Claude Prompt Commands 2026: The Complete Guide to Getting Elite Output Every Time

Most people use Claude Prompt Commands the same way they use Google. They type a question. They read the answer. They move on. And they wonder why the output feels generic.

The difference between average Claude Prompt output and elite Claude Command output is not the model. It is the instruction. Prompt commands are the fastest way to close that gap. They are shorthand instructions that tell Claude exactly how to write, think, research, and structure every word it produces. One command changes the entire character of a response.

This guide covers every major prompt command, when to use each one, how to stack them for maximum effect, and the best practices that most people never figure out. Whether you are writing scripts, building content, running research, or producing anything that needs to be sharp and human, this is the system.

What Claude Prompt Commands Actually Do (And What They Do Not)

Before getting into the list, one thing needs to be clear. Prompt commands are not secret codes. They do not unlock hidden Claude capabilities or bypass anything. They are efficient shorthand for specific output instructions.

When you type /punch, Claude reads it as a direct instruction to tighten every sentence, remove soft language, and make the writing land harder. The result is identical to writing out that instruction in plain language. The command is just faster and more precise.

The commands that deliver the most consistent results are the ones with the clearest instructions embedded in the word itself. /trim, /ghost, /punch, /hook --->> each of these communicates exactly what it wants. The vaguer the command, the less reliable the result. This is why knowing what each command actually does is the difference between using them as a gimmick and using them as a real production system.

The Writing and Style Claude Prompt Commands That Change Everything

/ghost — The Most Important Command for Any Published Content

/ghost tells Claude to rewrite the output so it is impossible to identify as AI-written. This is not about fooling detection tools. It is about quality. AI-generated content has patterns. Sentence structures that repeat. Transitions that feel mechanical. A cadence that signals to any reader that a system produced it rather than a person.

When you apply /ghost Claude strips those patterns. The output becomes specific, human, and grounded in the voice of someone who was there rather than a system generating plausible text.

Use /ghost on anything that will be published under your name. Blog posts, LinkedIn content, scripts, emails, case studies. If a reader will see it, /ghost should touch it.

Best practice: Apply /ghost after the content is structurally correct. Get the ideas right first. Then run /ghost to make it sound like a person wrote it.

/punch — When Every Sentence Needs to Hit

/punch removes softness. It eliminates hedging language, weak verbs, passive constructions, and sentences that exist to fill space rather than deliver force. The result is writing that moves fast and lands hard.

Use /punch on anything where each sentence needs to earn its place. Sales copy, script openings, hook lines, bullet points in a resume, social media captions, and any content where losing the reader in the first three seconds is a real risk.

Best practice: /punch works best on short-form content. On long-form pieces, apply it section by section rather than to the entire document at once. A full essay written at maximum punch becomes exhausting. Punchy opening sections, followed by a slightly more measured delivery, give the reader room to breathe.

/trim — The Command That Cuts Without Losing Meaning

/trim removes everything that does not carry weight. Filler phrases, redundant sentences, unnecessary qualifiers, paragraphs that restate what the previous paragraph already said. The meaning stays. The excess goes.

Use /trim when a draft feels longer than it needs to be, but you cannot identify exactly what to cut. /trim finds it. Apply it to blog posts before publishing, scripts before recording, and proposals before sending.

Best practice: Run /trim before /ghost. Get to the essential content first. Then make it sound human. Trimming after humanizing forces you to redo the humanizing work.

/hook — When the Opening Line Is Not Working

/hook rewrites the opening line of any piece of content to grab attention immediately. It applies the principle that the first sentence must create a prediction error, contradiction, or curiosity so strong that leaving before the second sentence feels impossible.

Use /hook on any piece with a flat opening. Blog posts, video scripts, email subject lines, social media posts, and any content where the first line determines whether anyone reads further.

Best practice: Give /hook context about the audience. A hook that works for a creator audience lands differently than one built for a B2B executive. The more specific the context the stronger the hook.

/mirror — When the Output Needs to Sound Like You

/mirror takes a writing sample you provide and matches the style, cadence, vocabulary level, and sentence structure across all subsequent output. It is the most personalized command in the set.

Use /mirror when you need Claude to write in your voice rather than its own. Paste three to five paragraphs of your own writing at the top of the conversation, call /mirror, and every response will match your patterns.

Best practice: Give /mirror your most natural writing, not your most polished. The goal is to match how you actually write, not how you wish you wrote.

/flow — When the Draft Reads Choppy

/flow restructures text so it moves smoothly from one idea to the next without jarring transitions or abrupt jumps. It is the editing pass that makes a draft readable rather than just correct.

Use /flow after /trim. Cutting content can leave gaps in the logical progression. /flow fills those gaps and makes the remaining content read as a continuous thought rather than a series of disconnected points.

/polish — The Final Pass Before Publishing

/polish makes rough text sound professional without changing the meaning. It is the last pass before anything goes out. It catches awkward phrasing, fixes minor inconsistencies, and elevates the overall quality of the writing without restructuring it.

Use /polish as the final step in any content workflow. After /trim, after /ghost, after /flow, run /polish, and the output is ready.

/rephrase — When the Idea Is Right, But the Words Are Wrong

/rephrase delivers the same information in a completely different way. Use it when a sentence or paragraph is technically correct but the phrasing is not landing. It is also useful for creating variation when you need to say the same thing multiple ways across different platforms.

The Research and Analysis Commands That Go Deeper

/deepdive — When Surface Level Is Not Enough

/deepdive pushes Claude far beyond the obvious information on any topic. It goes beyond the first layer of what is commonly known, identifies the underlying mechanisms, pinpoints what most analyses miss, and produces research that actually informs decisions rather than just summarizing what everyone already knows.

Use /deepdive before writing any piece of content that depends on factual accuracy or expert-level insight. For documentary scripts, MadeUnseen-style content, long-form educational videos, and any piece where being wrong costs credibility, /deepdive is the required starting point.

Best practice: Never start writing content directly from a /deepdive output. Use it to build an informed outline. The research shapes the structure. The writing comes after.

/investigate — Research With Journalistic Rigor

/investigate approaches a topic the way a journalist chases a story. It looks for what does not add up. What the official narrative leaves out. Where the money went. Who benefits. What the primary sources actually say versus what gets reported.

Use /investigate for any content where the interesting story lives beneath the surface. MadeUnseen topics, documentary research, any piece where the hook is the gap between what people believe and what is actually true.

FACTCHECK — Verify Before It Goes Live

FACTCHECK reviews claims, statistics, and statements and flags anything that may be incorrect, unverifiable, or commonly misattributed. Before any statistic goes into a script or article, FACTCHECK it. A single wrong number in a published piece of content costs more credibility than the stat was worth.

Use FACTCHECK on any draft that contains specific numbers, attributed quotes, historical claims, or statements presented as fact.

/blindspots — Find What You Did Not Think to Ask

/blindspots surfaces the risks, angles, and missing information you have not considered. It identifies what your current framing assumes, what the obvious objections are, and what a critical reader would immediately question.

Use /blindspots before locking a content strategy, a script outline, or a business argument. It is the command that prevents you from publishing something that is immediately dismantled by the first person who reads it carefully.

DOSSIER — The Full Briefing Before You Write Anything

DOSSIER compiles a comprehensive briefing on any topic, person, company, or system. It is the research foundation that everything else builds on. Use it at the beginning of any major content project before a single word of the actual content is written.

/extract — Pull Exactly What You Need From Long Documents

/extract reads a long document and pulls only the specific information you need. Paste a research paper, a report, a transcript, or a long article and tell Claude what you are looking for. /extract finds it without making you read the entire document yourself.

The Creative and Content Commands Built for Retention

HOOKS10 — Ten Opening Lines for Any Topic

HOOKS10 generates ten distinct opening hooks for any topic or piece of content. Use it when you need options before committing to a direction. The best hook is rarely the first one. Having ten lets you identify which one creates the strongest prediction error and build from there.

Best practice: Give HOOKS10 the target audience alongside the topic. A hook built for curious adults who want to understand hidden systems is structurally different from one built for creators who want to grow their channels.

/storysell — Turn Information Into a Story People Stay For

/storysell converts information, arguments, or pitches into story form. It applies narrative structure to content that is currently just a sequence of facts or points. The result is content that people read to the end because they are following a story rather than processing a list.

Use /storysell on any content that currently explains rather than shows. Case studies, sales emails, pitch decks, educational scripts — anything where the reader currently receives information but does not feel it.

CLIFFHANGER — The Command That Makes People Come Back

CLIFFHANGER writes an ending that makes the reader or viewer desperate for what comes next. It plants a question, a consequence, or a contradiction that cannot be resolved without continuing.

Use CLIFFHANGER at the end of every section in a long-form script, at the end of every blog post that is part of a series, and at the end of any YouTube video where you want return viewers.

/angles — Ten Ways Into One Topic

/angles generates ten completely different approaches to the same topic. Use it before deciding on a content angle. Most creators and writers default to the most obvious angle. /angles reveals the nine other options, some of which will be significantly stronger than the first instinct.

/viral — Optimized for Sharing

/viral rewrites content so it is structured to travel. It applies the principles that make content shareable: a strong opinion, a counterintuitive claim, a specific and striking detail, and a payoff that makes the reader want to show it to someone else.

Use /viral on social media content, short-form scripts, and any piece where distribution depends on organic sharing.

POLARIZE — Make Your Take Bold Enough to Generate Response

POLARIZE makes a bold enough statement that people cannot ignore it. Not provocative for its own sake. Bold in the sense that it takes a clear position and defends it with force.

Use POLARIZE when the content is currently too safe. Safe content gets no response. Bold content that is grounded in truth and argued well generates engagement, debate, and shares.

The Thinking and Strategy Claude Prompt Commands for Decisions That Matter

/redteam — Tear Your Own Idea Apart First

/redteam finds every weakness in an idea, argument, plan, or piece of content before an opponent, critic, or market does. It is the command that prevents you from going public with something that has an obvious flaw you missed because you were too close to it.

Use /redteam on any major decision, content strategy, business argument, or creative direction before committing resources to it.

/premortem — Imagine It Already Failed

/premortem starts from the assumption that the plan failed and works backwards to identify why. It is more useful than a standard risk analysis because it forces specific reasoning rather than general caution.

Use /premortem before launching anything significant. A campaign, a product, a content series, a business decision. Knowing why it might fail before it launches gives you the chance to prevent it.

PARETO — Find the 20 Percent That Drives Everything

PARETO identifies the 20 percent of actions, content types, or strategies that produce 80 percent of the results. Use it when a strategy has too many components and you need to know where to focus.

/scenario — Play Out Multiple Futures

/scenario runs multiple versions of how a situation could unfold and identifies what to prepare for in each case. Use it for content strategy planning, business decisions, and any situation where the outcome depends on variables you cannot fully control.

How to Stack Commands for Maximum Effect

Single commands change the output. Stacked commands transform it. The key is applying them in the right order.

The standard production sequence for any published content:

1. Start with /deepdive or DOSSIER to build the research foundation.

2. Use /blindspots to identify what the research missed.

3. Run FACTCHECK on any specific claims before the outline is locked.

4. Build the outline and apply /angles to confirm you are using the strongest approach.

5. Write the draft with /hook applied to every opening section.

6. Run /trim to cut everything that does not carry weight.

7. Run /flow to make the remaining content move smoothly.

8. Apply /ghost to strip AI patterns and make it sound human.

9. Run /polish as the final pass before publishing.

For script writing specifically, add HOOKS10 before step four and CLIFFHANGER at the end of every section during step five.

For social media content: /punch, /hook, /ghost, /viral in that order. Nothing else is needed.

For research-heavy content: /deepdive, /investigate, FACTCHECK, /blindspots before a single word of content is written.

The Full Command Reference List

Core Writing and Style

• /ghost — Rewrites text so it is impossible to tell AI wrote it

• /mirror — Matches your exact writing style from a sample you provide

• /raw — Strips all formatting and gives a clean plain response

• /voice — Locks in a specific tone for the entire conversation

• /punch — Makes every sentence hit harder and more direct

• /flow — Restructures text so it reads smoother from start to finish

• /trim — Cuts the fluff without losing any meaning

• /hook — Rewrites the opening line to grab attention instantly

• /rephrase — Says the same thing in a completely different way

• /polish — Makes rough text sound professional without changing the meaning
  

Creative and Content

• /viral — Rewrites content optimized to get shared

• HOOKS10 — Generates ten opening hooks for any topic

• /storysell — Turns a boring message into a story people want to read

• REMIX — Takes two unrelated ideas and combines them into something new

• /angles — Gives ten completely different ways to approach one topic

• CLIFFHANGER — Writes an ending that makes people desperate for more

• /captionme — Writes social media captions that get engagement

• POLARIZE — Makes your take bold enough that people have to respond

• /banger — Turns a flat line into something people screenshot and share

• THUMBNAIL — Writes a title or headline designed to get clicks
  

Research and Deep Dives

• /deepdive — Goes far beyond the surface on any topic

• FACTCHECK — Verifies claims and flags anything that might be wrong

• /sources — Backs up every statement with where the information came from

• COMPARE — Puts two or more options side by side with a clear winner

• /investigate — Researches a topic like a journalist chasing a story

• TRENDSCAN — Identifies what is happening right now in any space

• /extract — Pulls the exact information you need from a long document

• TIMELINE — Maps out the full history of a topic in order

• /digest — Reads everything and gives you only what matters

• DOSSIER — Compiles a full briefing on any topic, person, or company
  

Thinking and Strategy

• /redteam — Tears your idea apart to find every weakness

• PARETO — Finds the 20 percent of actions that drive 80 percent of results

• /swot — Runs a full strengths, weaknesses, opportunities, threats analysis

• WARGAME — Simulates how competitors or opponents would respond

• /premortem — Imagines your plan already failed and figures out why

• LEVERAGE — Finds the one move that changes everything else

• /audit — Reviews your work and flags every issue it can find

• BOTTLENECK — Identifies the one thing slowing everything down

• /scenario — Plays out multiple futures so you can plan for each one

• BLINDSPOT — Surfaces the risks and angles you have not considered yet
  

Analysis and Learning

• OODA — Analyzes a situation through observe, orient, decide, and act

• /deepthink — Forces Claude to reason through every layer before answering

• L99 — Pushes the response to the highest possible level of depth

• CHAINLOGIC — Walks through each reasoning step so you can follow the logic

• /unpack — Breaks a complex idea into every piece that makes it work

• INVERT — Solves the problem by thinking about it completely backwards

• /layered — Gives the answer at surface level, mid level, and expert level

• XRAY — Sees through the obvious answer to find what is really going on

• /teachme — Turns the response into a lesson instead of just an answer

• MASTERCLASS — Teaches a topic like a world-class expert would
  

Power Commands

• /godmode — Unlocks the most detailed and comprehensive response possible

• /autoprompt — Takes your rough idea and builds the perfect prompt for it

• MEGAPROMPT — Generates a long, detailed prompt that gets exactly what you want

• /chain — Runs multiple prompts in sequence, each building on the last

• PERSONA — Makes Claude respond as a specific type of expert

• CONTEXT — Loads background information so every response is more relevant

• PROMPTFIX — Takes a bad prompt and rewrites it so it actually works

• OPERATOR — Handles a task end-to-end without needing more input

• SENTINEL — Reviews everything for errors, risks, and missed details

• CEOMODE — Responds like a CEO making high-stakes business decisions
  

The One Thing Most People Get Wrong About Prompt Commands

Commands do not compensate for a bad brief. If the input is vague, the command improves the style of a vague output. It does not fix the underlying problem.

The highest leverage move is to be specific before applying any command. Tell Claude exactly what you need, who it is for, what it must accomplish, and what it must not do. Then apply the command. The combination of a precise brief and the right command produces output that is genuinely difficult to improve on.

The second most common mistake is applying too many commands at once. Stack them in sequence rather than listing ten commands in a single prompt. Each command works better when it operates on output already shaped by the previous one.

The third mistake is using commands as a substitute for thinking. /godmode does not know what you want to say. It only knows how to say it with maximum depth. The thinking still has to come from you. Commands amplify direction. They do not replace it.

Use them right, and the output speaks for itself.