About project
A common situation in many offices is someone opening ChatGPT, typing a vague request like “I need a budget plan”, and then wondering why the result is not very useful.
Others end up frustrated, asking the one colleague who is more comfortable with AI, “How do I get it to do what I want?!”
The answer is usually simple: keep talking to it. Have a conversation!
Working with AI should be an ongoing back and forth. Add context. Explain what you are trying to do. Tell it when something is wrong. Ask what information it needs from you, and refine the result as you go.
This sounds obvious, but it solves a surprising number of the problems people have when they first start using tools like ChatGPT, Claude or Copilot.
Challenge
People often expect far too much from the first message they send.
“I need a budget plan.”
“Write an email.”
“Make me an image.”
Technically, the AI can respond to all of these. The problem is that it has almost nothing to work with, or arguably far too much to work with, given that there are infinite possibilities of what "an image" or "an email" might be.
What kind of image? What is the budget for? Who will read the email? How detailed should the answer be? What information already exists? What constraints matter?
If those details are missing, the model has to fill in the gaps itself. The result might be too formal, too vague, completely the wrong style, or full of phrases that nobody involved would actually use or understand.
This is usually the point where someone starts searching for the “right prompt” or asks a colleague who uses AI more often for help.
In many cases, there is no special prompt, and no need to ask the colleague everyone has decided is “good at AI”. The missing information can simply be added through a brief conversation.
Process
A useful starting point is to tell the AI what needs to get done in normal, conversational terms.
Instead of:
“I need a budget plan.”
Try:
“I need to put together a budget plan for a new training project. I have the expected costs, but I am not sure how to structure it. Ask me what you need to know before you create anything.”
That last sentence is incredibly useful. If you are unsure what context the AI needs, ask it.
It may ask about the audience, available information, deadlines, format, budget categories, level of detail and anything else that might affect the answer. You can then provide those details one at a time.
The same applies once the AI starts producing work.
If an email sounds far too formal, say:
“This sounds too formal. Make it sound more natural and closer to the way a normal colleague would write.”
If a paragraph is full of generic wording, say:
“This is too vague. Be more specific and remove anything that sounds generic or obviously AI-written.”
If an image comes back looking like a cartoon when you wanted a photograph, say:
“I need this to look truly photorealistic. The current result looks illustrated. What should I change in the prompt?”
You can even ask the AI to help create the prompt itself.
For example:
“I need to generate a realistic image of a safety training scenario, but I am not sure how to describe it properly. Ask me the questions you need, then write the image prompt for me.”
The useful part is the back and forth. The first answer gives you something to react to. Your reaction gives the AI more information. The next answer gets closer.
This also removes a lot of the pressure people put on themselves to write a perfect first prompt: you do not need to predict every detail before you start. You need to be able to explain what you want, recognise when the result is wrong, and say why.
Result
Once people start treating AI this way, it becomes much easier to use.
A vague first request can turn into something useful through a few short exchanges. A bad draft can be corrected instead of abandoned. A confusing task can be broken down by asking the model what information it needs. A useful prompt can be created with the AI, rather than written from scratch by the user.
The quality of the result still depends on the information you provide and your judgement of what comes back. AI cannot know that an email sounds wrong for your company, that a technical term is inaccurate, or that an image has missed the point unless you tell it.
The next time an AI tool gives you something disappointing, keep going. Tell it what is wrong. Give it more context. Ask what it needs to know.
Have a conversation!
Project shots
