There’s a conversation happening in many AI training sessions that I think is missing something important.

It usually goes like this: here’s a framework, here are the parts, here’s an example, now you try. And the framework is genuinely useful. I wrote about why I use two of them and when to switch between them. Structures like GCES and CO-STAR are real scaffolding. I reach for them regularly.

But when people say they find prompting hard, I’m not sure the framework is the missing piece. I think what they’re actually describing is a thinking problem, not a typing problem.

What prompting actually asks of you

When you sit down to write a prompt, you need to:

• Know what you actually want (which is harder than it sounds)
• Hold that goal in mind while you organise your thoughts around it
• Decide what context matters and what doesn’t
• Sequence your instructions so they make sense
• Then review what comes back and decide if it’s right

That’s planning, working memory, task initiation, and cognitive flexibility – all in one small box before you hit enter.

For anyone with ADHD or AuDHD, that list isn’t a checklist. It’s a wall.
Not because the person isn’t capable. But because those are exactly the executive functions that don’t always fire in sequence, or on demand, or under the low-level pressure of a blinking cursor.

The part the framework doesn’t cover

A framework like GCES gives you the what. It provides the goal, context, expectations, source. But it doesn’t help you figure out what your goal actually is when you’re in the foggy, half-formed stage of a task. It doesn’t help you pull relevant context out of a brain that’s simultaneously thinking about three other things. It doesn’t tell you how to start when starting feels impossible.

That’s the gap I keep noticing. And it matters, particularly because that gap is disproportionately felt by people who are neurodivergent.

Here’s the thing though: prompting, done well, can actually be the scaffold. Not just use it.

Prompting as externalised thinking

One of the ways I use AI that I don’t often hear talked about is using it to figure out what I think before I ask it to do anything.

On a hard day, when I can’t find the thread of a task, I’ll start with something messy and unfinished – almost like thinking out loud in the prompt box. Not because I expect a great output. But because the act of writing something, anything, gets the thinking moving in a way that staring at a blank document doesn’t.

The prompt becomes a thinking tool before it becomes a task tool.

For neurodivergent people, especially people with ADHD or AuDHD, this matters. Working memory challenges mean we don’t always hold the full shape of a task in our heads. Initiation difficulties mean the blank page can be paralysing. Writing a rough, incomplete prompt can be enough to externalise the shape of the thing – to get it out of your head and onto a surface where you can see it and start working with it.

The framework comes in later, once you know what you’re trying to do. Not before.

What this changes about how I think about prompting

I used to think of prompting as a technical skill. Here are the parts, here’s the structure, here’s the shortcut.

I’m shifting toward thinking about it as a cognitive skill that happens to have a technical layer.

The technical layer (the frameworks, the syntax, the refinement loop) is learnable and it matters. But underneath it is the harder question of how you get from a fuzzy, unformed idea to a clear instruction. That transition is where most people get stuck. And that’s not a prompting problem. It’s a thinking problem. Specifically, it’s an Executive Function problem.

When we design training that skips that layer, we design training that works for people who already find it easy. Which means we design training that leaves behind exactly the people who would benefit most from getting good at this. To include it and design training that suits most of a group is more inclusive, and that matters.

A few things that actually help

For anyone who finds prompting hard, especially if you suspect Executive Function is part of the picture, a few things worth trying:
 

Start messy 
Write something rough and incomplete into the prompt box. Don’t try to structure it first. Let AI help you find the shape of what you want. Ask it: “I’m trying to do [vague thing] — what questions would you ask me to help me be clearer?”
 

Use prompting as a planning tool, not just an output tool 
Before you write the polished prompt, write a scratchpad version. Talk yourself through it. What do I actually want here? Who is it for? What would good look like? The answers to those questions become your prompt.
 

Separate the thinking from the doing 
The cognitive work of knowing what you want, and the technical work of writing the prompt well, don’t have to happen at the same time. You can do the thinking first — even just in dot points or rough notes — and then shape it into a prompt.
 

Give yourself permission to iterate 
The first prompt is almost never the last one, and that’s fine. The loop of ask, review, refine isn’t a sign you did it wrong the first time. It’s just how this works.
 

Why this matters beyond individuals

I spend a lot of time thinking about AI adoption at an organisational level. And one thing I notice is that prompting confidence tends to split fairly quickly along lines that we don’t always name directly.

Some people find the structure intuitive and take off fast. Others feel like they’re doing something wrong, even when they follow the framework exactly, because the output still doesn’t match what they had in their head. Not because they didn’t follow the instructions. But because the step before the instructions, knowing what they actually want, was the hard part.

If we want AI to be genuinely inclusive as a work tool, we need prompting education that reaches that earlier layer. The one where the thinking happens. The one that doesn’t fit neatly on a laminated card.

Frameworks are a good start. But they’re the scaffolding around the skill. The skill itself is thinking clearly enough to tell AI what you need. And for a lot of people, particularly those who’ve spent a lifetime working around brains that process differently, that’s where the real work is.

Interested in the frameworks I mentioned? Read the earlier post on GCES and CO-STAR and how I decide between them.