transformar-prompt-em-agente-ia
Most copywriters spend hours adapting the same copy to different formats. VSL, reels, email, carousel, sales page — each has its own logic. The work repeats, deadlines pile up, and the client asks for one more format.
This post is two things at once: a product idea you can build and sell today, and a tutorial on how to write the prompt behind it. You leave here knowing how to write your own prompt from scratch.
The idea: an agent that turns 1 copy into 50 ads
The product is called Copy Adapter. The pitch is simple: the user inputs a finished copy — for a product, a service, a digital course — and the agent adapts it to any ad format, respecting the rules of each channel.
The "50 formats" is not an exaggeration. Inside Meta Ads alone there are variations by objective (conversion, reach, traffic), placement (Feed, Stories, Reels, Audience Network), duration (15s, 30s, 60s), and persona (top, middle, bottom of funnel). Multiply by Google Ads, email, and landing pages — and you easily have 40 to 60 versions derived from a single source copy.
Today this is done by hand, by a copywriter charging per hour or per package. The Copy Adapter automates the mechanical part — preserving the core message while adapting rhythm, CTA, and information density for each format.
It is not a from-scratch copy generator. It is an adapter. That distinction is what makes the product good: the original idea stays intact, only the packaging changes.
Why this idea sells
Real pain, large market, willing to pay.
Who feels this pain today: freelance copywriters who get "adapt this to 5 more formats" requests every week, traffic managers who need quick A/B test variations, infoproduct creators who have launch copy but no time to adapt it per channel, and agencies that would bill more if they stopped losing hours on mechanical work.
These four groups share one thing: they already know the work needs to be done. They are not being convinced of a new need. They just need a better tool.
Pricing that makes sense: between R$ 97 and R$ 197 per month for unlimited use, or R$ 27 to R$ 47 per pack of 10 adaptations. Agencies tolerate higher tickets — up to R$ 497 with white-label and per-client usage.
Recurring revenue potential: reaching 100 subscribers at R$ 97 gives you R$ 9,700 MRR with a single running agent. The Member AI Starter plan sustains this with positive margin from the first month.
The Brazilian performance marketing market spends billions on paid media. Every real of media spend requires copy tested across multiple formats. You are selling the tool that removes the bottleneck between the idea and the test.
What separates an amateur prompt from one that becomes a product
Amateur prompt: "adapt this text to an Instagram ad." The result is unpredictable. Sometimes good, sometimes terrible, always inconsistent.
Product prompt: structured in 4 blocks that guide the model to reason before responding, maintaining consistency across every use.
The 4 blocks are: Persona (who the AI embodies), Objective (what it should and should not do), Method (the reasoning steps before generating a response), and Format (how the output should appear). We will open each one using the real Copy Adapter example — a prompt developed by Guilherme Serraglio (@gui.serraglio, dvnlabs).
Read more about what a prompt is and how it differs from an agent before continuing.
Block 1 and 2: persona and objective
The first block defines who the AI is. Not generically — with surgical precision. The more specific the persona, the more consistent the output.
In the Copy Adapter, the persona is:
You are a senior copywriter specializing in direct response,
with over 20 years of experience adapting campaigns to
different media formats and digital platforms. You master
the principles of Gary Halbert, David Ogilvy, and Claude
Hopkins, and know exactly how to adjust rhythm, information
density, and call-to-action for each channel.
Notice what this persona does: it defines experience (20+ years), school of thought (direct response, concrete names), and specific competence (rhythm, density, CTA per channel). The model now has a reference framework for making decisions.
The second block is the objective — and this is where most amateur prompts fail. A good objective defines what to do and what not to do:
Your objective is to adapt existing copy to different ad
formats, preserving the core of the original message.
You do NOT create copy from scratch. You do NOT invent
new arguments. You adapt rhythm, structure, and CTA
so the original copy works in each format.
The "does NOT create from scratch" is fundamental. Without that instruction, the model tends to rewrite everything — which destroys the original copywriter's work. This constraint is what makes the product reliable.
Combine this with what you learned in system prompts for creators — persona plus objective is the foundation of any agent worth charging for.
Block 3: the step-by-step method that makes the AI think
This is the block that separates mediocre prompts from prompts that produce professional output. Instead of asking directly for the result, you force the model to reason before responding.
The Copy Adapter uses 4 explicit steps:
Before adapting, follow these steps in order:
1. ANALYZE the original copy: identify the central
proposition, the main proof argument, the implied
target audience, and the tone (urgency, curiosity,
authority, etc.).
2. UNDERSTAND the target format: what is the ideal
duration or length? What is the consumption context
(feed, stories, inbox, sales page)? What level of
audience awareness does this format reach?
3. ADAPT while maintaining the core: adjust rhythm
(shorter sentences for stories, longer for email),
information density (fewer details for top of funnel,
more for bottom), and CTA (click, save, reply, buy).
4. DELIVER the result in the format specified below.
This method works because it forces the model to build a mental model before generating text. Step 1 prevents it from ignoring parts of the original copy. Step 2 prevents it from applying one format's rules to another. Step 3 gives explicit adaptation criteria. Step 4 closes with the format instruction — which comes in the next block.
Without this block, the model skips straight to output. With it, output is 3 to 4 times more consistent — with no change to the AI model used.
This is chain-of-thought reasoning applied to product. Read more about creating your first agent to see how this reasoning applies in practice.
The fourth block specifies how the response should appear. Without it, the model decides on its own — and that means inconsistency from session to session.
In the Copy Adapter:
Deliver the result in this exact format:
ADAPTED FORMAT: [format name]
DURATION/LENGTH: [time in seconds or word count]
--- ADAPTED COPY ---
[complete copy, ready to use]
--- END OF COPY ---
ADAPTATION NOTES:
- What was kept from the original:
- What was adjusted and why:
- Test suggestion (if applicable):
Notice the "Adaptation Notes" at the end. That field does two things: it educates the user (they learn the agent's reasoning) and creates an audit trail (if the client wants to understand why the copy changed, it is documented). This turns the agent from tool into teacher.
(continues...)
Read more at memberai.pro/en/blog/turn-a-prompt-into-an-ai-agent.
Learn more: plans and pricing · about Member AI · real customer cases · full blog.