0:00 System Prompt One Habit That Changes Everything
0:25 Who This Is For PushLeads Background
0:53 Why Decent Answers Are Not Useful
1:28 Four Components of a Strong System Prompt
2:11 Role Definition Specificity Is the Key
2:41 What System Prompts Actually Do
3:10 Mistakes That Cost You the Most
3:43 Multi-Turn Conversations for Complex Work
4:58 XML Tags for Structured Output
5:37 Thinking Prompts for Strategic Decisions
6:18 Building a Customer Service Bot Right
7:34 Five-Part Prompt Formula You Can Use Now
Full transcript
0:00 System Prompt One Habit That Changes Everything
One habit upgrades every response you get from Claude Sonette 4 .6, and most business owners skip it entirely, write a system prompt before your first question. That single step assigns Claude a role, gives it your business context, and defines the output format you expect. This video breaks down exactly how, plus the most common prompting mistakes, and how to fix them fast.
0:25 Who This Is For PushLeads Background
Let’s get into it. I’m Jeremy Ashburn, the founder of PushLeads. If you’re one of the small business owners and marketers using AI tools out there, this channel is built for you. By 2008, my own sites were generating $20 ,000 a month in sales, and that’s when I knew this worked. So, let’s get into how to prompt Claude Sonette for marketing, customer emails, proposals and content, the prompt structure that works, and the mistakes that produce generic output.
This is part of our AI search visibility guide.
0:53 Why Decent Answers Are Not Useful
Most business owners open Claude, type of question, and get a decent answer. Decent is not the same as useful. The gap between a vague output and a polished deliverable almost always comes down to how the prompt was built. Claude is capable, but it is not a mind reader. It performs best when you treat it like a skilled contractor, tell it who it is, what the job is, what format you need, and what standards it must meet. Without that framing, you keep getting generic answers to specific problems. A strong system prompt has four components.
1:28 Four Components of a Strong System Prompt
First, role definition, tell Claude what expert it should operate as. Second, business context, and share enough about your company, so responses feel relevant. Third, tone guidance, how you communicate, so output matches your voice. Fourth, hard constraints, rules like avoiding certain phrases, sticking to a word count, or always recommending a next step. Real example for a plumbing company, you are a direct response copywriter with 15 years of experience writing for local service businesses. Avoid technical jargon, always end with a specific call to action. Our business is a plumbing company in Asheville, Deputy, North Carolina. That single block changes everything that follows.
2:11 Role Definition Specificity Is the Key
Assigning a specific role is the fastest single step improvement you can make to any Claude prompt. Compare these two. You are a marketing expert, broad, forgettable, versus you are a Google Ads strategist managing campaigns for contractors in competitive local markets. The second pulls the right knowledge, vocabulary ban, and structure. Specificity is the key. Even a short phrase like you are an experienced local SEO strategist shifts the quality of what you receive.
2:41 What System Prompts Actually Do
System prompts do one thing. Give Claude a specific role or behavior to follow for the entire session. They do not make responses longer. Length comes from the task and explicit word count instructions. They do not increase processing speed. That is model infrastructure. They do not reduce API costs. Longer system prompts can actually add token usage slightly. The goal is behavioral control, not formatting tricks. When you understand that, you use them correctly.
3:10 Mistakes That Cost You the Most
The mistakes that cost the most. Biggest, no system prompt at all. You type a raw question and every response starts from scratch. You pay for that in editing time on every output. Second, vague roles. You are a writer tells Claude almost nothing. Without domain focus, outputs feel like they could be for any business anywhere. Third, expecting a finished deliverable from one message. That is like handing a contractor a sticky note and expecting a finished room. Multi -turn conversations are the standard for complex work.
3:43 Multi-Turn Conversations for Complex Work
Multi -turn conversations consistently outperform single prompt attempts for detailed deliverables. Here is the flow. Set the system prompt with role, context, tone, and constraints. Provide task specific details. Ask Claude to confirm its understanding before drafting. This catches misalignment early. Review the first draft and give targeted feedback one section at a time. Restate the core goal at each new stage. Build section by section. By the end of a 10 exchange session, you can have a complete customized strategy document that would have taken hours on your own. Claude Sanne 4 .6 has a context window of 200 ,000 tokens, roughly 150 ,000 words per session. Most business documents fit comfortably. Claude’s projects feature lets you save a system prompt that loads automatically at the start of every session inside that project. Set at once and it carries across every session without manual reentry. For business owners using Claude daily, setting up project instructions for your top four use cases, writing, research, planning, and customer communication is the highest return setup step you can take.
4:58 XML Tags for Structured Output
One of the most effective techniques for longer deliverables is using XML tags to tell Claude exactly how to organize its response. Without structure, Claude may deliver one large paragraph requiring manual reformatting. With XML tags, you get organized labeled sections from the start. A basic request, please structure your response using these tags. Headline, Intro, Body, Call to Action. Claude follows that structure reliably once defined. No technical background needed. XML tags are simply labels wrapped in angle brackets typed directly into your message. Anyone who can type can use this.
5:37 Thinking Prompts for Strategic Decisions
When you need Claude to work through something complex, a competitive pricing analysis, a content gap audit, a market positioning decision, tell it to think step by step before responding. Ask for reasoning first. Include a line like before answering. Think through the key factors involved. Request a pros and cons list for decisions with multiple options. Ask Claude to walk through its logic and numbered steps. Set the stakes. Tell Claude what is writing on the decision. Thinking prompts work especially well for strategic planning and offer development where a well reasoned answer beats a fast one.
6:18 Building a Customer Service Bot Right
What is the right approach for building a customer service bot that only discusses your products? A system prompt that defines role, scope, and tone before any user interaction begins. Telling users to self -police does not work. Adding be professional to every message is inefficient. Setting high temperature increases unpredictability. Exactly wrong for a customer service bot. A well written system prompt handles scope, tone, and off topic redirects in one place. Add a scripted fallback like I can only help with questions about our products and services and test it yourself before it goes live. The same logic applies to a tutoring app where you want Claude to give hints instead of direct answers. A system prompt explaining the tutor role is the correct tool. Asking users to say please, changing font sizes, or scheduling messages on weekdays. None of that affects how Claude responds. A system prompt assigning Claude the role of a patient tutor and instructing it to withhold direct answers is the only tool that changes how Claude teaches. Tell it to ask guiding questions, offer conceptual hints, and reveal the full solution only after the student has made a genuine attempt.
7:34 Five-Part Prompt Formula You Can Use Now
A formula you can use right now for any business task. Start with context, who you are and what your business does. Assign a role. The specific expert Claude should act as. Describe the task. Exactly what you need produced. Specify the format, bullet points, numbered list, XML labeled sections, or a word count. Set your constraints, tone, topics to avoid, required endings like a call to action. Build a few prompt templates for your most common tasks and you will have a repeatable system that works every session. Every refinement raises the quality floor on every future output that uses it. Better prompts produce better content briefs, stronger calls to action, and faster production cycles. When you get polished, targeted drafts from Claude on the first or second pass, you spend less time editing and more time publishing. That advantage compounds over time, especially in competitive local markets. If you want to turn better AI outputs into actual leads, traffic, and revenue, the next step is understanding exactly where your current online visibility stands. PushLeads offers a free teardown so you can see the gaps and opportunities specific to your market before making any decisions. Get started at the link in the description or call 828 -348 -7686.
Learn how to write effective system prompts for Claude Sonnet to get better AI outputs for your small business.
Read the full written guide: Claude Prompting Guide for Small Business Owners 2025
Up next
12:39how do you get your restoration company cited by ChatGPT and Perplexity?
10:11Perplexity Local Contractor Recommendations: Show Up
8:17Is the AI Bubble About to Pop? What It Means for Your Marketing
14:01Why AI Search Ignores Your Contractor Business (Fix This)
14:30How does a fully automated AI video pipeline turn a topic into a published YouTube video?
8:17Do YouTube Videos Show Up in Google Search? Yes, and in AI TooKeep Reading
Want more customers from Google & AI search?
Get a free SEO audit of your site — see exactly what to fix first.
