There's a particular kind of stuck that most people recognize: you have something you want to do โ€” a project, a side venture, a change you want to make at work, a conversation you need to have โ€” but it's still living in your head as a vague shape rather than an actual plan. You know roughly what you want. You don't know where to start. And because you don't know where to start, you don't start.

This is one of the places where AI tools turn out to be genuinely useful โ€” not because they have better ideas than you, but because they're patient, available at any hour, and surprisingly good at helping you externalize and organize thoughts that are currently tangled up in your own head. The process is less about getting the AI to think for you and more about using it as a surface to think against.

Here's what that actually looks like in practice.

Start by Describing the Mess, Not the Plan

The mistake most people make when using AI for this kind of thinking is trying to be too organized too early. They sit down, try to write a clean prompt, and end up either staring at the blank input box or producing something so vague โ€” "help me make a plan for my business idea" โ€” that the AI gives back something equally vague and useless.

The better approach is to describe the mess. Open ChatGPT or Claude and just talk. Tell it what you're trying to figure out, what you know, what you don't know, and what's making it feel stuck. Don't tidy it up first. The messiness is useful information.

Something like: "I want to start running workshops for people in my industry, but I'm not sure if I should do them in person or online, whether I should charge for them or offer them free to build my reputation, and I don't have a clear sense of who the audience is yet. I've been thinking about this for about six months and keep going in circles."

That's enough. A good AI tool will respond by either asking clarifying questions (Claude tends to do this well) or by pulling the threads apart and showing you what the actual underlying questions are. Both responses move you forward, because they show you what you're actually dealing with rather than letting it stay as an undifferentiated cloud of uncertainty.

Use the AI to Name the Real Decision

Most messy ideas feel messy not because they're complicated, but because somewhere inside them there's a decision that hasn't been made โ€” and everything else is just noise around that decision. When you're going in circles, it's usually because you're circling that unacknowledged choice without quite landing on it.

AI is good at surfacing this. After you've described your situation, you can explicitly ask: "What do you think the core decision is here that I need to make first, before the other things can get clearer?"

In my own work, I've done this with projects that felt complicated and had the AI hand back something like: "It sounds like the core decision is whether you're building this primarily for reputation or for revenue โ€” because once you've answered that, the in-person vs. online question and the pricing question both become easier." That kind of reframe is often worth more than any amount of detailed planning, because it tells you where to focus your actual thinking energy.

You may disagree with what the AI identifies as the core issue. That's fine โ€” and it's useful. Arguing with a wrong answer is often how you figure out what the right answer is.

Ask for a Starting Point, Not a Complete Plan

One of the places AI planning assistance goes wrong is when people ask it for a comprehensive plan too early. The AI will produce one โ€” it's quite good at generating structured outlines โ€” but a comprehensive plan for something you don't fully understand yet is mostly an illusion of progress. You end up with a document that looks organized but doesn't actually reflect your real situation, your real constraints, or the decision you haven't made yet.

A more useful ask, once you've identified the core questions, is: "What's the smallest next step I could take in the next two weeks to get better information about whether this is worth pursuing?"

This reframes the task. Instead of planning the whole thing, you're planning the first move โ€” the thing that will tell you whether a fuller plan is warranted. AI tools handle this kind of question well because it's concrete and bounded. The response is usually something you can actually act on rather than something that requires three more planning sessions before anything happens.

Let the AI Play Devil's Advocate

Once you have a rough direction, one of the most practically useful things you can do is ask the AI to push back on it. Ask it what could go wrong. Ask it what assumptions you're making that might not hold. Ask it what someone skeptical of your idea would say.

This feels slightly uncomfortable, which is exactly why it's valuable. Most of us are not great at self-critiquing ideas we're excited about. We tend to look for confirmation rather than challenges. AI tools don't have an emotional stake in your idea, which makes them reasonably good at generating counterarguments you might not have surfaced on your own.

You might ask: "What are the three most likely reasons this wouldn't work out the way I'm imagining?" or "What would a skeptical person ask me if I pitched this to them?"

Perplexity is also worth using at this stage if your idea involves an industry or market you don't know well โ€” it will pull in current information and help you check whether the landscape matches your assumptions. Claude and ChatGPT are better for the thinking-through-problems conversation; Perplexity is better when you need real-world data to ground the thinking.

Build the Plan in Stages, Not All at Once

Once you have a clearer sense of what you're doing, AI is useful for breaking it into phases. But the key is to do this iteratively, not all at once. Ask for the first phase. Review it. Add your own knowledge and corrections. Then ask for the next stage.

This matters because AI doesn't know your specific situation as well as you do โ€” your time constraints, your actual skills, the specific people involved, the organizational politics if it's a work project. If you let it generate everything in one pass, it will make assumptions that fill in the gaps you didn't mention. Some of those assumptions will be wrong in ways that make the plan unworkable.

The iterative approach keeps you in the driver's seat. You're providing the judgment; the AI is providing the structure and asking the questions you might not have thought to ask yourself. That division of labor is where these tools are genuinely helpful rather than just producing the appearance of a plan.

A Note on What This Isn't

AI isn't going to tell you whether your idea is a good one. It doesn't know that. It will engage with almost anything you bring to it with the same level of apparent enthusiasm, whether the idea is sensible or not. That's not a feature โ€” it's something to stay alert to.

If you ask ChatGPT to help you plan something and it responds helpfully, that is not the same as the idea having been validated. The only validation that matters comes from talking to real people in the world โ€” potential customers, colleagues, people who have tried something similar. AI can help you figure out what questions to ask those people, and how to make sense of what they tell you. It can't substitute for asking them.

What AI does well in this process is remove the paralysis that comes from not knowing where to start. Most messy ideas stay messy partly because starting the thinking feels like a commitment, and the stakes feel high before you've done anything. Using AI to start that conversation costs you nothing โ€” it's low stakes, you can delete the conversation, and you're not accountable to anyone for how it goes. That permission to think messily and out loud, with something responding, turns out to be genuinely useful for a lot of people.

The plan that comes out of this process will still need your judgment, your revision, and your actual effort to execute. But it will exist โ€” which is already further than a lot of ideas get.