Many people use AI with a simple pattern: writing a question, hitting send, and then hoping to receive a perfect final answer. This method is quick, but often results in responses that are too general, miss the mark, or sound like generic writing.
The problem is not always with AI's capabilities. Often, we ask for too much in one step. The notes are not organized, the goals are unclear, the audience is not defined, yet AI is already asked to produce the final output.
A more effective workflow is to break the work into four stages: collecting materials, organizing information, developing results, and then reviewing them. This pattern can be used to create articles, prepare presentations, summarize meetings, draft proposals, or think through technical solutions.
Why is AI Better Used in Stages?
Imagine asking a coworker to create a report from a desk full of papers without explaining the purpose of the report. They might produce something, but it may not meet the needs. AI faces the same issue when given brief instructions without context.
Complex work has several layers. There are raw materials, goals, constraints, communication styles, and standards for results. If everything is mixed into one prompt, important parts may be overlooked.
With a staged approach, each conversation has a clearer task. AI does not just act as a "answer machine," but also as a tool to help unravel problems.
Stage 1: Input Raw Materials Without Forcing AI to Complete It
The first step is not to ask for a final write-up. First, input the information you have, even if it is still in the form of short notes, a list of ideas, interview results, or meeting transcripts.
The goal of this stage is to help AI understand the available materials. You can ask AI to categorize the information, identify incomplete parts, or highlight ambiguous statements.
Example prompt:
I will provide raw notes about a project. Do not create a final report yet. Your tasks are:to categorize the notes by theme,to separate facts, assumptions, and open questions,to highlight conflicting information,to mention important data that is still lacking.
After that, paste your notes. This instruction helps prevent AI from immediately filling gaps with assumptions.
Stage 2: Ask for Structure, Not Beautiful Sentences
Once the raw materials are organized, the next step is to create a structure. At this stage, do not focus too much on word choice. What matters more is the order of ideas and the relationships between sections.
For example, if you want to write an article, ask for an outline that includes the problem, causes, solutions, examples, and practical steps. If you are preparing a presentation, ask for a slide arrangement based on the audience's goals.
Example prompt:
Based on the categorized information, create three alternative structures for an article aimed at general readers.The first structure should be the simplest and most practical.The second structure should highlight case examples.The third structure should be suitable for readers who want to understand technical aspects.For each structure, explain its advantages and disadvantages.
Such commands give you options. You do not have to accept the first structure created by AI. In fact, comparing several alternatives often helps clarify the direction of the work.
Stage 3: Develop Results with Specific Constraints
Once the structure is chosen, only then should AI be asked to develop the results. This is where many users need to provide more concrete constraints.
Constraints do not mean making the prompt complicated. Constraints are a way to explain what kind of results can be considered successful. Some things to mention include:
- who the readers are;
- the main purpose of the writing or document;
- the desired length;
- the language style;
- things to avoid;
- output format;
- examples or data that must be retained.
Example follow-up prompt:
Develop the second structure into an article of about 900 words.Use natural and easy-to-understand Indonesian.Start with a problem that is close to the readers' experiences.Explain technical terms when they first appear.Include realistic examples, but do not fabricate statistical data.Use subheadings and practical step sections.If there is uncertain information, mark it with [needs verification].
The final instruction is important. AI can construct sentences very convincingly, but a convincing writing style does not automatically mean the information is correct.
Stage 4: Make AI a Reviewer, Not the Final Judge
The finished results still need to be reviewed. At this stage, use AI to look for weaknesses, not just to polish sentences.
You can request reviews from several different perspectives. For example, once for logical accuracy, once for clarity for general readers, and once to find overly excessive parts.
Example review prompt:
Review the following text as a critical editor. Do not just rewrite the entire text. Create a table with columns:potentially confusing parts,claims that require sources or verification,repetitions of ideas,sentences that sound too general,suggestions for improvement.
By asking for a list of issues first, you maintain control over the final decision. After reading the review results, you can determine which parts need improvement and which ones are intentionally retained.
Examples of Application in Daily Work
This workflow can be used for various needs. After a meeting, for example, do not immediately ask AI to create the final minutes. Ask AI to separate decisions, tasks, responsible parties, deadlines, and unanswered questions.
For technical work, first input the symptoms of the problem. Ask AI to categorize possible causes and list the information that needs to be collected. After that, only then ask for the order of checks. This method is safer than directly asking for one solution that may not fit the system's conditions.
For content ideas, use AI to compare several perspectives, identify the most relevant readers, and find information gaps. Do not just ask for "20 content ideas," as a long list may not help if it does not align with your goals.
Things to Watch Out For
The staged workflow is not a guarantee that every AI output is correct. There are several risks that still need to be considered.
- Incorrect information can become neater. AI may turn erroneous notes into paragraphs that sound professional.
- Assumptions can be taken as facts. Ask AI to highlight parts that are not supported by your information.
- Sensitive context can be carried into AI services. Remove personal data, company secrets, identification numbers, and unnecessary information before uploading materials.
- Results can be too uniform. Use AI to speed up the thinking process, but still add your own judgment, experience, and perspective.
What You Can Do Now
Choose one small task today—such as tidying up meeting notes or drafting an outline—and run it through four conversations or four stages: raw materials, structure, development, and review.
Do not measure success by how short the prompt you write is. Measure it by how few major revisions are needed after AI provides results. If the results are more directed, easier to review, and closer to your needs, it means the workflow is working.
AI is most useful not when asked to guess the entire job from one sentence. Its value emerges when you use it to break down large tasks into clear small steps—while important decisions remain in human hands.
– Rio Yotto @rioyotto