From one question to a whole workflow: six ways to get more out of AI
Tools like ChatGPT and Claude can explain concepts, draft client proposals, test ideas and help with complex tasks. The quality of what you get depends, among other things, on how you define the task: what information you give, which examples you attach and how you check that the work is done.
The six methods below build on each other. They start with asking for a simple explanation and work up to a process where the AI does the work, checks it and improves it against criteria you set in advance.
To keep it concrete, one example runs through the whole guide: starting a new business and writing a proposal for a client.
Every part ends with a ready-made prompt. Replace the text in square brackets with your own details and paste it into the chat.
ELI10 is short for Explain Like I’m 10. With AI, it is a way to ask for a clear explanation that works for someone who does not know the field.
Someone starting a business may suddenly face questions about company structure, invoices, payment terms, contracts and finding clients. When every term is new, even the first step can look complicated.
Asking for a simple explanation helps break the topic into pieces:
Explain to me, ELI10 style, how to start a consulting business. Start with the basic concepts, explain every professional term and lay out the steps in a logical order.
The same approach works for a clause in a contract, a financial concept or a piece of technology. The goal is an accessible starting point you can build on step by step.
ELI10 is not a special command you need to install. It is just a well-known way of phrasing a request. You can also write: "Explain this in plain language, without assuming I know anything about it."
Explain [the topic] to me as if I were ten years old, without assuming any prior knowledge. Use simple language, a real-life example and a short step-by-step explanation.
Even an AI that explains things well does not necessarily know your business, your goals or your preferences.
So before an important task, ask it to ask you questions first:
I want to write a proposal for a new client. Before you draft it, ask me five clarifying questions that will help you tailor it to my business and to the client.
The questions might cover the service you offer, your audience, the problem the service solves, price, payment terms and your track record. Knowing what you have already tried, what worked and what did not adds important context.
For example, "Write me a proposal for consulting services" leaves a lot of room for guessing. A request that says who the service is for, what the client needs and why your business fits the job produces a much more focused proposal.
Context: the clearer and more accurate the relevant information, the more likely the result fits what you need. The number of questions is flexible; what matters is that they fill the gaps that affect the work.
My task: [the goal or task]. Before you answer, ask me five clarifying questions that will help you tailor the answer to my needs. Ask one question at a time.
Words like "professional", "persuasive" or "interesting" can mean different things to different people. A concrete example shows the AI exactly what you mean.
When writing a client proposal, you can attach a previous proposal that closed a deal, a template you like or a few examples of the style you want:
Here are two proposals that represent the level and style I am after. Analyse their structure, tone and how they present value, then write a new proposal for the client I described.
It also helps to explain what works in the examples: a short opening, speaking directly to the client’s problem, proof of results or a clear next step.
The method works for CVs, posts, emails and presentations too. In every case, examples give a reference point that general words struggle to convey.
Show alongside the explanation. The example teaches the structure and style you want; the information about the task lets the AI adapt them to the new case.
Here is an example of a result I like: [paste an example]. Identify its structure, its style and what makes it work, and use them to create [the result you want] for [the business or audience].
Once the AI has information about your business, has seen examples and has gone through a few rounds of corrections, you have a useful working base. Saving it means you do not have to start from scratch next time.
Here it helps to separate two kinds of information.
Memory keeps ongoing context over time: your field, your audience, your preferences and your communication style, depending on what your tool supports and how it is set up. You can ask the AI to summarise what is worth keeping:
Based on this conversation, summarise the important details about my business and the preferences to keep in mind for future tasks.
A skill is a set of reusable instructions for one specific process. For example, a process for writing client proposals can include the opening questions, the proposal structure, examples, style rules and quality checks.
You might ask:
Turn the proposal-writing process we built into clear instructions for reuse. Include the examples, preferences and corrections we agreed on.
You can store instructions like these as a skill, a custom assistant such as a Custom GPT or a saved template, depending on the tools you use. Some tools let you trigger a saved process with a command that starts with a slash (/), for example /write-pitch. That name is only an example of a custom command, and using one depends on it being set up and installed.
The idea is similar to the programming principle DRY, short for Don’t Repeat Yourself: you keep instructions you repeat in one place, so you can reuse them and improve them there.
The improvement comes from the knowledge and instructions that were saved and are available to the AI. It does not mean the model itself is retrained after every conversation.
Summarise the important information about my business and my preferences from this conversation. Separately, turn the process we built into instructions I can save as a skill or a reusable template, including the examples and corrections we agreed on.
AI can also help you look at an idea critically. Its answers sometimes lean towards agreeing with you or encouraging you, so it is worth asking it explicitly to look for weaknesses.
For example:
Review this proposal critically. Find blind spots, weak assumptions and information I am missing. Separate the facts I gave you from assumptions, and explain what is worth checking.
For a new business, this might raise questions like: is the audience defined well? Is the promise to the client backed by experience? Does the price fit the scope of the work? Is it clear what the client gets?
Think of it as practising your reasoning with a partner, sometimes called "sparring": you put forward an idea, test the objections and sharpen your reasoning.
The goal is to see what is solid, what is still an assumption and what data you need to decide. The AI’s criticism needs checking too; confident, persuasive wording is not proof that something is right.
Critically review [the idea or plan]. Find weak assumptions, blind spots and information I am missing. Explain what needs checking and suggest practical ways to improve it.
In everyday work with AI, you write a request, read the answer, ask for a fix and check again. That creates a loop of writing, feedback and improvement, and you run every step of it.
The approach described here as loop engineering hands part of the checking and fixing to the AI. For that, you need to define two things: what result you need and how to check it.
Instead of only asking "write a good proposal", you can write:
Write a client proposal of up to 150 words. Open with their specific problem, include one piece of proof from the information I gave you and end with a clear next step. Check the draft against every requirement and fix it before showing me the final version.
A definition like this lets the AI check whether the proposal is too long, whether it speaks to the client and whether every required part is there.
In more complex processes you can use a separate checker, sometimes called a verifier. Its job is to compare the work with the criteria and point out what is wrong. The check can rely on a list of requirements, a testing tool or another model.
When building an app, for example, a check might mean opening the app in a browser, clicking the buttons and looking at the phone layout. In writing, you can check length, structure and whether the required details are in. A quality score such as nine out of ten is more useful when it is clear what is being measured and how the score is set.
Claude Code also has a /goal command, an example of a mechanism that separates doing the work from checking it. Whether it is available and how it behaves depends on your setup and version; the principle is to define a goal, do the work, check it and fix it.
An instruction like "keep going until the task is done" needs a practical definition of what counts as done, plus the right tools to check it. With those in place, you can cut down how often you need to step in.
Do [the task]. The result must meet these requirements: [list of requirements]. Check the work against each requirement and fix any problems before showing it to me. If there is something you cannot check, say so explicitly.
When the methods come together as one process, you can build more consistent work with AI and spend more of your attention on the goals and decisions the work is meant to serve.