How to get better output from your AI by giving it what it really needs.

Ever ask ChatGPT or Claude for a marketing email, only to get a bland, generic template that sounds like it was written by a robot from 1995? You try to refine the prompt, adding adjectives like "punchy" or "professional," but the output remains stubbornly mediocre. You know the AI is capable of more, but you can't seem to unlock it. That frustrating experience isn't a failure of your prompting skills; it's a failure of context. In fact, a recent Workday study found that for every 10 hours of efficiency gained from AI, nearly four of those hours are lost to correcting, verifying, and rewriting the generic output.

The AI doesn't know who you are, who your customers are, what you've written before, or what makes your brand unique. It's guessing. To get the specific, high-quality output you need, you have to stop focusing on finding the perfect prompt and start focusing on providing the right context. The good news is, there's a systematic way to do this, and a powerful tool that lets you do it faster than you can type.

AI LESSON

Stop wrestling with generic AI and start giving it the specific information it needs to do its best work.

This lesson teaches you the context-first method for getting better results from any large language model. No technical background is required — if you can write an email, you can apply this method today. The core idea is simple: the quality of an AI's output is directly proportional to the quality of the context you provide upfront. We'll cover the shift from prompt engineering to context engineering, how to use meta-prompts to provide that context, and how a voice-to-text tool like Wispr Flow can transform your workflow by letting you speak your context instead of typing it.

When you give an AI a simple prompt like "Write a blog post about the benefits of our new software," it lacks the critical information needed for a great response. It doesn't know your role, your audience, your goal, your brand's voice, or the specific facts and figures you want to include. Without this information, the AI defaults to the most average, generic patterns it learned from its training data. The result is content that is technically correct but practically useless.

This is not a prompting problem. It is a context problem.

The AI industry is moving beyond simple prompt engineering. The new frontier, as described in Anthropic's engineering blog, is context engineering — the practice of curating the entire universe of information the model sees. This includes not just the immediate instruction, but the system prompts, reference documents, examples, and conversational history.

Think of it like briefing a new team member. You wouldn't give them a one-sentence task and expect a perfect result. You'd provide background documents, style guides, and examples of past work. Context engineering means doing the same for your AI — creating a rich, informative workspace before you ever ask for the final output.

A meta-prompt is a structured context block you provide to the AI before asking for the final output. Instead of jumping straight to the task, you front-load all the raw materials the AI needs to succeed. A well-built meta-prompt includes:

Role & Goal: "You are the head of marketing for a B2B SaaS company. Your goal is to write a blog post that drives sign-ups for a free trial."

Audience: "The audience is non-technical project managers at mid-sized companies."

Key Points: "The post must cover three points: our software saves 10 hours per week, it integrates with existing tools, and it has a 98% customer satisfaction score."

Tone & Style: "The tone should be helpful and professional, but not overly formal. Use the Oxford comma. Avoid jargon."

Structure & Format: "The post should be 800 words, with an H2 title, three H3 subheadings, and a call-to-action at the end."

Raw Facts & Data: "Include this customer quote: 'This tool changed our workflow overnight.' Mention that we were founded in 2022."

Once you've provided this context block, then you give the simple instruction: "Based on the context above, write the blog post."

Here is a full, copy-pasteable example of a meta-prompt you can adapt for your own work:

Typing out a detailed meta-prompt for every task is time-consuming. This is where voice dictation becomes a genuine productivity advantage. According to research from the National Center for Voice and Speech, the average English speaker talks at around 150 words per minute, while the average typing speed is approximately 41 WPM. That's nearly a 4x difference.

Wispr Flow is an AI-powered voice-to-text tool that works in any application with a text field — Gmail, Notion, Google Docs, ChatGPT, Claude, and more. Instead of typing your meta-prompt, you speak it. As you talk, Wispr Flow transcribes your words in real time, automatically removes filler words, and formats the text. This allows you to perform a rapid brain dump of all the necessary context in a fraction of the time it would take to type.

Step 1: Download and install Wispr Flow. It runs in the background and activates with a configurable hotkey.

Step 2: Open your AI tool of choice (ChatGPT, Claude, etc.) and place your cursor in the chat input field.

Step 3: Activate Wispr Flow and speak your meta-prompt naturally, as if briefing a colleague. Cover your role, audience, goal, key points, tone, and any specific facts. Don't worry about perfect sentences — just get the information out.

Step 4: Wispr Flow transcribes your speech directly into the text field. Review the transcription for accuracy, then add your final instruction ("Now write the email.") and send.

Time: Creating a detailed meta-prompt by typing typically takes 5–10 minutes. With Wispr Flow, the same context dump takes 1–2 minutes of speaking.

Wispr Flow is available on Mac, Windows, iPhone, and Android. It operates on a freemium model:

All new accounts include a 14-day free trial of Flow Pro, with no credit card required.

Wispr Flow is a strong tool, but it has real constraints worth knowing before you commit:

It requires an internet connection. Wispr Flow processes audio in the cloud, so it does not work offline. If you work in environments with unreliable connectivity, this is a meaningful limitation.

The free tier has weekly word limits. At 2,000 words per week on desktop, the free plan is sufficient for occasional use but will constrain daily power users. A single detailed meta-prompt can run 300–500 words, so heavy users will hit the ceiling quickly.

Background noise affects accuracy. Like all dictation tools, Wispr Flow performs best in quiet environments. The "whisper mode" feature helps in shared spaces, but loud environments will reduce transcription accuracy.

Cloud processing raises privacy considerations. Audio is processed on Wispr Flow's servers. For sensitive business content, review their privacy policy and Zero Data Retention option before using it with confidential information.

Alternatives exist if Wispr Flow doesn't fit your setup. Apple Dictation and Windows Speech Recognition are free built-in options, though they lack the AI formatting and filler-word removal that make Wispr Flow particularly effective for meta-prompt dictation.

Step 1 — Right now: Download the Wispr Flow free version (this link will give you 1-month free) and activate it on your primary device.

Step 2 — This week: The next time you open ChatGPT or Claude, don't just write a prompt. Use Wispr Flow to speak a full meta-prompt first — your role, your audience, your goal, and the specific facts you need. Then give the simple instruction. Compare the output to what you normally get.

Step 3 — Next week: Identify one repetitive writing task in your workflow — status updates, client emails, internal reports. Build a meta-prompt template for it and save it as a Wispr Flow snippet. The next time you need that output, speak the variable details into the template and let the AI do the rest.

The shift from prompting to context engineering isn't a technical skill. It's a communication skill. You already know how to brief a colleague. Now you know how to brief your AI.

Want to go deeper on the ideas in this lesson? I recently gave a talk at the MarketingProfs AI Friday Forum called Beyond the Prompt: Creating Reproducible Marketing Workflows That Actually Work. It covers how to move past one-off prompts and build repeatable AI-powered workflows — a natural next step once you have the context-first method down. The slides are attached to the LinkedIn post.

I appreciate your support.