Last week a client sent me a screenshot from LinkedIn claiming /godmode unlocks an “uncensored Claude.” I tested it. Claude responded normally, like the words /godmode were never there. So I tested twelve more viral Claude.ai commands. Most did nothing. A handful genuinely changed the output. This guide separates the two for you — every command I could verify, what it really does, and which ones are folklore dressed up as cheat codes. If you’ve been hoarding cheat-sheet PDFs and wondering why your prompts don’t feel different, you’re in the right place.
By the end you’ll know exactly which Claude.ai commands earn space in your workflow, and which to ignore.
What Are Claude.ai Commands, Really?
Here’s the thing. The phrase “Claude.ai commands” gets thrown around to mean three very different things, and that’s where most of the confusion comes from.
First, there are real slash commands — the ones you trigger by typing / in the chat box. On Claude.ai web, the menu is short. On Claude Code (the developer CLI), it’s huge. People mix them up constantly.
Second, there are prompt prefixes like /ghost, L99, or PERSONA. These aren’t features Anthropic built. They’re text patterns the community discovered that nudge Claude into specific behaviours, because the model has seen those patterns in its training data. Useful, sometimes. Magic, no.
Third, there are flat-out fake codes — things like /godmode, /nofilter, /unlocked. They went viral on Instagram reels. None of them work. I’ll show you why later.
Anthropic launched Claude in March 2023, and the team has slowly added a real command system across its surfaces — claude.ai, the desktop app, the iOS and Android apps, plus Claude Code. Not every command works on every surface, which trips up most users.
The Real Slash Commands on Claude.ai (the chat interface)
If you click the + button or type / in the Claude.ai message box, a menu actually pops up. Anthropic’s own help center confirms this. The list is shorter than you’d expect from social media posts, but every item below does something concrete.
/search — pull live web results into the chat
Type /search [your query] and Claude pulls fresh information from the web. Handy when you ask about something past the model’s knowledge cutoff. I use it for stock prices, fresh news, and “did this product launch yet” kind of questions.
/artifact — open the side panel
Artifacts are the side window where Claude builds documents, code, dashboards, and small interactive apps. You can also just ask Claude to “build it as an artifact” in plain English. Both work.
/think (Extended Thinking toggle)
Not a slash command in the strict sense, but worth flagging. There’s a toggle that forces Claude to reason longer before answering. On Opus 4.7 it’s noticeable — you trade speed for depth. Turn it on for hard reasoning, off for quick lookups.
Project, Style, and Memory selectors
These aren’t commands so much as built-in shortcuts. Projects pin context across chats. Styles let you save a tone (formal, concise, your-voice). Memory, when enabled, remembers facts about you between sessions. So you stop re-explaining your job every Monday morning.
The official command list inside Claude.ai chat is genuinely small. That’s not a flaw. Anthropic deliberately kept the chat interface simple and pushed the heavy command system into Claude Code where it belongs.
The Famous /ghost Command — Does It Actually Work?
/ghost is the most-searched Claude command on the planet right now. So let’s settle it.
/ghost is not an Anthropic feature. It’s a community prompt prefix. What it does: when you paste /ghost [your AI-written paragraph], Claude tends to rewrite that paragraph in a more human-sounding way — fewer em dashes, no “I hope this helps,” less of that polished symmetric rhythm AI loves.
Why does it work without being “real”? Because Claude has seen thousands of forum posts, GitHub issues, and Reddit threads where people use /ghost as shorthand for “rewrite this so it doesn’t sound like ChatGPT.” The model picks up the intent. So the label triggers a learned behaviour, not a hidden feature.
Honestly, in my testing it works maybe 70% of the time. Sometimes the rewrite is genuinely better. Sometimes Claude just shrugs and produces the same kind of polished prose. The fix is to spell it out: “rewrite this so it sounds like a tired human typed it at 11pm — no em dashes, no friendly closers, vary the sentence rhythm.” Same effect, more reliable.
So /ghost works. Just not for the mystical reason people on YouTube claim.
Prompt Prefixes That Genuinely Shift Claude’s Output
Beyond /ghost, a small set of prefixes do produce measurable changes. Tested across hundreds of community trials and confirmed in my own.
L99 — prepend it to a question and Claude tends to commit to an opinion instead of hedging. Useful when you’re sick of “it depends” answers. Ask “L99 should I host on Vercel or Netlify for a Next.js portfolio site?” and you’ll often get an actual recommendation.
PERSONA: [specific expert] — this one is real and underrated. The trick is in the word specific. “PERSONA: Senior Postgres DBA at Stripe, 15 years, cynical about ORMs” works. Generic stuff like “act like an expert” barely moves the needle.
OODA — frames the answer using the Observe-Orient-Decide-Act framework. Good for incident response and competitive analysis. The military origin sounds cooler than the actual output, but the structure does help.
ULTRATHINK — pushes Claude into longer, more analytical responses. On Opus 4.7 with Extended Thinking already on, the difference is small. On Sonnet, it’s bigger.
/skeptic — this one I genuinely use weekly. Instead of answering, Claude challenges whether the question is the right question. Saved me from running an A/B test on 200 landing-page variants once. Worth its weight.
That’s about it for the prefixes that consistently produce different output. Everything else on the viral lists — /raw, /punch, /trim, /voice, BEASTMODE — works some of the time, fails some of the time, and you’d get the same effect from clearer plain-English instructions.
The Cheat Codes That Are Pure Folklore
Now the part nobody on Instagram wants to admit. Some “Claude.ai commands” simply don’t exist.
/godmode — does nothing. Tested in fresh sessions across Sonnet 4.6 and Opus 4.7. The model treats it as filler text.
/nofilter, /unlocked, /jailbreak — Anthropic does not have a hidden override switch. If a thread tells you these unlock an uncensored mode, the person posting is either confused or selling a $9 PDF.
L33T-speak and similar typo-tricks — old jailbreak relics from 2023. Patched.
ARTIFACTS as a typed prefix — Artifacts is a real UI feature, but typing the word in caps doesn’t summon it. The side panel opens based on what you ask Claude to build.
The pattern is depressing but consistent. Anything claiming to “unlock” Claude or bypass its guidelines is fiction. Everything that genuinely changes Claude’s output works because it’s clearer instruction, not secret code.
What About Claude Code Slash Commands?
Quick detour, because people Google “Claude.ai commands” and end up on Claude Code pages. Claude Code is the terminal CLI for developers. It has 50+ real slash commands — /clear, /compact, /cost, /model, /init, /memory, /agents, /diff, /effort, and many more. These are built into the CLI binary and genuinely change behaviour.
If you write code, Claude Code is worth installing. But none of those commands work on claude.ai chat. Different product.
How Claude.ai Stacks Up Against ChatGPT and Gemini
You probably want to know if Claude’s command system is better than the competition. Quick honest take.
ChatGPT has Custom GPTs, Memory, Canvas, and a richer slash-style menu inside the message box. More features, broader ecosystem, weaker reasoning on complex tasks.
Gemini has Gems and tight Google Workspace integration. Great if you live in Docs and Gmail. Its prompt-prefix culture is way smaller because Gemini’s training corpus has fewer of those community conventions baked in.
Claude wins on writing quality, code, and long-context reasoning. It loses on multimodal range and feature breadth. For our readers — solo creators, content folks, freelance SEO consultants — Claude’s smaller but cleaner command surface is usually a feature, not a bug. For deeper head-to-heads, see our full library of AI assistants compared across writing, coding, and research tasks.
Claude.ai Pricing in 2026 (Quick Recap)
Quick context, because plan tier affects which commands and tools you can actually use.
Free — Sonnet 4.5 access, basic chat, limited daily messages. No Opus, no Claude Code. Fine for trying things.
Pro — $20/month (~₹1,700) or $17/month annually. Unlocks Projects, Artifacts, Connectors, Claude Code on terminal, and a healthy message budget. The right tier for most AIDigitalBox readers.
Max — $100 or $200/month for 5x or 20x Pro usage and priority Opus 4.7 access. Overkill unless you’re using Claude all day.
Team — from $25/seat/month, minimum five seats. Adds admin controls and shared Projects.
For verifying current rates, check Anthropic’s official pricing page directly — Claude tweaks tiers more often than most SaaS.
Who Should Actually Care About Claude.ai Commands?
Three groups get real value here.
Content writers and SEO folks — /ghost plus a clear humanisation prompt is genuinely useful for cleaning up first drafts before they hit a CMS. If you’re still shopping around, our roundup of the best AI writing tools we’ve tested covers the alternatives worth comparing.
Researchers and analysts — /skeptic, OODA, and Extended Thinking save you from confidently wrong answers.
Power users on Pro or Max — the real value is in Projects, Memory, and Artifacts, not in viral prefixes. Set those up once and your workflow compounds.
Who shouldn’t bother? Casual users who ping Claude twice a week. The default chat is already great. Don’t overthink it.
Final Verdict on Claude.ai Commands
Here’s my honest take after testing every command and prefix I could find.
Claude.ai commands are mostly oversold on social media. The official slash menu is small and useful. About five community prefixes — /ghost, L99, PERSONA, /skeptic, OODA — produce measurable improvements when used right. The rest is noise.
Rating: 4 / 5 for the actual command system. One star deducted because Anthropic’s documentation around what’s official versus folk-knowledge could be much clearer. Most users are confused, and Anthropic isn’t helping.
Next step: open Claude.ai, type / in the chat box, see what your plan actually shows. Then bookmark this guide for the prefixes worth trying.
FAQs
Q: What is the /ghost command on Claude.ai? /ghost is a community prompt prefix, not an official Anthropic feature. Paste /ghost before AI-generated text and Claude tends to rewrite it in a more human-sounding way — cutting em dashes, polished closers, and balanced sentence rhythm. Works most of the time, fails sometimes. Plain-English instruction works just as well.
Q: Are Claude.ai commands like /godmode and /jailbreak real? No. These don’t exist as real Claude.ai commands. They’re viral folklore. Anthropic has no hidden override system that unlocks an uncensored Claude. Anyone selling you a “secret cheat code PDF” is, charitably, mistaken.
Q: How many real slash commands does Claude.ai actually have? The chat interface keeps the official slash menu deliberately small — search, artifacts, project switching, and a few utility options. Claude Code, the developer CLI, has over 50 built-in slash commands. The two are different products.
Q: What’s the difference between Claude slash commands and prompt prefixes? Slash commands are built into the product — you type /, a menu appears, the system handles them. Prompt prefixes like L99 or PERSONA are just text patterns the community popularised. Claude responds to them because it has seen them in its training data, not because they’re hard-coded features.
Q: Do Claude.ai commands work on the free plan? Most do. The official slash menu, Artifacts, and basic prompt prefixes work on Free. What you lose is Opus 4.7 access, Claude Code, Connectors, and higher message limits. Pro at $20/month unlocks the rest.
