ChatGPT's 'hidden codes' on TikTok: fact vs. fiction

You’ve probably seen a video on TikTok or Instagram: someone types /3dbillboard into ChatGPT, pastes in a photo, and gets a giant ad on a billboard in the middle of a city within seconds. Or /magazine cover, and there they are on a magazine cover. These videos give the impression that there’s a list of “secret codes,” a bit like cheat codes in a video game, that unlock hidden ChatGPT features.

The problem is that this whole hidden-codes story is largely a myth. Understanding why saves you time from hunting for the “magic formula,” and more importantly teaches you what actually works — whether for generating an image or for organizing your revision.

What actually happens when you type a slash

An article that reviewed several of these viral shortcuts, including /handwritten, is clear on this point: these tags aren’t official commands documented by OpenAI. When ChatGPT correctly responds to /handwritten, it’s not because some hidden mode got activated. It’s because the model interprets the word “handwritten” as a plain style instruction, exactly as if it had read the sentence “write this by hand, scribbled on a sticky note.”

In other words: the slash itself doesn’t activate anything. There’s no internal switch that flips because a / character appears at the start of a message. What the slash does, in practice, is help the model quickly spot that a style keyword is about to follow — a bit like bold text draws the eye on a page. The model then understands that keyword through natural language, not through some secret mechanism.

That’s exactly why /3dbillboard, /makeup, /magazine cover, or /product explosion work: they’re perfectly understandable descriptions of a visual style or staging, in plain English. You’d get a very similar result by writing directly “put this photo on a 3D billboard on a busy street, with nighttime lighting.” The slash just packages that instruction into a format that shares well in a short video.

The article sums it up: these shortcuts “aren’t official OpenAI features, and they aren’t hidden anywhere in the interface.” They spread because they work through natural-language understanding, not through some behind-the-scenes technical activation.

Why it still works (sometimes quite well)

Don’t throw these tags away just yet. If they produce good results, it’s because they pack a precise style instruction into very few words. It’s actually a good lesson in method: the more concrete and visual your instruction is (a setting, a light, a framing, a mood), the closer the result will be to what you imagine. The “hidden code” is just a writing shortcut, not a technical mechanism.

That also means you can invent your own. If you type /press-clip, explaining once what you expect (a visual styled like a newspaper clipping, a bold headline, a black-and-white photo), ChatGPT will understand the next time you reuse the word — as long as you repeat it within the conversation: there’s no memory of a “real command” that creates itself.

The real slash commands do exist — elsewhere

Real tools do exist where a slash triggers a precise, documented, reproducible technical action every time. Claude Code, Anthropic’s command-line tool, is a good example. Its official documentation describes two genuinely real categories of slash commands.

First, built-in commands like /help (show help) or /compact (shrink the conversation’s context): these are hardwired into the tool and always trigger exactly the same action.

Then, and this is the more interesting part, commands users can create themselves. The documentation explains that a simple Markdown file placed in the right spot (for example, a file named deploy.md in a dedicated folder) automatically creates a /deploy command that always behaves the same way, with instructions written once and for all. That’s very different from a TikTok tag: here, the command genuinely exists, it’s registered, and it produces a stable, verifiable behavior every time it’s called — not an approximate interpretation of a keyword.

What a real, useful revision command looks like

This is exactly the principle behind a skill: a real, named procedure, written once, that you can reuse as often as you like without having to re-explain what you want every time. You’re not typing a keyword and hoping the model guesses your intent correctly among a thousand possible interpretations: you’re triggering a precise, documented action, built for a specific purpose.

Key takeaways

• Viral tags like /3dbillboard aren’t hidden commands: they’re style instructions that the model understands in natural language.
• The slash doesn’t activate anything by itself; it just helps quickly spot the style keyword that follows.
• A precise, concrete instruction (setting, light, framing) works just as well, with no need for a “code.”
• Real, documented, reproducible slash commands do exist, in tools like Claude Code — built on instructions written once and for all.
• For revision, this principle of a named, reliable command is exactly what the fiche-revision skill uses.