Among ChatGPT commands, the thing that changes your results the most isn’t long, complex prompts, it’s a few words used at the right moment. These words, which have recently been circulating on social media under the heading “ChatGPT’s secret modes” are actually just simple instructions that change the model’s tone and framing. There’s no backdoor, secret key, or special code involved! But when used correctly, they genuinely raise the quality of the answers you get.
In this article, you’ll see these 10 words one by one with real examples, and we’ll also address the real issue that nobody talks about. Beyond that: these commands change the style of the answer, but they don’t guarantee its accuracy. We’ll also explain what actually fixes accuracy, backed by official sources.
The “secret mode” myth: Why does it feel like it works?
Let’s first clarify the most critical point, because the framing circulating online is wrong: These words are not secret modes embedded inside ChatGPT.
Think of ChatGPT as a language model trained on a massive pool of data. When you type an expression like “truth mode” or “act as,” you’re not pressing a hidden button in the model. You’re simply telling the model which tone, which role, and which framing to use. So these aren’t “modes” in the technical sense, they’re requests for a style and a persona.
So why does it feel so effective? Because most people ask ChatGPT vague, framing-free questions and get an average answer in return. When you say “act as an SEO manager” it produces a more focused answer specific to that field. The difference is so noticeable that people interpret it as “I unlocked a secret mode.” What’s actually happening is much simpler: You told the model more clearly what you wanted.
Knowing this distinction makes you more credible and lets you truly understand why and how to use these words.
The 10 words that get you better answers
I’ve given each command below with an example prompt. You can copy and adapt them to your own query.
1. 80/20
Instead of listing every detail on a topic, it asks the model to focus on the critical 20% that delivers the result. You don’t waste time with long lists.
Example: “Using 80/20 logic in content marketing, tell me the 3 things that deliver results.”
2. lindy mode
It brings you not this week’s passing trend, but the proven method that has worked for years. The “Lindy effect” is based on the idea that the longer an idea has stayed alive, the longer it’s likely to endure. It can be especially valuable in producing SEO and content strategy.
Example: “Lindy mode for SEO: explain the methods that haven’t changed in 10 years and still work.”
3. eli5
Short for “Explain like I’m 5.” It simplifies a topic as if explaining it to a 5-year-old. Ideal when you’re learning a complex topic for the first time. The correct spelling is eli5; “eli10” is circulating incorrectly.
Example: “What is a backlink? Explain eli5.”
4. truth mode
The favorite of anyone tired of ChatGPT constantly flattering them! Because with this command, it drops the unnecessary approval and starts telling you what it actually thinks. In other words, it’s useful when you want honest feedback rather than blind support for your idea.
Example: “Truth mode on this content plan: where are its weak spots?”
Important note: This command sharpens the tone, but it doesn’t fix the accuracy of the answer. The model can state incorrect information with “brutal honesty” too. We’ll come back to this detail at the end of the article.
5. step by step
Instead of summarizing the job all at once, it asks the model to explain it in order, step by step. It prevents confusion in setup, process, and planning tasks.
Example: “Explain the Google Search Console setup step by step.”
6. act as
You give ChatGPT a specific expert role. Instead of a generic answer, you get one that’s specific to that field and more competent. OpenAI’s own guide also lists having the model adopt a persona as one of the ways to give clear instructions.
Example: “Act as an experienced influencer marketing consultant and prepare a brand collaboration brief for me.”
7. socratic
Useful when you want to learn the topic, not just get the result. Instead of giving you the answer directly, it makes you think by asking questions. This way it becomes a method that pushes you to learn while maturing an idea.
Example: “Question my content strategy using the socratic method; make me think by asking me questions.”
8. red team
It tries to find and knock down the weakest point of an idea, plan, or text. Ideal for seeing the flaws before you put something live and taking precautions.
Example: “Red team this campaign idea: find the 3 biggest risks.”
9. format as table
Instead of giving the answer as plain text, it asks the model to lay it out as a table. It significantly improves readability, especially in multi-dimensional content like comparisons, pros-and-cons lists, or plans. It lets you scan the same information at a glance.
Example: “Compare TikTok, Instagram Reels, and YouTube Shorts for short video; format as table: columns: platform, average reach, ideal video length, who it’s suitable for.”
A small tip: To get the best result from a table, specify which columns you want. Just saying “make a table” is enough, but when you define the columns yourself, the answer comes back much more useful and organized.
10. Persona assignment (e.g., “write like…”)
You borrow the style of an expert or a well-known figure. It changes the tone and perspective quickly. In the marketing world, for example, the Alex Hormozi or Gary Vaynerchuk style is often used.
Example: “Write the benefits of this product like Alex Hormozi, in a direct and clear voice.”
Note: When assigning a persona, be careful not to produce fabricated quotes in the voice of real people; the goal is to borrow the style, not to attribute statements to that person.
The real issue: These words don’t fix accuracy
Now let’s get to the part most “secret command” content skips. This is the part that sets you apart from the rest.
All of these commands change the style, framing, and focus of the answer. None of them guarantee the answer will be correct. Typing “truth mode” doesn’t stop the model from stating incorrect information in a confident voice. So you sharpen the tone, but the risk of hallucination (the model producing fabricated information) remains in place.
So what actually improves accuracy? There’s no need to guess here, because OpenAI’s official prompt guide states it clearly:
- Give clear and specific instructions
According to OpenAI, the ways to get a clearer result include giving more detail, having the model adopt a persona, and specifying the steps you want. A vague question brings a vague answer.
- Provide reference text
In OpenAI’s words: providing reference text helps the model answer with fewer fabrications, just as a sheet of notes helps a student on a test. If you have a source, data, or document at hand, put it inside the prompt; don’t expect the model to rely on its memory.
- Split complex tasks into parts
Again, OpenAI’s recommendation: break complex jobs into simpler tasks. Instead of asking for everything with a single giant prompt, dividing the work into steps lowers the error rate.
In short: A command like “step by step” makes the answer more readable and makes the chain of reasoning visible, which helps you a lot. But what delivers accuracy isn’t the magic of a command word, it’s the clear instruction, reference text, and task-splitting discipline you provide.
Frequently asked questions
1. Are these commands ChatGPT’s secret modes?
No. These aren’t secret modes that toggle on and off inside the model; they’re normal instructions that change the tone and framing of the answer. They contain no backdoor or special code.
2. Does “truth mode” stop ChatGPT from lying?
No. This command makes the tone more honest and direct, but it doesn’t stop the model from producing incorrect information (hallucination). The real methods that improve accuracy are clear instructions, reference text, and task-splitting.
3. Is it “eli5” or “eli10”?
The correct one is eli5 (“explain like I’m 5,” as if explaining to a 5-year-old). “eli10” is a variation circulating incorrectly online.
4. Do these words only work in ChatGPT?
No. These aren’t secret commands specific to ChatGPT; they’re requests for tone and framing that work across language models in general. That’s why you can use them with the same logic in other models like Claude and Gemini. The result just may not be exactly the same, because each model interprets in its own way — but their core functions are similar.
Sources:
OpenAI: Best practices for prompt engineering with the OpenAI API (official guide): https://help.openai.com/en/articles/6654000-best-practices-for-prompt-engineering-with-openai-api
OpenAI: GPT-4.1 Prompting Guide (OpenAI Cookbook): https://cookbook.openai.com/examples/gpt4-1_prompting_guide
