AI Prompting How-To: Stop Commanding and Start Collaborating
- Jul 29
- 9 min read
We have a communication problem.
AI prompting is the process of giving an artificial intelligence large language model the instructions, context, examples, and questions it needs to understand what you are trying to accomplish. We open ChatGPT, type a short command, and expect more accurate responses from AI without giving the model enough context about the goal, audience, or intended use.
“Write me a blog post.”
“Make me an image.”
“Build me a presentation.”
The AI produces something, but it is usually generic, incomplete, or different from what they imagined.
That does not necessarily mean the model failed. It often means the model did not have
enough context to understand the assignment.
In the first episode of How To AI, Jorge Hernandez sat down with Mark Quinn, Head of AI Operations at Pearl, to break down a more effective way to work with AI. If you use AI tools and keep getting generic results, this is for you: it explains practical AI prompting techniques, why context changes output quality, the difference between commanding and collaborating with AI, and Mark’s three-part framework for getting responses closer to what you actually want.
Mark’s advice is simple:
“Forget that you are talking to AI. Imagine the world’s best expert in crafting effective prompts just sat down next to you.”
That shift changes everything.
AI Is Not a Search Engine
A search engine is designed to retrieve information.
You enter a few keywords, scan the results, and look for the answer yourself.
AI can do more than retrieve information. It can help you think, plan, analyze, create, revise, and make decisions.
But many people still interact with it like Google.
They type one command and wait for the finished answer.
According to Mark, the difference between basic AI use and effective AI use often comes down to the difference between telling and asking.
Telling AI sounds like this:
Make me an image for my company.
Asking AI sounds more like this:
I have been asked to create an opening image for my company’s presentation deck. I am not a designer, and I would like your help developing the concept. What information do you need from me before we begin?
The first prompt gives AI a task.
The second gives it a role, context, an intended use, and permission to participate in the process.
That is the beginning of collaboration.
Talk to AI Tools Like You Would Talk to an Expert
Imagine a world-class designer walked into your office.
Would you immediately say, “Make me an image”?
Probably not.
You would explain the company, the audience, the purpose of the image, where it will appear, and what you want people to feel when they see it.
You might admit that you are unsure what the final design should look like.
You would ask for recommendations.
You would expect the designer to ask questions before beginning.
Mark argues that we should bring that same behavior into our conversations with AI.
The model cannot see the idea in your head. It only knows what you give it.
The more clearly you explain your situation, the more useful the response becomes.
Useful context can include:
Who you are
What you are trying to accomplish
Who the work is for
Where the final result will be used
What you already know
What you are uncertain about
What a successful result would look like
Any limitations, references, or requirements
This does not mean every prompt needs to be a perfectly structured essay.
It means the conversation should contain enough information for the AI to understand the assignment.
Commanding Produces an Answer. Collaboration Produces a Process for Complex Tasks.
When you command AI to complete a task, it usually begins immediately.
That can be useful when the task is simple.
But when the work requires judgment, creativity, strategy, or several decisions, immediate execution can become a problem.
The model starts filling in missing information on its own.
It guesses your audience.
It invents a visual direction.
It assumes your priorities.
It chooses a tone.
It may give you a polished result that solves the wrong problem.
Collaboration creates a different interaction.
Instead of asking for the finished output immediately, ask AI to help define the assignment.
You can say:
I need help creating a launch strategy for a new product. I have some early ideas, but I am not sure how to organize them. Help me think through the strategy before writing the final plan.
That prompt encourages AI to participate as an adviser rather than operate as a vending machine.
It can identify missing information, challenge assumptions, suggest approaches, and help you make decisions before producing the final deliverable.
Mark demonstrated this during the episode by asking ChatGPT to help design an image for Pearl. Instead of forcing the model to create the image immediately, he explained the intended use, acknowledged what he did not know, and asked what information the model needed.
The interaction changed from execution to collaboration.
Mark Quinn’s Three-Part Prompt Engineering Framework
Mark ends many of his first interactions with AI using three requests.
Summarize your understanding
Ask the AI to repeat the assignment in its own words.
For example:
Before we continue, summarize your understanding of what I am trying to accomplish.
This gives you a chance to catch misunderstandings early.
The AI may have interpreted your audience incorrectly, missed an important requirement, or focused on the wrong objective.
It is easier to correct the direction before the work begins than after a full draft has been produced.
Identify the expertise required
Ask the model what kinds of expertise would help solve the problem.
For example:
Identify the expertise and perspectives we should use to approach this effectively.
A marketing project might require brand strategy, copywriting, audience research, analytics, and creative direction.
A product project might require user research, technical architecture, operations, and go-to-market planning.
This step encourages the AI to approach the assignment from several relevant perspectives instead of producing the first obvious answer.
Ask clarifying questions
Finally, invite the AI to identify what it still needs to know.
For example:
Ask me any clarifying questions that would strengthen your understanding and improve the final result.
This is one of the most valuable habits in prompting.
You do not need to predict every piece of information the model might need. You can ask the model to surface the gaps.
Mark’s complete framework is:
Summarize your understanding, identify the expertise needed, and ask any clarifying questions required to optimize the support.
A Prompt You Can Use
Here is a simple starting template:
I am working on [describe the project].The goal is [describe the intended outcome].This is for [describe the audience].The final result will be used for [describe where or how it will be used].Here is the relevant context: [include background, requirements, examples, and limitations].I would like you to collaborate with me rather than immediately produce the final answer.First, summarize your understanding of the assignment.Second, identify the expertise or perspectives that would help us complete it.Third, ask me any clarifying questions you need before we continue.
You do not have to use those exact words.
The goal is not to discover a magic prompt.
The goal is to create a better working relationship with the model.
AI Cannot Read Your Mind
A common frustration with AI is that the result does not match the vision.
But the vision often exists only in the user’s head.
AI can infer. It can predict. It can make reasonable assumptions.
It cannot know what you never told it.
That is why prompting is not only about writing better instructions. It is about creating enough conversation for the model to understand your intentions.
As Jorge summarized during the episode, commanding AI will usually produce a result. Context creates conversation, conversation creates collaboration, and collaboration gets you closer to the result you actually had in mind.
The Best Prompt Is Often a Conversation
You do not need to become a prompt engineer to use AI effectively.
You need to communicate.
Explain what you are working on.
Share the context.
Admit what you do not know.
Ask for help.
Invite questions.
Review the model’s understanding before asking it to execute.
The next time you open an AI tool, do not treat it like an empty search box.
Imagine the right expert just sat down next to you.
Then start the conversation.
Frequently Asked Questions About AI Prompting
What is AI prompting?
AI prompting, often discussed as prompt engineering for large language models, is the process of giving an AI model instructions, context, examples, and questions so it can understand what you are trying to accomplish.
A prompt can be a simple request, but stronger prompts usually explain the goal, audience, intended use, relevant background, and any requirements the model should follow.
How can I write better AI prompts?
Start by explaining the situation instead of immediately demanding a finished result.
Tell the AI:
What you are working on
What outcome you want
Who the work is for
Where the result will be used
What constraints or preferences matter
What you are unsure about
Effective prompts usually include clear input queries for specific tasks.
Then ask the AI to summarize its understanding and identify any information it still needs.
Should I talk to AI like a person?
You do not need to pretend that AI is human, but it can be helpful to communicate with it the way you would communicate with a knowledgeable collaborator.
Mark Quinn recommends imagining that the world’s best expert in the subject has just sat down next to you. You would probably give that expert context, explain the problem, ask for advice, and answer clarifying questions rather than issue a vague one-line command.
What is the difference between commanding and collaborating with AI?
Commanding AI means giving it a direct instruction and expecting it to complete the task immediately.
Collaborating means explaining the assignment, discussing possible approaches, answering questions, reviewing its understanding, and refining the result together.
Commands can work well for simple tasks. Collaboration is often more effective for work involving creativity, strategy, judgment, or several interconnected decisions.
Why does AI give generic answers?
AI often gives generic answers when it receives a generic prompt.
When important details are missing, the model has to make assumptions about your audience, goals, tone, priorities, and desired format. Those assumptions may be reasonable, but they may not match what you had in mind.
Providing more relevant context reduces the amount the model needs to guess.
Should I include a role in my prompt?
Assigning a role can help, but the role should match the task.
For example, you might ask the AI to approach a project from the perspective of a brand strategist, financial analyst, product manager, editor, or designer.
However, assigning a role is not a substitute for providing context. “Act as a marketing expert” is still vague unless you also explain the product, audience, objective, and constraints.
What clarifying questions should AI ask me?
The questions will depend on the assignment, but useful clarifying questions often cover:
The desired outcome
The target audience
The intended format
The tone or style
The available information
Important constraints
Examples or references
How success will be measured
Rather than trying to anticipate every question yourself, ask the AI what it needs to know before beginning.
What are the three questions I should include in a prompt?
Mark recommends ending an initial AI interaction with three requests:
Summarize your understanding of what I am trying to accomplish.
Identify the expertise or perspectives needed for this task.
Ask any clarifying questions that would improve your support.
These requests help confirm that the model understands the assignment before it begins producing the final work.
Do longer prompts always produce better results?
No.
A prompt should be detailed enough to explain the assignment, but length alone does not make it effective. Better prompts matter more than length alone. A long prompt filled with irrelevant information can make the objective less clear.
Focus on information that materially affects the result.
Can AI create a good result from a one-line prompt?
Sometimes.
Simple, predictable tasks may only require a short instruction, and zero-shot prompting can work for direct requests. More complex work usually benefits from additional context and conversation.
A one-line prompt may produce an acceptable first draft, but it is less likely to match a specific idea that exists only in your head.
What should I do when the first answer is wrong?
Do not immediately restart the conversation.
Tell the AI what is wrong, what is missing, and what should change. Be specific about the gap between the current result and the result you expected.
You can also ask:
What assumptions did you make when creating this answer?
That can reveal where the model misunderstood the assignment.
Can AI read my mind or understand my intent automatically?
No.
AI can infer intent from the information you provide, but it cannot know details that were never communicated. Conversation gives the model the context it needs to produce something closer to your intended result.
As Jorge explained in the episode, commands can generate an answer, but context and collaboration help move the answer closer to what you actually envisioned.
What is a good beginner prompt template?
Use this structure:
I am working on [project].My goal is [desired outcome].This is for [audience].The final result will be used for [intended use].The relevant context is [background, requirements, references, and constraints].Before producing the final result, summarize your understanding, identify the expertise needed, and ask me any clarifying questions.
The wording does not need to be exact. The important part is creating a conversation before asking the model to execute. A few examples can help when you want a specific style or format.
Comments