The short answer
/ghost is a Claude prompt prefix that removes the writing patterns AI tools fall into by default. You add it before any text you want rewritten, like /ghost [paste your paragraph], and Claude rewrites the text without the tells that make AI writing recognizable.
The output reads like a tired human typed it at 11pm on a Tuesday. Slightly imperfect rhythm, no flowery transitions, no "I hope this helps," no balanced sentence pairs, no em-dashes everywhere. Just text that does not trigger your brain's "this was written by ChatGPT" detector.
I have run this prefix on maybe 400 pieces of writing across cold emails, blog posts, and Slack messages over the last year. The pattern is consistent enough that it belongs in every writer's default toolkit.
What gets stripped
After running Claude and ChatGPT daily for the last year, you start recognizing the AI rhythm. /ghost was built to remove these specific tells:
1. Em-dashes everywhere
Default Claude uses em-dashes constantly because it has been trained on writing that overuses them. Real humans use em-dashes once a paragraph, maybe. /ghost replaces most of them with commas, periods, or just removes them.
2. "I hope this helps" and similar closers
Claude likes wrapping up answers with friendly closers: "I hope this helps!", "Let me know if you have questions!", "Happy to clarify!". /ghost cuts them. The text just ends when it is done.
3. Balanced sentence rhythm
Default AI writing has this weird poetic rhythm where every paragraph alternates short and long sentences in a way that sounds polished but artificial. /ghost breaks the rhythm. You get text that flows like actual speech: uneven, occasionally clunky, but human.
4. "In conclusion" and signposting
"First, let me explain X. Next, we'll cover Y. Finally, we'll discuss Z. In conclusion..." Real writers do not talk like this. They just move forward. /ghost cuts the meta-commentary about what you are about to read.
5. "Furthermore," "Moreover," "Additionally"
These transition words are fine in academic writing but they are a dead giveaway when used by AI. /ghost replaces them with simpler joins ("and", "also", or just a new sentence).
6. Three-item lists where two would do
Claude has a deep love for groups of three. "Powerful, flexible, and easy to use." "Efficient, reliable, and scalable." /ghost cuts the third item if it is just there for rhythm.
7. Hedging like "perhaps", "could", "might consider"
Default Claude hedges everything. /ghost removes the softeners and lets statements land directly.
Before and after example
Here is the same paragraph before and after /ghost.
Before (default Claude):
In today's competitive landscape, leveraging AI tools effectively can be a game-changer for solo founders. By utilizing platforms like Claude, particularly when paired with thoughtful prompt engineering, you can streamline your workflow, enhance productivity, and unlock new levels of creative output. The key is to start small, iterate often, and remain open to experimentation. I hope this gives you a clearer picture of how to get started!
After /ghost:
AI tools can save solo founders a ton of time, but only if you actually learn how to prompt them. Start with one tool. Run real tasks through it. Pay attention to what comes back. The people who get the most out of Claude are not the ones who memorize prompt formulas. They are the ones who notice when the output is off and figure out why.
Same information. Same length. Completely different feel.
When to use /ghost
/ghost shines for any text where "sounds like a human" matters more than "sounds polished":
1. Cold emails and outreach
The biggest reason cold emails fail is they sound like every other AI cold email. /ghost makes your outreach blend in with messages from real humans. Reply rates noticeably go up.
2. Slack messages and DMs
Nobody types "I hope this finds you well!" in Slack. /ghost text fits the medium. Short, direct, slightly imperfect.
3. Blog posts (if you do not want them to feel AI-written)
More people are filtering out AI content because it is everywhere now. /ghost gets you past the filter. Your post does not prove it is human-written, but it does not immediately scream AI either.
4. Twitter and X posts
AI-generated tweets are the easiest to spot. Too polished, too balanced, too "value-driven". /ghost makes a tweet feel like something you would actually type with your thumbs while annoyed about something.
5. Customer support responses
Customers can tell when they are getting an AI response. /ghost makes the help feel personal, even when you used Claude to draft it.
When NOT to use /ghost
/ghost is the wrong tool for some writing.
Do not use it for technical documentation
Docs benefit from clarity, structure, and signposting: exactly the things /ghost strips out. You want "first, install X. Then run Y" in docs, not a casual flow.
Do not use it for legal text
Legal writing needs precision and standard phrasing. /ghost will make your terms of service sound informal in a way that might create ambiguity.
Do not use it for academic writing
The "polished AI rhythm" /ghost strips out is actually what academic writing wants. /ghost will make your paper sound like a Reddit post.
Do not use it on text you wrote yourself
/ghost is for rewriting AI output to sound more human. If you wrote the text yourself, running /ghost on it will inject fake roughness (small inconsistencies that do not reflect your voice) and make it feel less like you, not more.
/ghost combos that stack
/ghost gets more powerful when combined with other prefixes:
/ghost + /punch: humanize the writing AND make every sentence sharper. Best combo for cold emails./ghost + /trim: humanize AND cut the filler. Best for tightening Slack messages./ghost + /voice: humanize AND hold a specific tone for the whole conversation. Best for writing in your own voice./mirror + /ghost: clone someone else's voice from samples, then strip the AI tells from the result. The closest thing Claude has to a perfect ghostwriter mode.
Does /ghost actually fool AI detectors?
Mostly yes, but not always. /ghost reduces the obvious signals (sentence rhythm, transition words, em-dashes) but AI detectors look at deeper statistical patterns too: perplexity, burstiness, predictability of word choices. /ghost helps a lot with the surface-level tells but it is not a guaranteed bypass.
For most use cases (cold emails, Slack, casual blog posts), /ghost output passes any reader's smell test. For high-stakes use cases where being detected as AI matters (academic submissions, contracts), do not rely on it alone.
A small caveat
Here is the honest part: /ghost is not an official Anthropic feature. It is a community-discovered convention that Claude has learned to recognize because so many people use it. There is no documentation on Anthropic's site that says /ghost does anything. It just works because the model has seen the pattern enough times to recognize the intent.
This means it could change. Future Claude versions might handle /ghost differently. The community pattern is what makes it work, not a hardcoded behavior. Most of the codes I tested for the Cheat Sheet are like this: community conventions that Claude consistently picks up on. That is also why testing them matters more than trusting the folklore.
Try it now
Open Claude. Pick any AI-generated paragraph you have lying around (a draft email, a blog intro, anything). Run it through this:
/ghost [paste your paragraph]
Compare the before and after. If the difference does not make you immediately want to send the after version, the prompt is not right for this task. If it does, congratulations, you just unlocked one of the most useful prefixes in the Cheat Sheet.
Where the rest of this lives
/ghost is one of about 160 Claude prompt prefixes I have tested over the last year. Most of them are noise, roughly 5 shift real reasoning, and a handful of stacks (like /ghost + /punch) do more together than either alone. The Cheat Sheet is where the tested set lives, with before and after outputs on the ones that showed signal and the combos that actually stack. $15 for Full, $35 for Pro with the deeper analysis.
The free 75-page Claude guide covers the setup basics if you want the long version first: clskillshub.com/guide