Let’s talk about AI Prompts…
Scrolling through my social media posts, I’ve seen lots of ‘how to’ guides on the best way to prompt your AI assistant. Many of them are useful, but they are often framed within a given context which means you have to figure it out yourself anyhow. Getting a prompt ‘wrong’ can be a little frustrating at times, but it is all part of the learning process – and as it turns out, it is a two-way street.
Whilst considering how I go about sharing my relative inexperience on this subject, for this article, it made sense to also address what I could have done better. Much of my prompting to-date has been unstructured, but I’ve been able to get the results I wanted. What should I be doing? Well I thought I’d go straight to the source…
The question that started it…
In order to get the information I require on subject ‘X’, how should I phrase my prompt to get what I need?
That seemed a perfectly reasonable generic question to ask my AI assistant (ChatGPT in this instance). There was just one problem. I was asking it how to help me write a good prompt without first giving it much information about what I was trying to achieve.
Its response helped identify the information that would make the request more useful.
In other words, I asked AI how to ask AI what I should ask AI. It demonstrates the point rather nicely. AI wasn’t simply producing an answer to a question. The conversation helped me work out what information I needed to provide in order to get a better answer – and in doing so it also provided much of the draft text for this article.
Instead of guessing, ask AI…
So, rather than trying to work out a perfect prompt myself, I should be asking AI what information it needs from me in order to answer the question properly.
That provided a much more useful way of thinking about how I approach this.
Instead of:
What is the perfect prompt for this?
I should ask:
I need to research [SUBJECT]. What information do you need from me to give me the most useful and relevant answer?
The important part isn’t the exact wording. It is the idea behind it. If you don’t know what your AI assistant needs to know, ask it.
That turns the process from trying to guess the magic words into something much closer to having a conversation with someone who knows a lot about the subject, but doesn’t know your particular circumstances. So, what did AI tell me it needed to know (in general terms)?
A prompt is really a brief…
A good brief gives somebody enough information to understand what you want them to do. AI prompting works in much the same way. Depending on the task, useful information might include:
- The Task
What do I actually want AI to do? - The Context
What does it need to know about the situation? - The Purpose
Why do I need the information? - The Audience
Who is going to use or read the result? - The Scope
What should be included, and what is outside the job? - The Constraints
Are there limits on length, sources, style, terminology or anything else? - The Output
Do I want a list, article, table, summary, email, code or something else? - Examples
Is there an existing example that demonstrates what I am looking for? - Verification
What should happen if AI is uncertain or the information needs checking?
This response looks very familiar to all those lists on social media. I wonder what came first – an AI-generated response now spun on social media or an original user guide being referenced by AI?
Anyhow, you don’t necessarily need all of these every time. For a simple question, a short prompt may be perfectly adequate. For a more complicated piece of work, however, providing more context can make a significant difference.
And this is where asking AI what it needs can be useful.
An updated approach: let AI help build the brief…
This is the approach I am going to trial further as I have been randomly using it.
Start with what I actually want to achieve, rather than trying to construct a technically perfect prompt.
Then ask AI:
What information do you need from me to do this properly?
The answer becomes my checklist. I can then provide the information, refine the request and get on with the actual task. It also means I don’t have to know in advance exactly how an AI system wants a prompt structured.
I’m effectively asking the assistant to help me brief it. I don’t know if this is the best way to do it as I guess it depends on context, but it certainly appears to be a logical approach.
A simple starting point…
The suggestion from my AI assistant; If you are unsure how to start, try something along these lines:
- I want to [TASK].
- The context is [CONTEXT].
- The result is for [AUDIENCE/PURPOSE].
- I need the answer in [FORMAT].
- My constraints are [CONSTRAINTS].
- What else do you need to know from me to give me the most useful answer?
You don’t have to fill in every line. In fact, that last question is the bit I find most interesting. What else do you need to know? It gives your AI assistant an opportunity to identify missing information before it attempts to complete the task. That can be particularly useful when you know what you want to achieve but don’t necessarily know how to describe it.
This isn’t a magic formula…
There is a temptation to treat prompt writing as though there is one perfect formula that will suddenly make AI produce brilliant answers.
I don’t think it is quite that simple. Current guidance on prompt writing (and I’ll also reference those social media posts again) commonly comes back to the same broad principles: explain the task, provide relevant context, set useful constraints, specify the desired output and, where appropriate, provide examples or ask for verification.
Those are useful principles. But the best prompt for a simple question might be one sentence. A complicated research task may need considerably more information. The trick is not necessarily to make every prompt longer. It is to make it more useful.
And if you don’t know what would make it more useful, ask AI.
The takeaway…
For me, the adaptation in thinking is this:
- You don’t necessarily need a perfect prompt. You need a useful conversation.
- Start with what you are trying to achieve.
- Tell AI what you already know.
- Explain the context where it matters.
- Give it any important constraints.
Then, if you aren’t sure what is missing, ask.
- What else do you need to know from me to answer this properly?
That may be one of the most useful prompt-writing questions you can have in your toolkit. And if you are wondering how to ask your AI assistant to help you write a good prompt… Well, you could always start by asking it.
