There's a new genre of content taking over your feed.
It goes like this: a guru posts "100 secret ChatGPT codes nobody knows about." Or "hidden commands that unlock GPT's full power." Or "type this one word and ChatGPT becomes 10x smarter." The post shows a long list of shortcuts — /godmode, ELI5, DELTR, TLDR, X10THINK — each with a promise attached.
The pitch is always the same: AI tools like ChatGPT, Claude, and Gemini have secret built-in commands. Insiders know them. You don't. Buy in, save the post, unlock the machine.
Here's the problem. These codes don't activate anything. There is no hidden command system inside ChatGPT. No cheat menu in Claude. No developer console in Gemini that opens when you type the magic word.
And yet — and this is what makes the whole genre so slippery — some of these codes actually work.
Both things are true at the same time. The codes work, and the gurus are misleading you about WHY. Once you understand the real mechanism, you'll never need another "100 secret codes" list again — and you'll prompt better than the people selling them.
Let me explain.
What's Actually Happening When You Type a "Code"
ChatGPT is not a video game. It's a language model. It doesn't execute commands — it predicts what response fits your words.
So when you type ELI5 and get a beautifully simple explanation, nothing was "unlocked." No mode was "activated." The model simply recognized a phrase it has seen millions of times across the internet — "explain like I'm 5" — and did what people usually want when they say it.
That's the entire magic trick. There is no trick.
Every single "secret code" is just a compressed prompt — a shortcut for a full sentence you didn't feel like typing:
TLDR= "summarize this briefly"ELI5= "explain this in very simple terms"/formal= "rewrite this in a more formal tone"SWOT= "analyze the strengths, weaknesses, opportunities, and threats"
The model isn't obeying a command. It's interpreting language — the same thing it does with every message you send.
Which leads to three conclusions the guru posts conveniently skip:
Codes work because of training data, not programming. The model understands them only because humans have used those phrases publicly, millions of times.
If the model hasn't seen the pattern before, the code does nothing reliable. It just guesses.
A clear, specific sentence always beats a cryptic acronym.
Keep that last point in mind. It's the whole game.
The Three Tiers of "Codes"
Not all shortcuts are created equal. Every viral list mixes three very different things together — and that's exactly what makes them misleading.
Tier 1: Real shortcuts (widely understood, reliable)
These are phrases so common that every major model instantly understands them:
ELI5 — everyone on the internet knows this. So does the model.
TL;DR — universally understood as "summarize this."
SWOT — a real business framework. The model knows exactly what four boxes to fill.
Pros/cons — self-explanatory. Works every time.
These work across ChatGPT, Claude, and Gemini because they're not codes. They're vocabulary.
Tier 2: Lazy prompts (work sometimes, worse than plain English)
Things like /improve, /shorten, /formal, BULLETS, TABLE.
Do they work? Usually. But you're gambling on the model's interpretation. /improve — improve what? Clarity? Tone? Grammar? Length?
Compare:
/improve
vs.
"Tighten this up. Cut filler words, keep my voice, don't make it sound corporate."
The second one takes six extra seconds and gives you a 10x better result. Vague shortcut in, vague output out.
Tier 3: Pure fiction (made-up acronyms sold as insider knowledge)
DELTR. KILLCRITIC. X10THINK. LENSSTACK. /godmode.
Nobody trained the model on these. The ‘gurus’ invented them, wrote a definition next to each one, and packaged it as leaked internal commands.
Here's the irony: in the original article, the code only "works" because the definition is sitting right next to it. Paste DELTR into a fresh chat with no explanation, and the model will guess. Sometimes it guesses well. Sometimes you get word salad delivered with total confidence.
You're not unlocking hidden power. You're rolling dice.
But Wait — Some Concepts Underneath Are Genuinely Good
Here's where the viral lists deserve partial credit. Strip away the fake-code packaging, and some entries point at real thinking techniques:
Steelmanning — asking the model to build the strongest version of the opposing argument
First principles — breaking an idea down to base truths before rebuilding it
Red teaming — attacking your own plan to find weaknesses before reality does
Second-order thinking — "and then what happens?" applied to every consequence
Socratic mode — making the model teach through questions instead of dumping answers
These aren't codes. They're prompting patterns — and they're excellent. You just don't need an acronym to use them. You need a sentence.
The Shortlist: Shortcuts That Actually Earn Their Place
After filtering out the fiction, here's what I'd actually keep. Each one is short enough to type fast and clear enough that every major model gets it right:
For understanding things:
ELI5— dead simple explanation"Explain this using an analogy" — the fastest way to actually get a concept
"What am I likely misunderstanding about this?" — surfaces blind spots
For condensing:
TL;DR— instant summary"3 key takeaways" — forces prioritization instead of a lazy recap
"One paragraph, no fluff" — my personal favorite
For decisions:
"Pros and cons" — the classic
"SWOT this" — for anything business-related
"Steelman the opposite position" — the single most underrated prompt in existence
"What's the strongest argument that I'm wrong?" — same idea, plainer words
For pressure-testing ideas:
"Red team this plan" — brutal, useful
"What assumptions am I making here?" — catches the stuff you can't see
"What are the second-order effects?" — thinks past step one
For learning:
"Quiz me on this" — turns passive reading into active recall
"Teach me this Socratically — ask, don't tell" — slower, but it sticks
For writing:
"Make this sound human" — kills the robotic AI cadence
"Cut this by 50% without losing the argument" — better than
/shortenbecause it sets a target
Notice the pattern? The best "shortcuts" aren't acronyms. They're short, specific sentences. The specificity is the feature.
The Real Cheat Codes Do Exist — You Just Have to Build Them
Here's what the viral lists never tell you: official, reusable command systems exist. They're just not secret — and they're not universal. You define them yourself:
ChatGPT — Custom Instructions and Custom GPTs let you bake your preferences into every conversation
Claude — Skills let you package entire workflows into reusable instructions the model follows every time. Claude Code has actual, real slash commands
Any model via API — system prompts are the original "secret code," and they're fully in your control
The difference: a viral "code" hopes the model guesses your intent. A custom instruction guarantees it, because you wrote the definition.
That's the real upgrade. Not memorizing 100 fake commands — building 5 real ones tailored to your work.
The Bottom Line
There are no secret codes. LLMs understand language, not commands.
Common shortcuts (
ELI5,TL;DR,SWOT) work because they're everyday vocabulary, not magic.Made-up acronyms are less reliable than the plain sentences they replace.
The thinking techniques underneath (steelman, red team, first principles) are genuinely powerful — use the words, skip the acronym.
Want actual reusable commands? Build them: Custom Instructions, Skills, system prompts.
The people sharing "100 secret codes" aren't lying about AI being powerful. They're lying about where the power is.
It's not in the codes. It's in knowing exactly what to ask for.
