You Do Not Need the Perfect Prompt to Work Well with AI

Better results come from clear goals, useful context, practical tests and checking the work.

One of the most useful things I have learned from working with AI is that I do not need to know exactly what to ask before I begin.

Sometimes an idea is clear in my head but hard to describe. I can tell when the result is wrong, yet struggle to explain what needs to change.

Talking things through with AI helped me find another way to develop those ideas.

It also taught me something less comfortable: a long conversation does not necessarily mean progress.

An AI can misunderstand you, agree too readily or turn a simple problem into an unnecessarily complicated solution.

So I now start with what I know, use AI to help structure what is missing and, most importantly, define what I am actually trying to achieve.

You can work this way too, even if you are just getting started.

1. Research is useful. Research that considers your preferences is better.

Imagine you want to watch a film.

Searching Google for "best films of 2026" might bring up reviews, ratings and recommendation lists.

But do those films match your taste?

In ChatGPT, you could start like this:

I want to watch a thriller. I enjoy investigations, intelligent stories and satisfying twists. I do not like supernatural horror. Recommend five films and explain why you picked each one. Ask a few questions first if you need to understand my taste.

Now the search is shaped by what you actually enjoy.

You can mention films you have already seen, what you loved and what did not work. You can ask for less obvious choices or check where a film is streaming, provided the AI consults current information.

Over time, available context and memory features may also help the assistant tailor future suggestions.

That does not make Google less useful. Search engines also offer conversational AI features.

The point is that you can keep refining a search without starting from zero every time.

The same approach works for planning a trip, comparing products or finding tools for work.

2. Struggling to explain an idea? Ask AI to help organize it.

Imagine you need a better way to manage customer orders.

You might ask for a spreadsheet, search for an app or consider building a system.

But perhaps you do not yet know which solution is appropriate.

Try starting here:

I need to organize my customer orders. Right now, information arrives through messages, and I struggle to track progress. My budget is limited. Help me find the simplest workable solution. Ask no more than three important questions, then make a proposal.

You have already explained the problem, the context and one constraint.

AI can help you identify what is still missing.

One of my most useful habits is asking AI to help improve the request itself.

For example:

I am going to describe an idea in my own words. Organize what I say, identify important missing details and turn it into a clearer request. Do not complicate something that could be simple.

You do not need to memorize dozens of prompts.

What helps is learning to communicate your goal, your situation, the limits that matter and what a useful result should look like.

And there is no need to spend an hour answering questions. If something is essential, the AI can ask. Otherwise, it can make a reasonable assumption, label it and move forward.

3. Create separate Projects for different kinds of work

This is one of the most useful options for people who use AI regularly.

ChatGPT Projects let you group conversations, files and instructions around one topic.

You might have one Project for finding deals on Mercado Livre, taking your budget and preferred brands into account.

Another could help explore haircut styles, compare reference photos and think through what suits your preferences.

A third might help organize your sales.

Each Project keeps its own context, reducing how much you need to explain again.

Let AI help you set up the Project

Suppose you want a Project called Wilson Sales.

In a regular conversation, write:

I want to create a ChatGPT Project called Wilson Sales. It should help me review prices, products, customer service and sales opportunities. Ask the questions you need, then draft the instructions I should put in the Project settings. Tell me which information and files I should add to get started.

The assistant can help clarify the goal, draft the instructions and prepare the material for you to copy.

Then create the Project, add those instructions and include relevant files.

The interesting part comes when you give it a real task.

Imagine uploading a spreadsheet with products, prices and costs. Rather than asking for a vague analysis, try:

Analyze this spreadsheet and identify the products with the lowest margins. Show your calculations, list the costs included and highlight missing information needed to calculate profit correctly. Do not invent numbers.

Now Wilson Sales has a concrete job.

If the spreadsheet omits fees, taxes or other expenses, the assistant should point that out before recommending price changes.

Creating a Project does not automatically make AI an expert or guarantee accurate analysis.

The improvement comes from relevant information, useful instructions and a way to check the results.

This can also help teams, provided confidential information is shared only in authorized environments.

4. Work with documents, but learn to verify

Another practical use is analyzing files in a conversation.

Imagine receiving a fifteen-page contract.

You could attach the PDF and ask:

Identify my main obligations, deadlines, penalties and cancellation terms. For each item, cite the relevant clause and show the supporting passage. Separate what the contract says from your interpretation.

That last instruction matters.

AI can locate information and explain difficult language, but it can also misread clauses, miss exceptions or overlook details.

The next question should not simply be whether you can sign.

Try:

What is ambiguous? What could not be verified? Prepare questions I should clarify before making a decision.

Then check the relevant passages in the original document. Seek professional advice for important legal decisions.

The same principle applies to spreadsheets, reports and commercial proposals.

Use AI to make review easier, not to replace verification.

And be careful with personal documents, customer information and confidential material. Before uploading anything, check whether sharing it is permitted and whether the service has suitable protections.

5. A good idea has to survive contact with reality

This is one of the most important lessons I have learned while building projects with AI.

It is easy to spend hours imagining features, discussing possibilities and producing ever larger plans.

The hard part is turning those plans into something that works.

When I have a new idea, I now try to investigate whether it is viable before building it.

You can do the same:

I have a project idea. Evaluate whether it is feasible, research similar solutions and identify likely difficulties. Before suggesting anything complex, look for a simpler alternative. Then help me design a small test to see whether it is worth pursuing.

Suppose you want a system to organize orders.

You might not need to build an entire application.

A well-organized spreadsheet used for a few days could reveal what information really needs tracking.

If it works, you have learned something. If it does not, you have evidence to improve the idea before investing more time or money.

The key is choosing a result you can actually observe.

Apply the same question to AI's proposals:

Does this really solve my problem? What still needs testing? Are we making something more complicated than necessary?

A convincing plan is not the same as a well-executed project.

6. Over time, AI can fit your way of working better

There is another part of my experience I find interesting: continuity.

Through many conversations, I have learned to explain my needs more clearly and set expectations for how I want to work.

For visual projects, for instance, I make better progress when I share references and compare alternatives instead of trying to describe every detail upfront.

I have also learned to demand more critical responses, grounded reasoning and verifiable information.

In ChatGPT, you can record preferences through Custom instructions in the personalization settings.

For example:

I prefer direct, practical answers. Ask questions when essential information is missing. Do not agree with me automatically. Avoid complicated solutions when a simpler one would work, and clearly identify what still needs checking.

If Memory is enabled, ChatGPT may also use relevant information from earlier conversations and other available sources.

That can make replies more tailored, but it does not mean the assistant remembers everything or understands your preferences perfectly.

I think of this as a working relationship that develops over time.

You learn to guide the tool more clearly, while it can use available context to respond in ways that better suit your needs.

As you gain experience, you may encounter more advanced ideas such as Skills, MCP and RAG, which can help structure workflows, connect tools and draw on specific knowledge sources.

They are worth exploring in their own articles. You do not need them to get started.

7. Do not be afraid to disagree with AI

This may be the most important advice of all.

The first response will not always be good.

I have received suggestions that sounded promising but were generic, unnecessarily complex or simply not what I needed.

I learned not to accept an answer just because it was well written.

If something does not make sense, challenge it:

Why are you recommending that? What are the alternatives? What have you verified, and what are you assuming?

If the result misses the mark, explain why.

If the solution is too complicated, ask for a simpler one.

And if the conversation is drifting away from your goal, return to that goal.

AI can be a valuable thinking partner, but the responsibility for judging the result remains yours.

The best way to learn is to start with something real

Choose a task you actually want to solve.

It might be finding a film, comparing products, organizing information, analyzing a document or developing an idea you have been carrying around for months.

Start by explaining what you know. Ask for help structuring the rest. Define the outcome you want and inspect what the tool produces.

As you practice, you will learn which instructions work, when to explore further and when it is time to stop planning and start doing.

You do not need to learn how to speak the AI's language. Start by speaking your own.

There is a step beyond that, though.

Talking to AI is easy. Learning to turn that conversation into something genuinely useful is the skill worth developing.