ChatGPT Secret Codes: Do They Actually Work in 2026?

12 min read

Here are the 6 codes

/human    write it like a person, not a bot
/expert   answer like a specialist in that field
/ceo      answer with a business and money lens
/viral    give me content angles built for reach
/blog     write it search-optimised for Google
/brief    explain it short, no padding

Came from Instagram? That is what you came for. Copy it. Now read the next 30 seconds, because there is one thing about these codes nobody mentions and it is the reason they sometimes stop working.

ChatGPT secret codes are short prefixes like /expert or /brief that you type before your prompt to switch the tone and depth of the answer. They are not official OpenAI commands. They work because the model reads that token as a role signal and adjusts, which also means they can drift or get ignored on a busy chat. The fix takes two minutes and I will show you exactly how to lock them in.

That last part is the whole article. Almost nobody covers it.

Key takeaways

• The six codes are shorthand prompts, not built-in features. Nothing is hidden in your account.
• They work because a single word can carry a full role instruction. That is real and it is useful.
• They are unreliable until you define them once, which is the step every "secret codes" post skips.
• ChatGPT does have genuine slash commands. Type / in a fresh chat to see the ones your account actually has.
• The same codes carry over to Claude, Gemini, and Grok because the mechanism is not OpenAI-specific.

Are ChatGPT secret codes actually real commands?

No. Let me be straight with you, because this is where most posts lie by omission.

There is no hidden /expert feature sitting inside OpenAI's code waiting for you to unlock it. When you type /expert explain compound interest, ChatGPT is not looking up a command. It is reading the word "expert" and inferring that you want a specialist register: denser vocabulary, fewer disclaimers, more precision.

It is a compression trick. One word standing in for a sentence you did not feel like typing.

So does that make them fake? Not at all. It makes them understandable, which is better. Once you know why a code works, you can invent your own instead of collecting other people's.

But it also explains the annoying part. Ever used one of these, got a great answer, then used it again the next day and got the usual generic mush? That is not you doing it wrong. Read on.

What does each code actually do?

Here is the real instruction hiding inside each one.

/human

Tells the model to drop the corporate register. Shorter sentences. Contractions. No "in today's fast-paced world."

It works, but only partly, because "human" is vague. Pair it with one detail and it gets sharper. /human write like you are texting a friend who is smart but busy.

/expert

The single highest-value one on the list. It swaps the model out of explain-to-everyone mode and into talk-to-a-peer mode.

Add the field and it doubles in strength. /expert as a tax accountant beats /expert alone every time, because "expert" without a domain is still a guess.

/ceo

Reframes the question around money, risk, and tradeoffs instead of features. Ask "should I build a mobile app" normally and you get a list of considerations. Ask it with /ceo and you get cost, payback period, and what you are giving up by doing it.

Genuinely useful. Also the one people over-trust. It is a framing device, not a financial adviser.

/viral

Pushes toward hooks, angles, and formats rather than polite balanced content. Good for a first pass when you are staring at a blank page.

Warning from experience: it produces the same five hook shapes over and over. Use it to break the block, then rewrite in your own voice. Everything I have published that did numbers got rewritten after this step, never shipped straight out of it.

/blog

Signals structure: headings, keyword placement, skimmable sections. It gets you the shape of an SEO post.

What it will not do is give you an angle nobody else has, and in 2026 that is the only part that actually ranks. Structure is table stakes now. If you want the deeper version of this, I broke down the full five-block prompt framework in this piece on AI prompt structure.

/brief

Underrated. It kills the intro paragraph, the summary, and the "I hope this helps." You get the answer.

The best use is on follow-up questions. Ask the big question normally, then /brief everything after it.

The codes are not magic. They are a shortcut for people who got tired of typing the same instruction twice a day.

Why do these codes stop working sometimes?

Because you are relying on the model to guess your definition of the word every single time.

/expert on Monday might mean an academic tone. On Thursday, in a long chat about marketing, the same token gets pulled toward whatever that conversation has been about. Nothing broke. The word was just never pinned down.

Three things push a code toward being ignored.

1. A long chat. The further back your instruction sits, the more the current thread outweighs it.
2. A conflicting request. /brief write a detailed 2000 word guide cancels itself.
3. No definition anywhere. The model interprets fresh each time, so you get variance.

Which brings us to the part that makes the difference.

How do you make your ChatGPT codes work every time?

Define them once, in writing, where ChatGPT reads them on every chat. Then they stop being suggestions and start being rules.

Open Settings, then Personalization, then Custom Instructions. In the "how should ChatGPT respond" box, paste this.

When my message starts with a shortcut, follow that rule for the whole reply.

/human   Plain, conversational language. Short sentences. No corporate
         filler. Contractions are fine.
/expert  Answer as a senior specialist in the relevant field. Assume I
         know the basics. Skip disclaimers. Be precise about tradeoffs.
/ceo     Frame around cost, revenue, risk, and opportunity cost. Give a
         recommendation, not a list of considerations.
/viral   Give 5 distinct content angles with hooks. Vary the format.
         No generic listicle shapes.
/blog    Search-optimised structure: H2 headings, one clear intent,
         answer in the first 100 words, FAQ at the end.
/brief   Answer in under 100 words. No intro, no summary, no offer to
         help further.

If two shortcuts conflict, follow the last one and say so in one line.

That last line matters more than it looks. It turns a silent failure into a visible one.

A 2026 detail that helps here: OpenAI raised the Custom Instructions limit from 1,500 to 5,000 characters in July 2026 for Plus, Pro, Business, Enterprise, and Education accounts. Free and Go plans stay at 1,500. That block above is roughly 800 characters, so it fits everywhere, but Plus users now have real room to add their own codes on top.

If you want the rules to apply to one body of work only, put them in a Project instead. Custom Instructions are the base layer that applies everywhere. A Project overrides them inside that workspace, which is how you keep a /blog that means one thing for client work and something else for your own site.

[Image: Screenshot-style graphic of the ChatGPT Custom Instructions panel with the shortcut block pasted into the second box, one arrow pointing at the 5000 character counter.]

Which slash commands are actually built into ChatGPT?

Some are real, and this is worth knowing because the real ones behave differently. They fire a feature instead of nudging a tone.

Type / in a fresh chat. A menu appears listing exactly what your account has. That menu is the only accurate list, because what you get depends on your plan, your platform, and whatever is currently rolling out.

There is no official public master list of every ChatGPT slash command. Anyone selling you one is selling you a snapshot of their own account.

Real ones you may see include the canvas trigger for the side-by-side editing surface, and /goal and /plan in the composer for setting a persistent objective the model works toward across a conversation. You also get @ mentions to pull a specific custom GPT into the current chat, which is closer to a real command than anything on the reel list.

The honest split is simple.

• Native commands open a feature. The interface visibly changes.
• Shorthand codes change how the model writes. Nothing in the interface moves.

Both are useful. Confusing them is how people end up frustrated.

Do these codes work in Claude, Gemini, and Grok?

Yes, and this surprises people.

The codes are not exploiting anything specific to OpenAI. Every modern chat model does the same thing with a leading role token, so /expert and /brief carry over to Claude, Gemini, Grok, and whatever launches next month.

Each platform has its own version of the "define it once" step:

• Claude has Projects with custom instructions, and Styles for tone.
• Gemini has saved info in settings plus Gems.
• Grok has custom instructions in settings.

Paste the same block into any of them. The behaviour transfers. That is the strongest evidence that this is a prompting pattern, not a hidden feature.

Where do people get this wrong?

Three mistakes, most common first.

1. Collecting codes instead of using them. People screenshot lists of 50 shortcuts and use none. Six is already more than most people need. Use two for a week.

2. Stacking too many at once. /expert /ceo /viral /brief gives you a confused answer. One code, one job. Two at most, and only if they do not fight.

3. Treating output as final. /blog gives you structure, not judgement. /viral gives you angles, not taste. The code does the boring 40%. You still do the part that makes it yours.

If you are writing for social after all this, our X post generator handles the format side, and you can check your length before posting so nothing gets cut off mid-sentence.

What to do in the next 10 minutes

• Paste the shortcut block into Custom Instructions. This is the step that makes all of it stick.
• Pick two codes you will actually use. /brief and /expert are the highest return for most people.
• Type / in a new chat and look at what your account genuinely has.
• Write one code of your own. /roast for honest feedback on your work is a good first one.
• Stop hunting for more codes. Six defined properly beats fifty saved in a screenshot.

Frequently Asked Questions

What are ChatGPT secret codes?

ChatGPT secret codes are short prefixes such as /expert, /brief, or /blog that you type before your prompt to change the tone, depth, or format of the answer. They are not official OpenAI features. They work because the model reads the prefix as a role instruction and adjusts its response style accordingly.

Are ChatGPT slash commands official?

Some are and some are not. ChatGPT has genuine built-in commands that open features, and you can see the ones available to your account by typing / in a new chat. Popular "secret code" lists circulating online are mostly community shorthand prompts, which work by interpretation rather than by triggering any hidden feature.

Why do ChatGPT codes stop working?

Usually because the code was never defined anywhere, so the model reinterprets it fresh each time. Long conversations also dilute early instructions, and conflicting requests like asking for a brief but detailed answer cancel each other out. Defining your codes inside Custom Instructions fixes most of this.

How do I create my own ChatGPT shortcut?

Open Settings, then Personalization, then Custom Instructions, and write a line such as "/roast means give blunt critical feedback with no encouragement." From then on the shortcut applies across your chats. As of July 2026 paid plans allow up to 5,000 characters there, which is enough for a full personal command set.

Do these codes work on Claude, Gemini, and Grok?

Yes. The technique relies on how language models interpret a leading role word, not on anything specific to OpenAI. The same prefixes work across Claude, Gemini, and Grok, and each of those has its own settings area where you can define the shortcuts permanently.

Paste the shortcut block into your Custom Instructions right now, then read the five-block prompt structure that makes every prompt after this one better.

Written by Adithya Guttha, Founder of Art of Code.