Change One Setting and Claude Gets 10 Times Better

My name is Brandon Galey, and this is the Claw Genius Podcast, where I teach you the top growth hacks from Clawed Power users in short no-fluff episodes. Every episode has a companion sub-stack post that goes deeper with copy and paste prompts, meta-proms, or skills. That means you do not need to take any notes because everything is in the companion post that I'll link to in this episode's description. Today's companion post will have a complete set of copy and paste global instructions that the pros use that you can make your own in a couple of clicks. On top of that, I am including the meta-prompt that I use with my clients that analyzes their entire chat history and then delivers a set of customized global instructions. Let's go ahead and get into the episode. There is a small card that every person who works at the Ritz Curleton carries. On it are the rules of the house, how to greet a guest, how to answer the phone within a few rings, and what to do the moment something goes wrong. First, Schultzi, who co-founded the Ritz Curleton, did not want the quality of your stay to depend on who happened to be working that night. He wanted a guest in any city and in a year to feel the same care. So he wrote down the standard and put it in everyone's pocket. One line on that card still surprises people who hear it. Any employee from the general manager to the person cleaning out your room can spend up to $2,000 to fix a guest problem on the spot. No form and no manager's approval. Schultzi understood something that most managers miss. You don't get great service by reminding your employees to be great each morning. You get it by writing the standard down one time and letting it ride into every interaction, whether you are standing there watching or not. Okay, the vast majority of clawed users are not running a hotel, but they have a built-in instruction card that is read by clawed every time they start a chat. The sad thing is that almost nobody fills it in or they fill it in with the wrong information. When you open clawed's settings by clicking on your initials in the bottom left corner, you will find a field labeled "Instructions for Claude". Power users call it your "Global Instructions" because what you put there is not for one chat. It loads at the top of every conversation you ever start, today and a year from now. Most people leave it empty. An empty box means you are running on Claude's factory sitting. Here's the thing about a factory sitting. Pull a new Canon camera out of the box and it starts in auto mode. Auto mode is a careful compromise. It is built to take a decent photo for everyone, which means it is tuned perfectly for no one. And this is why a professional photographer does the first thing a casual photographer never does. They turn off auto mode and sit the camera by hand because they are not the average person taking the average shot. They are after a specific kind of shot because one could be a sports photographer and another could be a wedding photographer. The "Global Instructions" box is where you switch Claude out of auto mode. A few minutes there pays you back in every chat that follows. So let me walk you through the sitting's worth dialing in. The first sitting is the simplest and it shapes everything after it. Rob tailors every answer to who it thinks it is talking to. A founder, a marketing manager and a college student can ask the same question and should get a very different answer. So tell it who you are before you ask it anything. Your role, your business and the people that you serve. Claude will always read this first and adjust the depth and the examples to match who you are. A good example of this part of the instructions would be, "I'm Sam, founder of a six-person boutique branding studio. We help service-based small businesses with positioning, websites and brand voice. I do most of the marketing sales and client delivery myself." Those three sentences allow every answer to land already pointed at your world. The next two settings that I'll cover shape what comes back to you in a chat. When Claude edits your writing, its instinct is to sand you down. It drifts your words toward something flatter and more formal and quietly removes the part that sounds like you. You can shut that off by adding the following instruction. When editing my writing, preserve my voice and word choices, don't make it more formal or more polished. If a sentence has a real problem, explain what's wrong before you change it. The second setting that I'll give you buys back time on every long reply. Claude likes to lay out its reasoning first and reveal the recommendation at the end. The way a mystery saves the killer for the last page. You can turn this around to save hours each week by not having to wait through the mystery over and over again. Just add this to the instructions. Lead with your direct answer or recommendation in the first sentence. Put supporting reasoning in detail afterward so I can stop reading once I have what I need. These two settings ensure your writing voice is kept and in your chats the answer you seek will always be first. Now let's move on to the two settings that change what Claude does before a single word of output shows up. For anything bigger than a quick question, Claude fills the gaps you left out with guesses and writes around them. To keep this from happening on the important, deeper dive queries, you will add this to the instructions so that Claude will do a quick interview before it starts on the output. On any substantial writing or planning task before you start, ask me the questions you need answered to do it well. Then wait for my answers before producing the full output. The next sitting that add tells Claude to flag problems you didn't ask about. On its own, Claude answers exactly what you asked and stays quiet about the thing you should have asked. The fix is to add this to your instructions. If you spot a risk, an error, or a clearly better approach, flag it even if I didn't ask, if there's a relevant tool, feature, or method, I might not know about, mention it. With these two settings in place, it interviews you and it warns you. That is the line between a tool and a partner. The next five settings are for upgrading honesty and accuracy. This is the cluster that probably matters the most, so it gets the most room in this episode. Claude's default mode is warmth. It opens feedback with what is good and reaches the real problems late if it reaches them at all. For work you are about to ship, that politeness is a liability. The fix for this is adding the following. No flattery and no validation for its own sake. Don't open feedback with what's good. If my work or idea is weak, tell me directly and explain why. I'd rather hear it now than after I've shipped it. Next, guard against the cave. For the most part, the moment you push back, Claude tends to fold and agree with you. We don't want a tool that drops its position the second that you frown. Add this to stop Claude from caving in to your reactions. Push back when my reasoning is weak, my trade-offs are ignored or I'm missing something. Question my assumptions. Don't default to agreement and don't change your position just because I object unless I give you a real reason. The next instruction is the quiet one, the one that bites without warning. Claude says a hard fact and a soft guess in the exact same confident voice. A real number and a false number sound identical coming out. Here's the instruction for this fix. just clearly between what you know. what you're inferring and what you're unsure about. If you don't know something, say so. Never invent facts, stats, quotes, or your sources. After that, add this proof instruction that asks law to cite sources you can verify and to mark any claim if it is shaky by saying the following. When you state facts or stats, cite a source I can verify. Mark any claim you're not confident about, so I know to check it before I use it. And the last one of the honesty instructions is to fix the way grades. When you ask law to score your work, the number tends to drift up. The way a teacher who likes you rounds the grade. Add this to the global instructions for the fix. When I ask you to assess or score something, be critical and use the full range. Don't inflate. Lead with the biggest problems, then the smaller ones. The final two items of your global instructions are for Claude's ticks. A handful of words turn up again and again and readers have learned to treat them as a fingerprint left by AI. To fix this, give it this instruction. Never use words and phrases like these that can make my content get flagged as content being made by AI. Delve, leverage, robust, seamless game changer, unlock, dive into its worth noting in today's fast-paced world and at the end of the day. My final recommendation for an instruction to add to your global instructions is to retire the M-dash for good. This is the ultimate AI tell for anyone that reads blogs, social, or anything else you plan to create with Claude. Just round out your instructions with this. Don't use M-dashes. Use commas, periods, or parentheses instead. As I wrap up this episode, I wanna go back to the Ritz Carlton one more time. Schultzie did not make the Ritz grade by hoping each new hire would guess right. He wrote the standard down and then let it travel. Your global instructions box is your standard and it travels into every chat you will ever open. So here's the one thing for you to do this week. Open your sittings, find instructions for Claude and fill in three lines about your professional identity with your role, your business, and who you serve. Then copy and paste the rest of the global instructions I covered so the Claude starts using your standard. In the description of this episode, I link to the special substack companion post that has all of the instructions ready for you to copy and paste. In addition, I am including the meta prompt that I use with all of our clients that have been using Claude for at least 90 days. When you run this meta prompt, Claude reviews all of your past chats and gives you a personalized set of global instructions. And that will do it for this episode. Thank you very much for taking the time to listen and I'll catch you in the next one.