Many people use AI with a simple pattern: write a request, copy the answer, and then finish. This method is indeed quick, but it often results in text that sounds convincing without truly addressing the needs. The problem is not always with the AI model, but rather with a process that is too short.
For important tasks—writing articles, drafting proposals, summarizing documents, creating work plans, or preparing emails—it is safer if AI is not directly asked to produce a final version. Use a three-stage workflow: create an initial draft, ask AI to critique it, and then revise based on that critique.
Why is one prompt often not enough?
AI does not know the entire context that exists in our minds. When we write, "Create an article about productivity," AI has to guess many things: who the audience is, how formal the tone should be, what the article's purpose is, how long the text should be, and what perspective is desired.
As a result, the answer may be generally correct but not suitable for use. The text might be too long, too generic, lack examples, or make claims that need to be rechecked.
A staged workflow helps separate three different tasks:
- Drafting: generating initial material as quickly as possible.
- Critiquing: identifying weaknesses, assumptions, unclear parts, and unsupported information.
- Revising: improving the draft without losing the original intent.
With this separation, AI is not just asked to "produce something," but is also invited to check the quality of the output.
Stage one: ask for a draft, not a perfect answer
At the initial stage, do not overload the prompt with too many demands. Provide enough context, then ask for a working version that can still be improved.
Example prompt:
I am preparing an article for a general audience on how to use AI to plan weekly work.Article goals:- explain practical benefits- provide a simple workflow example- remind readers not to include sensitive dataCreate a draft of about 700 words in natural Indonesian that does not feel like an advertisement. Do not assume all readers understand technical terms.
Note that the prompt mentions the goals, audience, style, and limitations. You do not need to ask for a perfect sentence right away. What matters is that the draft has a clear structure and direction.
Stage two: turn AI into a critical editor
Once the draft is available, do not immediately ask AI to rewrite it. Ask it to review the draft against specific criteria. If you simply say, "Please improve it," AI tends to change many things without explaining the issues found.
Use prompts like this:
Review the draft below as a critical editor.Evaluate based on the following aspects:1. Does the opening directly address the reader's problem?2. Is the structure of the writing easy to scan?3. Are there claims that are too general or unsupported?4. Are the examples provided concrete enough?5. Are there any repetitive, unclear parts, or jargon-like phrases?6. Are the suggestions realistic for beginners?Do not rewrite yet. Create a list of findings with three sections: issues, why the issue is important, and proposed improvements.
The instruction "do not rewrite yet" is important. You want to get a diagnosis before moving to the revision stage. This way, you can see if AI truly understands the weaknesses of the draft or is just replacing words.
Do not accept all critiques automatically
Critiques from AI still need human assessment. AI can identify useful problems, but it can also suggest changes that do not align with the writing's goals.
For example, AI may find the article's opening too casual and suggest a more formal style. If your target audience is small business owners or lay users, a casual style might actually be more effective.
Use three simple questions when evaluating critiques:
- Does this issue truly interfere with the writing's purpose?
- Are the suggested improvements appropriate for the intended audience?
- Does this change enhance clarity, or just alter the style?
This section is a crucial point that should not be entirely handed over to AI. You remain the decision-maker, while AI serves as a second reader that can help identify overlooked aspects.
Stage three: ask for revisions with clear boundaries
After selecting relevant critiques, ask AI to revise the draft based on those findings. Include rules that protect what is already correct.
Revise the draft based on the following critiques I agree with:- the opening needs to be more direct- examples in the second section need to be made more concrete- remove repetition between the third and fourth sectionsMaintain:- warm and easy-to-understand Indonesian- length of about 800 words- structure with subheadings- a warning for users not to include sensitive dataDo not add new statistics or factual claims without clear sources.
The final instruction helps reduce AI's tendency to fill gaps with information that sounds plausible. For articles containing numbers, rules, product features, or recent developments, continue to conduct separate source checks.
Use a checklist before using the results
This workflow does not guarantee perfect results. However, a brief checklist can help prevent the most common mistakes.
- Purpose: Does this writing or plan truly address the initial needs?
- Context: Is AI using correct information, not assumptions?
- Clarity: Do readers know what to do after reading?
- Accuracy: Have facts, figures, names, and dates been verified?
- Privacy: Does the draft contain personal data, company secrets, or customer information?
- Human voice: Does the final result still sound like reasonable communication?
What does this mean for us?
Using AI productively is not just about finding prompts that sound smart. More importantly, it is about designing a process that makes mistakes easy to spot before the results are used.
Start with small tasks: important emails, meeting summaries, article outlines, or weekly plans. Ask AI to create a draft, instruct it to critique with clear criteria, and then carry out targeted revisions. After a few attempts, you will have a workflow pattern that can be reused for various types of tasks.
AI can speed up writing, but quality still depends on how we provide context, check results, and make decisions. These three simple stages make the process more organized—without having to make prompts long and complicated.
– Rio Yotto @rioyotto