ChatGPT Prompts That Work: The Meta-Prompt, the 6-Slot Template, and Sales Prompts

Bad prompts aren't a ChatGPT problem. They're a specification problem, and most people skip the specification entirely.

This post collects the three prompting frameworks I recorded videos about in 2023 with their examples kept together here. First, one meta-prompt that turns ChatGPT into your prompt engineer. Second, a six-slot template for any one-off task. Third, the sales prompts that cover outreach, objections, lead qualification, and CRM prep. You can adapt the structure to another model, but I have not rerun these examples across current Claude, Gemini, and Grok versions.

The master prompt#

Copy this exactly and paste it into a fresh chat:

I want you to be my prompt-creating expert. Your goal is to help me craft the best possible prompt for my needs. The prompt will be used by you, ChatGPT. You will follow the following steps: Your first response will be to ask me what the prompt should be about. I will provide my answer, but we will need to improve it through continual iterations by going through the next steps. Based on my input, you will generate three sections: A) Revised Prompt, B) Suggestions, C) Questions. We will continue this iterative process, with me providing additional information and you updating the Revised Prompt section, until it's complete.

That's it. That's "the one prompt to rule them all."

What happens when you run it#

ChatGPT responds with a single question: what should the prompt be about?

You give it something rough. Let's say you type: I want to create copy for my SaaS startup.

That's vague, and the model knows it. Instead of guessing, it generates three sections:

A) Revised Prompt, a cleaned-up version of what you just said, usually still incomplete but more structured.

B) Suggestions, variables that would make the prompt stronger. Things like: target audience, unique selling points, key features, tone and style, which pages need copy.

C) Questions, direct questions for you to answer before the next round.

You answer the questions. The model revises the prompt. Repeat until the prompt is specific enough to produce something real.

The before and after#

Vague prompt: I want to create copy for my SaaS startup.

ChatGPT responds with generic copywriting advice: define your value proposition, focus on benefits, keep it simple. Useful if you've never heard of copywriting. Useless if you want actual copy.

After the iterative build: The refined prompt specified a playful yet assertive tone, a target audience of entrepreneurs aged 22 to 30, a product that generates 10 business ideas, a co-founder matching feature, and a free trial offer.

The output from that prompt:

"Welcome to Idea Sparks, the place where brilliant business ideas come to life in 10 minutes or less. Calling all Risk Takers, go-getters, and success-driven entrepreneurs aged 22 to 30. Are you tired of staring at a blank page for hours on end trying to come up with the next big thing?"

That's a homepage headline and hook, ready to publish. Same tool, same model, completely different result because the prompt did the work of defining who this is for and what it needs to sound like.

Why this works when writing prompts from scratch doesn't#

Most people write prompts the way they'd text a friend: quick, contextless, assuming the other side can fill in the gaps. A model can't fill in those gaps. It defaults to the average of everything it's seen, which is why you get generic output.

The meta-prompt works because it systematically surfaces the variables that determine output quality: audience, tone, USP, context, format. You're not learning prompt engineering. You're just answering questions about your own project, which you already know the answers to.

The six-slot template for one-off tasks#

When you know what you want and don't need the interview, this template gets you there in one message. I recorded it as its own video, and it's the framework I reach for most often.

I want you to act as an expert in [Topic 1] and [Topic 2]. My goal/objective is to [describe your goal] for [target audience] in [optional: location or context]. I want you to create [desired output] in the format of [format]. Please use a [tone] tone. Keep in mind that [any additional details or constraints].

Six slots. Each one narrows what the model generates, so it spends less time guessing and more time being useful.

• Expert role. "Expert in blog writing and surfing" produces different output than "expert in marketing." This primes vocabulary and reference points before it writes a word.
• Goal. "Create a 400-word blog post" is a goal. "Help me with content" is not.
• Target audience. Age range, interest level, background. The model writes differently for surfers aged 15 to 50 than for corporate executives.
• Desired output. Name the deliverable: a blog post, a product description, an email subject line, a LinkedIn caption.
• Format. Paragraph count, word count, bullets, headers. Skip it and you get whatever default the model prefers.
• Tone. Playful, professional, conversational, technical. This is where your brand voice lives.

A filled-in example, for a surfing blog:

I want you to act as an expert in blog writing and surfing. My goal is to create a 400-word blog post about surfing in Costa Rica for surfers aged 15 to 50. Please create a playful blog post using surfer slang and include tips on the best surf spots in Costa Rica. The format should be a blog post, 400 words long, divided into three paragraphs. Keep in mind that the target audience is surfers aged 15 to 50 and the post should focus on surfing Costa Rica.

Yes, it's repetitive. That's intentional. Repetition in a prompt reinforces the constraints, and the output was three paragraphs with slang, solid hooks, and a spot list. Not "done," but "worth editing" instead of "worth deleting." That's the bar.

The self-interview trick. If filling the slots feels slow, paste the template and say: "I want to create a prompt, please ask me questions." The model runs through each slot as a question, you answer conversationally, it assembles the prompt, and you paste that into a fresh chat. The same slot logic works for visual output too. I use it when making diagrams and flowcharts with ChatGPT, where naming the format and the audience up front separates a usable chart from a mess.

The sales prompts: outreach, objections, qualification, CRM#

The third video was built for sales teams, and the pattern is different from the two above: give the model a base prompt, tell it to ask clarifying questions, then compound follow-ups inside the same conversation until the output is a shareable asset.

Start with context, not a perfect prompt. For cold outreach:

Craft a personalized prospecting email for [Name], demonstrating research and understanding of their needs. Ask me more questions if you need more information to write an effective email.

It comes back asking about the industry and the pitch. Answer (in the video: SaaS software for mining equipment companies) and you get a full cold email with a plausible discovery story, a subject line, and a value proposition you swap your details into. You're editing, not writing from scratch.

Objection handling is the highest-leverage use. Once product and industry are established in the conversation:

Identify common objections for [product/service] and develop persuasive responses to address them.

For the mining-equipment SaaS it surfaced "we already have a system that works," "we're worried about data security," and "our team isn't very tech savvy," each with a response. Then compound: "Make a list of five more objections." Then format: "Please format these objections into a table. First column: objection number. Second column: objection. Third column: response." Two prompts later a chat session is a spreadsheet your team can use.

Lead qualification and CRM prep. "Identify high-value leads from the following list, focusing on their potential long-term revenue," followed by a request for a scoring matrix, tells you where to spend call time this week. For CRM cleanup, paste raw customer data and ask for output in the exact format your CRM imports.

When no template fits. Ask directly inside the same conversation. Because the product and industry are already established, "how can I find qualified leads?" gets an answer about mining equipment and SaaS, not a generic B2B lecture.

If you'd rather skip the building, the ChatGPT Prompts for Sales pack covers all five categories, plus the master prompts for structuring inputs, for $14.

The only step that matters#

Paste the master prompt. Answer the questions honestly. Stop when the revised prompt is specific enough that you'd be satisfied seeing it as an assignment brief. For a task you'll repeat, fill the six slots once and save the result. For sales work, keep one conversation per product and let the context compound.

You don't need to understand how prompts work. You just need to know your product, your audience, and what you want the output to do. The model handles the rest.

If you want to package a refined prompt so you never rebuild it, how to build a Custom GPT covers that. For another way to compare model outputs, see my ChatLLM review.

Watch the full video on YouTube: https://youtu.be/gwFb8oAKb0Q

This post contains affiliate links. I only recommend tools I actually use.