500 ChatGPT Commands: The Complete Prompt Shortcut Reference

Most articles present these as hidden features that almost nobody knows about. They are not hidden and they are not features. ChatGPT has no command registry, and typing /eli5 into a fresh chat with no context is a coin flip — sometimes you get a simple explanation, sometimes a question asking what you meant. What makes commands reliable is a one-time definition block: a short list mapping each shortcut to the instruction it stands for. Paste that once, and the model treats the rest of the conversation as a lookup table.

To stop re-pasting that block, put your definitions in Custom Instructions under Settings and Personalization, or in a Project's instructions if you use Projects. Definitions stored in a Project apply to every chat inside it, which is the closest thing to a real command system you can build today. Keep any single active list under about twenty entries — beyond that the model starts blending similar commands together, and you lose the precision that made shortcuts worth using in the first place.

That constraint is the reason this list is organized the way it is. Below are 500 commands across ten domains, 50 each. It is a menu, not a configuration. Read the section closest to your work, take the ten you would genuinely use this month, and define only those. The other 490 are there for when the work changes.

Explain and Simplify

Start here. These are the commands that change how an answer is shaped rather than what it contains, and they pay off from the first use.

Writing and Editing

These work best pointed at your own drafts rather than at generated text. Most of them make writing shorter, and shorter is almost always better.

SEO

The search commands worth defining are the ones that encode a checklist you would otherwise retype every single time.

AEO and AI Search

Answer Engine Optimization is where the newest work is. These commands shape content so an AI engine can lift it cleanly and attribute it to you.

Ecommerce and Product

Store owners get more from a different set. Most of these are about removing doubt rather than adding persuasion.

Marketing and Copywriting

Positioning first, copy second. A command that generates headlines is worthless if the underlying claim has not been decided.

Social and Repurposing

One good piece should become ten. These commands are about extraction, not creation.

Research and Analysis

The most valuable commands here are the ones that make the model disagree with itself. Agreement is cheap.

Productivity and Workflow

These are the ones people discover last and end up using most, because they change how the work is run rather than what it produces.

Data, Code, and Automation

Useful whether or not you write code, because most of these are about making a machine explain itself.

How to make them stick

Write your definition block as one line per command in the form command equals instruction, with no extra prose around it. Short distinct names beat clever ones, because two shortcuts that overlap in meaning will collide and you will get the wrong behavior without knowing why. Test each new command twice on work you can already judge before you trust it on work you cannot.

Where commands fail

They degrade in long conversations as the definition block falls out of the model's effective attention, so re-paste it when outputs start drifting. They do not survive across chats unless stored in Custom Instructions or a Project. They break when a command's meaning depends on context the model cannot see, which is why /brandcheck needs your brand described first. And they cannot make the model know something it does not know: /sources will happily produce plausible-looking citations for facts it never learned, so verify anything that matters before you publish it.

None of this is specific to ChatGPT. The same definition block works in Claude, Gemini, or any other instruction-following model — you paste it into that tool's own system or project instructions and it behaves the same way. That portability is the point. You are not learning one product's feature set, you are building a small private interface to whichever model you happen to be using.

The real gain from commands is not speed on any single prompt. It is consistency: the same instruction produces the same shape of output every time, which is what turns scattered AI use into a repeatable process. Build your own list of ten from the tables above, define them in a Project, and add a new command only when you catch yourself typing the same instruction a third time. That habit is worth considerably more than any list of five hundred.