5 lessons on how to make artificial intelligence work better

Artificial intelligence (AI) has become an everyday work tool in marketing, communication, and sales. However, in practice, we see a very clear problem: although AI is widely used, the results are often superficial, generic, and difficult to apply when aiming for strong outcomes or real efficiency gains.

This is precisely why APG Media created the free AI prompt generator, PromptPro.

When building PromptPro, we relied on advanced prompt engineering practices and one very clear insight: the quality of AI output is determined not by the model, but by the process through which the task is given. Below are five lessons that guided us in developing this tool-and that you can apply yourself to better understand how to communicate effectively with AI tools.

1. The problem is not AI-the problem is an unclear task

Most poor AI results occur not because AI is “bad” or “stupid,” but because it does not understand what is expected of it. When a task is presented as a vague request (“write a text,” “come up with ideas”), AI automatically chooses the safest, most average response.

That is why the first principle we followed when creating PromptPro was simple: a prompt should not be a question, but a clearly structured task. It must define the goal, context, audience, and expectations for the result.

This is the foundation-without it, AI will always deliver a superficial outcome.

Example: a bad prompt vs. a good prompt

Bad prompt:

Write a Valentine’s Day advertising text.

Result you get from this prompt:
Love doesn’t have to be complicated. A gift doesn’t either. A Valentine’s Day offer for those who choose with both head and heart.

Why doesn’t this work?
AI does not know:

• who this ad is for,
  
  
• what outcome is expected,
  
  
• which emotion matters most,
  
  
• where the text will be used,
  
  
• what should be avoided.
  
  

As a result, it takes the safest route: generic statements, universal phrases, and an average tone that tries to appeal to everyone-and ends up appealing to no one.

Good prompt:

We are creating a Valentine’s Day campaign for a B2C e-commerce store.
The goal is to help the buyer quickly and confidently choose a gift for a loved one.
The audience is last-minute shoppers who are afraid of making the wrong choice and are looking for a safe solution.
Create three short ad copies for Meta ads.
The tone should be calm, reassuring, and emotionally warm, without aggressive sales language.
Avoid clichés (“the perfect gift,” “an unforgettable offer”) and avoid emphasizing percentage discounts.
At the end, evaluate each text: does it help the buyer feel confident about their choice? If needed, improve the wording.

Result you get from this prompt:

If you’re afraid of choosing wrong-choose a simple solution.
A gift that has often been chosen with a good feeling.

Evaluation:
✔️ Directly addresses the audience’s fear (“afraid of choosing wrong”).
✔️ “Good feeling” is emotional but not overly sentimental.
⚠️ The first sentence could be softened slightly to sound less strict.

Improved version:
If you want to avoid choosing wrong, it’s worth choosing a simple solution.
A gift that is often chosen with a good feeling.

Why does this work?
AI clearly understands:

• what we are doing,
  
  
• who it is for,
  
  
• what emotion we are aiming for,
  
  
• what a good result looks like,
  
  
• and that the first version is not final.
  
  

In this case, AI is no longer improvising-it is executing a task.

2. The first result is not the final one-it’s only the beginning

Another common issue is evaluating AI based on its first generated response. In practice, this turns AI into a fast text generator rather than a thinking assistant.

When creating PromptPro, we rejected this logic immediately. Good AI work does not start with the first text-it starts with the ability to evaluate and improve that text.

The key lesson here is simple: the first AI response should always be treated as a draft. Depth, consistency, and real value only appear when AI is forced to return to its own output and refine it.

3. Quality emerges when AI evaluates itself

One of the most effective modern AI workflows is based on a simple idea: AI should not only generate content, but also evaluate it itself. This means clearly instructing it:

• which criteria to use when assessing the result,
  
  
• where to look for weaknesses,
  
  
• what exactly needs to be improved.
  
  

We considered this principle fundamental when building PromptPro. Tasks are structured so that AI is encouraged to think not only as a creator, but also as an editor and quality controller.

This helps avoid situations where AI “spits out” a text and the work ends there.

4. Structure matters more than creative spontaneity

Another recurring issue we see is that every AI prompt is created “from scratch.” Sometimes it works. Other times, it doesn’t.

That’s why the fourth lesson is very clear: structure always beats improvisation.

What does “structure” actually mean in AI prompts?

When we talk about structure, we don’t mean technical templates or complex formulas. Structure is a consistent thinking framework that helps AI understand tasks with the same depth every time-regardless of who writes the prompt or what topic it covers.

A structured prompt means that each time you consciously answer the same essential questions, only with different content.

First: context and situation. Where will the result be used? In which campaign, channel, or business scenario? Without this, AI has no “stage” to operate on.

Second: a clear goal. What should this content do for the person reading it-calm them, inspire them, help them decide, guide their choice? Without a goal, AI creates “nice text,” but not functional content.

Third: a specific audience and their state of mind. Not demographics, but a real situation: the person is in a hurry, unsure, afraid of making a mistake, wants to show care but doesn’t want to take risks. This completely changes the tone.

Fourth: boundaries and direction. What is allowed and what is not? Which phrases should be avoided? What tone does not fit? What would be considered a poor result? This prevents AI from guessing and forces it to act deliberately.

Finally: quality expectations. What does a good answer look like? Should it be short or detailed, emotional or rational, focused on one scenario or multiple? Without this, AI has no criteria to evaluate itself.

PromptPro was created so that this framework appears automatically-so that every prompt, even when written in a rush, follows the same consistent foundation.

5. AI’s real value lies not in quantity, but in iteration

Finally, one of the most important lessons: AI is not meant to generate as much content as possible. It is meant to improve content quality through iteration.

Instead of dozens of average texts, far more value comes from a few iterated, reviewed, and refined versions. This is especially important in marketing, where emotion, clarity, and trust are critical.

This is why PromptPro is not a “fast content” tool. It helps create a process where AI becomes a partner-not a random phrase generator.

What does this mean in practice?

We created PromptPro because we saw a recurring problem: AI’s potential is only partially used. However, these five lessons apply regardless of the tool you use.

If you follow them:

• AI will better understand context,
  
  
• results will become more consistent,
  
  
• trial-and-error will decrease,
  
  
• real content quality will increase.
  
  

If you want to see how these principles work in practice, PromptPro is available for free here: https://prompts.apgmedia.lt/