Constraints: Here is the time, budget, audience or scope that matters.
Output: Here is the format and level of detail I need.
Add a shortcut when it helps describe the task. Then read the answer and correct what is off.
If it is generic, add the missing context. If it is too broad, narrow the question. If it contains claims you can’t verify, ask for the sources and check them.
You are allowed to direct the conversation. You don’t have to accept the first answer because it arrived confidently formatted.
Frequently asked questions about ChatGPT codes and prompts
Are ELI5, TRUTHMODE and /human official ChatGPT commands?
Treat these viral labels as natural-language prompting shortcuts. This article does not present them as documented built-in commands. Their effect depends on how the model interprets the instruction and the rest of the conversation. Define the behaviour when it matters.
Do you have to type the code before the prompt?
There is no special activation position for a label you define in ordinary text. Put the instruction where it is easy to understand, and clearly separate it from any material you want analysed. For longer tasks, state the request explicitly rather than relying on a single label.
Do capital letters or a slash make a prompt work better?
Use them for readability if you like. They don’t turn a self-defined label into a verified feature. What matters in these examples is the meaning of the instruction, the context you provide and whether the resulting answer meets your needs.
Can TRUTHMODE stop ChatGPT from making things up?
No prompt can guarantee factual accuracy. Ask it to distinguish facts from assumptions, acknowledge uncertainty and provide sources where appropriate. Then verify consequential claims against reliable sources. A forceful tone should never substitute for evidence.
What is the difference between REDTEAM and TRUTHMODE?
As defined here, REDTEAM focuses on possible failure points and tests. TRUTHMODE requests a balanced assessment of strengths, weaknesses and unknowns. The labels themselves are informal. You can get more clarity by writing the specific kind of critique you want.
Does a prompt like HORMOZI turn ChatGPT into a business expert?
A name may steer the style or ideas in a response, but it doesn’t transfer someone’s experience or judgment into the model. For a useful business review, name the actual questions: the offer, audience, price, acquisition assumptions and evidence. Don’t mistake a recognisable tone for expertise.
Can I combine several prompt shortcuts?
Yes, if the instructions fit together. You might ask for a critical assessment followed by a three-bullet summary. Avoid piling on conflicting requirements, such as an extremely short answer and exhaustive detail. Decide what matters most for the current task.
Will the same prompts work with Claude or other AI tools?
The plain-language requests can be tried in other conversational AI tools. Don’t assume a label has a shared official meaning or that every tool will respond identically. Paste the full instruction and check the result in the tool you are using.
Do longer prompts always produce better answers?
No. Add details that change the task or define a useful result. Repetition, elaborate personas and unrelated instructions can make a prompt harder to follow. A short, specific request is a good starting point; add context when the response shows what is missing.
Can AI critique replace app idea validation?
No. It can help you identify assumptions and design questions. It cannot establish that real users will adopt or pay for your app. Use the critique to choose what to investigate, then talk to people and observe behaviour.
Pick one instruction that fits something you are doing today. Give AI enough context to help. Then judge the answer by whether it moves the work forward.
If you want to go beyond prompting and start building with AI, AI Power User is my live workshop for that next step.
Sources and further reading
- OpenAI: Prompt engineering, for guidance on instructions, examples and relevant context.
- Sabrina Ramonov: Secret Codes for ChatGPT, the social-media shortcut discussion that inspired this article’s topic. The prompts and examples above are original adaptations, not a reproduction of her list.
- Jason Cohen: Simple, Lovable, Complete, for the product framework referenced in the scope-cutting section.