Generative AI has come to make our lives much easier by swiftly producing texts, code, or research information that would otherwise take longer to obtain. But there is a downside, though. The system needs to be properly guided.
It works more like the popular computer term, “garbage in, garbage out”. This simply means what you give is what you get. If the quality of your probe is poor or vague, the response of the system will also likely be poor.
To get the best out of generative AI, you may want to follow the steps below. It’s a bit lengthy but once you get used to it, it becomes a piece of cake.
Step 1: Properly state what you need
This part is very important, state what you want in very simple terms. This stage is the beginning and it needs specificity. Directly state your purpose and expectation. It is also important to mention your audience and the format that you expect. This will help the system to tailor its response to your need.
Avoid short or vague instructions like “write something about climate change.” Instead, say:
- “Write a 600-word article about climate change impacts on
agriculture.” - “Use a neutral tone and include subheadings.”
A clear goal tells the AI exactly what success looks like, saving you time later.
Step 2: Provide context and background
AI performs best when it understands the situation. Give it background details, such as product descriptions, industry focus, or previous examples. For instance:
- “Write a press-style summary for the Huawei Mate 70 Pro using these
specs.” - “Base the report on data from Q2 2025.”
Context helps the model stay accurate and prevents generic or irrelevant output.
Step 3: Pick the appropriate model that fits your needs
Don’t just jump on a model because its free and then lament when its output is poor. Different models are designed for different needs. Those who just want to write will not use the same model as coders.
Using the wrong model can limit your outcome. Here are some examples of generative AI types and what they do best:
• Text models – ChatGPT, Gemini, Claude
• Image models – DALL·E, Midjourney, Stable Diffusion
• Code models – GitHub Copilot, Replit AI
• Audio models – Suno, Udio, MusicLM
• Video models – Runway, Pika, Synthesia
• 3D or design models – Spline AI, Leonardo
Choosing the right model ensures your results are relevant, polished, and tailored to your specific goal.
Step 4: Write detailed prompts
Good results come from good prompts. Always give complete instructions that cover:
- The task:What you want it to do (e.g., write,
summarize, explain). - The format:
Length, tone, or structure. - The examples: What
style or outcome you prefer.
Example: “Write a 500-word neutral article about Samsung’s Galaxy Tab S10. Include display size, dimensions, and specs. Avoid promotional language.”
This structure gives the model a clear blueprint to follow.
Step 5: Break down complex tasks
If your request involves multiple parts, don’t do everything in one prompt. Divide it into smaller stages. For example:
- Ask for an outline first.
- Then request the introduction.
- Next, expand each section.
- Finally, ask it to polish the entire text.
Breaking tasks this way keeps the AI focused and ensures a smoother, higher-quality final result.
Step 6: Avoid very long texts and ask for multiple versions
Personally, I would recommend that the maximum number of words you request is 1000 words, preferably 500 words. The longer the text you require generative AI to write, the more likely it will pad your text with “rubbish”.
If your text is short, less than 400, then ask for multiple versions. For example, your probe could include “Produce three options for this text, and each option should use a different tone.”
You can compare the options and pick the one that best appeals to you.
Step 7: Review and refine the output
Treat the first result as a draft, not the final version. Read carefully and ask the AI to revise specific areas:
- “Make the tone more formal.”
- “Add more technical details about performance.”
- “Shorten the conclusion without losing meaning.”
Refining through follow-up prompts turns rough drafts into polished work.
Step 8: Verify information and sources
AI systems are very far from perfect. If you use it blindly, it will completely embarrass you. A lawyer in the US was recently fined $10,000 because he blindly used ChatGPT to prepare his case, and the system included legal citations that do not exist.
Therefore, there is a need to check and double-check whatever AI gives to you. Check the facts, data, and citations. Please, if any name is mentioned by the system, confirm the name. Even when you ask AI to produce references, it may produce fake references. Thus, verification is very important.
Step 9: Edit with human judgment
After swiftly writing the basics with AI, it is important to humanize the text with your own human judgment. In this stage, you will revise your text based on your findings. You may need to adjust the flow and maybe tone. In many cases, if you do this properly, the readability will be better. Ensure that you do the following
- Replace robotic phrasing with natural expressions.
- Remove redundancy or filler sentences.
Human editing ensures the final result sounds intentional and authentic.
Step 10: Save and reuse effective prompts
When you find a prompt that gives great results, save it. Create a small library of prompt templates for your common tasks. For example:
- “Article summary prompt.”
- “Technical explanation prompt.”
- “Data analysis prompt.”
Using templates makes your workflow faster and ensures consistent quality over time.
Final Word
While using generative AI, I recommend that you maintain good ethics and protect privacy. Do not share the personal information of somebody or company-sensitive information because AI gave it to you.
Source from Gizchina
Disclaimer: The information set forth above is provided by gizchina.com independently of Alibaba.com. Alibaba.com makes no representation and warranties as to the quality and reliability of the seller and products. Alibaba.com expressly disclaims any liability for breaches pertaining to the copyright of content.
