Director, Learning Design & Development| PMIAsheville, NC, United States
With Generative AI, iteratively refining and optimizing prompts can lead to better AI-generated results. This may involve adjusting the specificity or clarity of the prompt to increase relevance and accuracy of results.
What examples do you have of how improving a prompt drastically changed the output quality? What specific changes did you make that led to the improvement?
Refining a prompt leads to added specificity and helpfulness. The response becomes less generic and more precise. Saving Changes...
Trisha PflugerCEO / Program Manager / Senior Project Manager | Juno Biomedical, Inc.Ca, United States
In my experience, the more vague a prompt, the more vague the response is going to be. On the other side of that, the more specific and focused the prompt, the more sharp the response can be with more focused content. The prompt into a GenAI makes ALL the difference. In reality, this is the same with humans, too, we just don't notice as much. For example, ask a human to “cook dinner” and you don’t know what you’ll get. IF you prompt the human to make a very specific dinner with specific ingredients and cook to certain allergens, etc., the outcome will be very different. AI is no different.
Refining a prompt has consistently transformed my GenAI outputs from vague and generic to highly relevant and actionable. For instance, rephrasing a request from “Give me a project update” to “Summarize key risks and mitigation plans for the AI integration project as of last week” resulted in far more targeted and useful insights.
Saving Changes...
Ming YeungAdjunct Professor & Acting COO/CPO/CRO (contract)| Blockchain Venture Capital Inc.Toronto, Ontario, Canada
In my (humble) view, prompt refinement is like tuning an instrument—you can play the right notes, but with the right adjustments, suddenly the music resonates.
Here's a clear example: A user once asked, “Give me some leadership advice.” The result was vague and generic—things like “communicate clearly” and “be authentic.” However, when we refined it to: “Act like an executive leadership coach. Give three practical tips for a mid-level manager leading a team through a digital transformation in a remote environment, with examples”—the output was vastly better. The response became precise, scenario-based, and actionable, with tailored examples like using virtual whiteboarding to promote inclusion during remote brainstorming. All from just adding role, context, and intent.
Another common one involves writing assistance. The prompt “Write a cover letter” often gives a bland draft. But: “Write a concise and persuasive cover letter for a junior data analyst applying to a Toronto-based fintech company. Emphasize curiosity, adaptability, and Excel/Python skills. Limit to 300 words.” This produced something clear, confident, and beautifully on-target. The shift? Injecting specific tone, audience, purpose, and constraints.
From what I’ve seen, the most dramatic improvements come when you:
-- Assign a role (e.g., “Act like a career coach”)
-- Add constraints (e.g., word count, tone, format)
-- Provide context (e.g., location, audience, platform)
-- State the purpose clearly (e.g., to persuade, to explain, to critique)
It’s like giving the AI a script, a setting, and a spotlight. Refinements may often work wonders; practice makes perfect as the cliché goes.
Refining prompts is one of the most effective ways to get useful and accurate outputs from GenAI. Providing a large, overly complex prompt often leads to confusion or hallucinated responses, as the model struggles to identify the core request. Instead, breaking the task into smaller, focused prompts—and using a step-by-step or chain-of-thought approach—allows GenAI to process information more clearly and deliver higher-quality results. This method helps users guide the AI more effectively and ensures each stage of the task builds toward a more reliable final output. Saving Changes...
There are framewoks to create prompt. This is part of the Prompt Desing discipline. Those that gave me and the initiatives where I was included are:R-T-F (Role-Task-Format), T-A-G (Task, action, goal), B-A-B (Before, after, bridge), C-A-R-E (context, action, result, example), R-I-S-E (role, input, steps, expectations).
Hello Sergio
Thanks for those additional prompt patterns – I have come to realise that I heavily use RTF, CREATE, CARE and RISE.
These patterns have really helped me better the outcomes of the prompt.
It is critical for us to understand how to break that large/big complex task into constituent parts and then critical as to how we create prompt for each of the constituent tasks.
I got better at this after multiple iterations and learning by failure. Saving Changes...
There are framewoks to create prompt. This is part of the Prompt Desing discipline. Those that gave me and the initiatives where I was included are:R-T-F (Role-Task-Format), T-A-G (Task, action, goal), B-A-B (Before, after, bridge), C-A-R-E (context, action, result, example), R-I-S-E (role, input, steps, expectations).
Booma Pugazhenthi - yes your callout is valid - it might become restrictive, constrained, stifle creativity, however, these are definitely something one can use to start off to work with LLMs and then expand once they are comfortable - atleast helps in increasing our problem solving approaches based on these "building blocks" Saving Changes...
This is a whole new world of GenAI, through prompt engineering one is able to refine the prompt to get a desired answer that can help to sky-rocket the results of one's question at the beginning. I realized that my thinking was only limited to one answer, however one is able to have a set of chain prompts that can lead to a conversation with the AI model and therefore result in a more refined answer. Saving Changes...