In partnership with
Find Out What's Holding Your Revenue Back
Your business didn't plateau overnight. Growth slowed one bottleneck at a time.
But more hours at the office, more hires, and more marketing won't remove a hidden limitation. They just make it harder to spot.
The Revenue Ceiling Report reveals the four most common growth ceilings so you can identify what's quietly costing your business revenue and what to do next.
BELAY’s U.S.-based Assistants help build the systems and support that turn sustainable growth into reality.
Most people don't struggle with AI because they lack ideas. They struggle to explain those ideas in one useful prompt.
You may know exactly what you want. You want a report, article, analysis, business plan, image, research task, or spreadsheet. But turning that intention into precise instructions takes another kind of work.
There is a simpler approach.
Tell the AI what you want to accomplish. Then ask the AI to write the prompt for that task. Read the generated prompt, change anything that matters, and submit it back to the AI.
The AI becomes the first writer of its own instructions.
This approach fits current guidance from major AI developers. OpenAI's prompting guidance says users don't need rigid technical syntax. They can begin in ordinary language, describe the goal, add useful context, define the output, and set boundaries where needed.
That changes how prompt writing should be understood.
Prompt writing is becoming a conversation
Earlier prompt advice often treated prompting like writing a small program. Users collected formulas, role instructions, templates, and carefully arranged phrases.
Some of those methods remain useful for difficult tasks. But stronger models can often help build those instructions themselves.
Suppose you need a market research report.
Instead of spending twenty minutes constructing the perfect prompt, you could begin with this:
"I need a research report about how AI may affect accounting jobs during the next five years. Write a detailed prompt I can use to research this properly. Include source requirements, research questions, structure, fact checking rules, and output format."
The AI now has a different job.
It isn't researching yet. It is helping define the research task.
The resulting prompt might identify questions you forgot. It may request employment data, distinguish automation from job elimination, require recent sources, define geographic scope, and separate evidence from prediction.
You can inspect those instructions before any research begins.
That inspection matters.
Your Boss Will Think You’re an Ecom Genius
Optimizing for growth? Go-to-Millions is Ari Murray’s ecommerce newsletter packed with proven tactics, creative that converts, and real operator insights—from product strategy to paid media. No mushy strategy. Just what’s working. Subscribe free for weekly ideas that drive revenue.
The technique is sometimes called meta prompting
The basic idea is simple. One prompt is used to create or improve another prompt.
The first interaction focuses on task design. The second focuses on execution.
Anthropic's prompt engineering documentation follows a related idea. It recommends defining success criteria and having a first draft prompt before trying to improve that prompt. It also points users toward prompt generation methods when they don't yet have a useful first draft.
This means the user doesn't always need to arrive with polished instructions.
You can arrive with an intention.
The AI can help convert that intention into specifications.
For everyday work, this can remove much of the anxiety around prompt engineering.
Start by explaining the outcome
The first request doesn't need impressive wording.
Tell the AI what you're trying to produce.
Mention the audience. Explain where the result will be used. Add source requirements if facts matter. State what should be avoided. Tell it what a good result should contain.
Then ask:
"Based on everything I told you, write the prompt I should use."
That sentence changes the interaction.
You are asking the model to translate your messy human intention into instructions another model can follow.
OpenAI describes useful prompting around several practical components, including the goal, context, required output, and boundaries. It also says users can start in their own words and refine the response through follow-up messages.
So you don't need to memorize a universal prompt formula.
You need to describe the problem accurately.
Then make the AI review its own prompt
The first generated prompt may still contain weaknesses.
Ask the AI to inspect it before using it.
For example:
"Review this prompt for missing context, vague instructions, conflicting requirements, weak fact checking, and unnecessary wording. Rewrite it so another AI can execute the task reliably."
Now the model is acting as an editor.
This type of iterative improvement has research behind it. The 2023 Self-Refine paper tested a process where language models generated an initial response, produced feedback about that response, and then revised it. Across seven tested tasks, the researchers reported better results from iterative refinement than conventional one-step generation.
That paper focused on improving outputs rather than everyday prompt writing. Still, the broader lesson applies here.
The first version doesn't need to carry the whole task.
Feedback can become part of the workflow.
The human still decides what the prompt means
There is one trap in this method.
A beautifully written AI prompt can still contain the wrong assumptions.
Suppose you ask AI to create a prompt for evaluating a business. The generated version might prioritize revenue growth. Your real concern may be cash flow.
It might suggest American market data while your business operates in Bangladesh.
It might request forecasts when you only wanted verified historical evidence.
So the generated prompt needs human review.
Read it for intent, not grammar.
Check whether the AI understood the real question. Remove requirements you don't need. Add information only you know.
The AI can structure instructions. It cannot automatically know your unstated priorities.
Use one AI to challenge another when the task matters
For higher value work, you can separate prompt creation from prompt review.
Ask one model to create the prompt. Ask another model to criticize it. Then return the criticism to the first model for revision.
This creates a small evaluation loop.
You could write:
"Here is a prompt another AI created for my task. Find ambiguities, missing evidence requirements, possible factual risks, and instructions that could produce weak results. Suggest specific changes."
The purpose isn't to create the longest possible prompt.
Length is a poor measure.
A strong prompt contains the information that changes the result. Everything else adds reading without adding direction.
A prompt can become a reusable working document
This method becomes more useful for repeated tasks.
Suppose every week you prepare an industry newsletter.
Instead of starting again each week, ask AI to build the master prompt once. Include research standards, writing rules, audience, source policy, structure, prohibited language, fact checking, and final formatting.
Run it several times.
Then ask the AI to study the failures.
Perhaps the introductions are too general. Sources are too old. Headlines sound promotional. Sections repeat. Important opposing evidence gets ignored.
Feed those failures back into the prompt.
The prompt gradually becomes a description of how you actually work.
This is where prompt writing starts moving away from clever wording.
It becomes task design.
The easiest prompt may begin with an imperfect request
There is a strange advantage here.
You don't need to know how to write the final prompt before asking AI for help.
You need enough knowledge to describe your problem and enough judgment to inspect the result.
That is a much lower barrier.
Someone preparing a presentation can describe the audience and decision involved.
A journalist can describe the story and sourcing standard.
A teacher can describe the lesson goal and student level.
A developer can describe the feature, stack, constraints, and expected behavior.
The AI can convert that information into a more formal instruction set.
Then the person who understands the real situation checks it.
Prompt engineering becomes less about memorizing phrases and more about giving useful context.
Ask AI to write the instructions, then stay in control
A practical workflow can be very short.
Tell the AI what you need. Ask it to create the prompt. Ask it to review that prompt. Correct the assumptions. Then use the revised version for the real task.
If the output fails, don't immediately start again.
Show the failure to the AI.
Ask what instruction was missing. Update the prompt and run it again.
That feedback loop is where much of the improvement happens.
AI can now help write the instructions used to guide AI. The human role moves one level higher. You define the objective, judge the assumptions, and decide whether the result deserves to be used.
The next time you're staring at an empty prompt box, don't spend ten minutes searching for the perfect wording.
Describe the work first.
Then ask the AI what you should have asked.
This newsletter can't do the work for you. It can keep useful methods in view and bring new evidence into your routine. The progress still comes from what you test, reject, revise, and use outside this email. Over time, those small decisions matter more than collecting another folder of prompt templates.
Half your market is one app away.
Your business is already on Instagram, SMS, and web chat. But 52 million immigrants in the US rely on WhatsApp to connect with businesses they trust — not email, not phone calls.
Wati helps you show up on WhatsApp and every channel they use. Are you still not there?
