Some say that AI tools like ChatGPT, Deepseek, or Gemini are as close to magic as tech has ever gotten. Others are less impressed; for them, these things are little more than glorified search engines. But the truth may lie somewhere between these extremes. People who understand this tend to know how to get the best responses from generative AI.<\/p>\n
Read more about Tech<\/a><\/em><\/strong><\/p>\n There\u2019s more to using Artificial Intelligence than just requesting answers and getting them. The quality of output you get from it greatly depends on the precision of your requests. It\u2019s for this reason that the skill set called \u201cprompt engineering\u201d exists.<\/p>\n But this article isn\u2019t a course on prompt engineering. What we\u2019ll do here instead is to explain how to improve the output you receive from generative AI. Here are our recommendations:<\/p>\n Generative AI is not your friend or acquaintance. It has very little information about your background, way of thinking, or interests (unless you provide it with this information). So it lacks sufficient context to answer your vague questions to your satisfaction.<\/p>\n If you\u2019re going to collect definitive answers with this technology, your questions will have to be clear and specific. So, instead of asking:<\/p>\n You could ask:<\/p>\n The more detail you provide (goals, audience, constraints, etc.), the better the response you\u2019ll get from AI.<\/p>\n This is an extension of the first tip. Again, AI can only help you to the extent that you enable it. Quite often, the enablement that it requires is background information or context.<\/p>\n For example, if you want ideas for a project you\u2019re working on, you should describe your role, audience, the purpose, and key details. The following is an example of a well-written prompt for AI:<\/p>\n If you input a single, multi-part and complex request, your generative AI model may be \u201coverwhelmed\u201d by it and end up producing surface-level answers. To achieve decent results, you\u2019ll want to break the task into a series of steps.<\/p>\n Instead of asking:<\/p>\n Follow these steps:<\/p>\n You are more likely to get output with greater depth and accuracy this way.<\/p>\n Sign up for the Connect Nigeria daily newsletter<\/a><\/strong><\/em><\/p>\n The first response generated by AI may fall short of your expectations. But you can make it better by asking follow-up questions, correcting inaccuracies, or demanding revisions.<\/p>\n When your generative AI tool (e.g. ChatGPT or Claude) delivers a less-than-satisfactory answer, you could continue the session with this:<\/p>\n Make the previous response more concise, and add real-world examples<\/em>.<\/p>\n You should consider the expectations and preferences of your specific audience when you\u2019re trying to address them, even if it\u2019s with the aid of AI. Because this technology is heavily reliant on the quality of your input, you\u2019re better off including a style prompt when you want it to produce content.<\/p>\n Here are a few prompts that specify tone or style:<\/p>\n These types of instructions are useful sometimes and enable your AI model to supply output that fits your needs better.<\/p>\n You could see improved responses when you rephrase your requests or ask the same question from a different angle. Instead of asking:<\/p>\n You could try this:<\/p>\n As you use generative AI tools, make a mental note of what works and what doesn\u2019t. As you try different formats, tones, and strategies, identify the ones that deliver decent results and use them more often.<\/p>\n You could keep a record of successful prompts or templates for use in the future, so you don\u2019t have to recall them from memory every time you need them.<\/p>\n
\nBe Specific<\/strong><\/h2>\n
\n
\n
Include Background or Context<\/strong><\/h2>\n
\n
Break Down Complex Requests<\/strong><\/h2>\n
\n
\n
\n
\nRefine by Tweaking<\/strong><\/h2>\n
Use a Style Prompt<\/strong><\/h2>\n
\n
Try Different Ways to Phrase the Same Request<\/strong><\/h2>\n
\n
\n
Learn from Previous Experiences<\/strong><\/h2>\n
\n