AI as an Executive-Function Aid (ADHD-Friendly Workflows)

TL;DR: Using AI for ADHD-style attention usually works best as an executive-function aid: something that holds the plan, the first step, and the state of an interrupted task so your own attention does not have to. This is workflow scaffolding, not treatment, and it comes with a privacy trade-off worth knowing before you write anything down.

What does it mean to use AI as an executive-function aid?

Executive function is the unglamorous background work of getting anything done: holding a plan in mind, sequencing the steps, starting the first one, and noticing when you have drifted off it. None of that is about intelligence or effort. It is closer to bookkeeping than to willpower, and bookkeeping is exactly the kind of task worth outsourcing to something that never gets tired of holding a list.

Some people find that treating an AI chat as external scaffolding, something that holds the plan so their own attention does not have to, changes how a day actually goes. That is the whole idea behind AI for ADHD as a workflow rather than a slogan: not a tool that understands your brain, but a place to put the parts of a task that are easy to lose track of mid-task. The value is entirely in the scaffolding. It is not a claim about what ADHD is, how common it is, or how it should be treated, and nothing in this post is written as if it were.

That distinction is worth stating plainly before going further: this describes a workflow, not a condition, and it works the same way regardless of whether "ADHD" is the word you would use for how your own attention operates.

None of the techniques below are exotic. They are ordinary planning habits, breaking a task down, writing a note to your future self, that get more valuable the more expensive it is for you personally to reconstruct a plan from nothing. What makes them worth writing down as a set is that an AI chat can do the reconstructing part for you on demand, cheaply, without judgment, at eleven at night when nobody else is available to help you figure out where you left off.

What should you know before you write any of this down?

This is worth deciding upfront rather than mid-workflow, because every technique below involves writing something down: a plan, a status, a note to your future self. Keep it task-shaped. "Break the report into three parts and start with the outline" is task-shaped. A running log of what triggers a bad day is not, and none of what follows needs it to be.

A simple filter, if you are ever unsure whether a line belongs in something you save: would you be comfortable with a coworker reading it over your shoulder. If the honest answer is no, it is either too personal for the task at hand, or it belongs in a private note rather than a stored context entry, and either way it does not need to go where an AI tool can see it.

How small does a "first step" actually need to be?

Smaller than feels reasonable, almost always. A step that still requires deciding between three sub-approaches is not actually a first step, it is a smaller version of the original decision. "Outline the report" sounds small until you notice it still requires knowing what sections a report like this needs, which is exactly the kind of decision that stalls things in the first place.

A useful calibration question is whether the step could be done in under two minutes without having to think about what comes after it. "Write the report" fails that test. "List five section headings for the report, no content yet" usually passes. If a proposed first step still fails it, the fix is not more willpower, it is asking the model to break its own suggestion down one more level: "That's still too big. Give me the smallest possible next thing."

This matters more than it sounds like it should, because a first step that is genuinely too big produces the same stall as no first step at all, just one level removed. The goal is not a good plan. It is a step small enough that starting it does not require one.

How do you turn a vague obligation into a first concrete step?

The most common failure with a big task is usually not the task itself. It is that "deal with the taxes" or "sort out the client contract" is not a thing anyone can actually start. There is no verb in it small enough to do in the next five minutes. That gap, between an obligation and a first move, is where an AI chat earns its keep as an executive-function aid.

The prompt pattern is simple: describe the obligation exactly as vaguely as it actually sits in your head, and ask for the smallest possible first action, not a full plan.

The pattern in every row is the same: you are not asking the model to do the task, you are asking it to find the smallest legitimate entry point into the task. That is a different question from "help me write the report," and it produces a more usable answer. Our guide to writing a prompt for a task you don't understand yet covers the same first-step problem from the angle of not knowing the domain, rather than not knowing where to start; the two overlap more than they differ.

A second version of the same trick helps when even describing the obligation clearly is the hard part: ask the model to interview you first. Getting ChatGPT to ask you questions before it answers turns "help me write the report" into three or four short questions that do the sequencing for you, one at a time, instead of leaving you to hold the whole plan in your head before you can even begin.

What does externalizing working memory actually look like day to day?

The scaffolding above only helps once, for one task. The bigger win is not re-deriving it every single time. If you have found a phrasing that reliably gets you a usable first step, or a version of the interview-me prompt that fits how you actually talk, saving it somewhere you can reuse it is the difference between a trick and a workflow.

That is the real argument for building a personal AI prompt library in this specific context: not organization for its own sake, but not having to reconstruct a working prompt from memory on a day when reconstructing anything from memory is exactly the problem. The same logic covers standing preferences, the default tone you want, the project names you are tired of re-explaining, which is what reusing prompts without copy-pasting is really solving: fewer things you have to hold in your head just to ask a question.

A quieter, second form of the same idea is keeping an open-loops list separate from whatever task list you already use. Not what needs doing eventually, but what you are currently mid-way through and would otherwise have to reconstruct from scratch: one line per open thread, updated the moment you switch away from it. Three half-finished threads held only in your head, with a fourth thing interrupting, is exactly the situation this list exists to prevent.

How do you write a re-entry note for when focus breaks?

Losing your place mid-task and having to reconstruct where you were is one of the more expensive things that can happen to a piece of work, because reconstruction takes real attention on top of whatever the task itself needed. A re-entry note trades a few seconds now for skipping that reconstruction later: before you step away, however briefly, write down what you were doing, why, and the exact next action, not a summary of the whole task.

RE-ENTRY NOTE
Task: [one line, what this actually is]
Where I stopped: [the literal last thing you did]
Why I stopped: [interruption, decision needed, out of time]
Next action: [one verb, specific enough to start without re-reading anything]

The reason the last line matters most is that "next action" is not "continue the report." It is "write the second bullet under the budget section," something you could do without first figuring out what you were even doing. Filled in, a real one might read: task, draft the Q3 budget memo; where I stopped, finished the headcount section, have not touched software spend; why I stopped, needed the actual invoice numbers from finance; next action, message finance for the November invoice, then draft the software-spend paragraph. Four lines, and the entire reconstruction cost of picking this back up two days later is reading them.

Paste the filled-in note wherever you already keep things, a saved prompt, a pinned note, a context entry, and treat coming back to a task as reading four lines, not re-opening the whole file and reorienting from scratch.

Does this replace an assessment, therapy, or medication?

No, and it is not trying to. Everything described here is workflow scaffolding: prompts, saved templates, and a note-taking habit, aimed at a specific practical friction, starting and resuming. None of it diagnoses anything, treats anything, or substitutes for an assessment, a clinician, or medication, and finding it useful or not useful says nothing about whether any of those would help you. If you are weighing whether to seek an assessment, or you are already working with a clinician, that conversation belongs with them, not with a prompt template.

What if a workflow stops fitting how you think?

Every technique above is a starting shape, not a fixed rule. If the re-entry note format feels like overhead instead of relief, shorten it to one line. If the interview-me prompt asks the wrong kind of questions for how you actually plan, rewrite it until it does not. Some people find a five-line skeleton useful for starting a report and a nuisance for starting an email; both reactions are legitimate information about what to keep.

Once a version of a workflow has proven itself more than once, it is worth turning into something you save rather than something you re-improvise each time, which is the entire subject of turning a repeated workflow into a template: the point where a personal trick becomes a personal system.