Wednesday · Understand it

Start with the muddle

Hello,

You open an AI chat, look at the empty box and realise you don't quite know what to ask.

Perhaps you're preparing for a meeting. Perhaps you need to explain what your business does, or you've got a list of jobs that keeps getting longer. You know something needs doing. You haven't worked out what would help.

Start there. Write the untidy version, including what you're stuck on. Then ask the tool to help you work out the task before it tries to finish it.

Today, we'll turn a muddled note into a small plan. The useful bit is learning what to put into the conversation, and what to do when the answer misses the point.

The essentials

Everyday jobs account for much of the use

In September 2025, OpenAI reported that roughly three-quarters of consumer ChatGPT conversations concerned practical guidance, finding information or writing. Its research with Harvard economist David Deming describes use of its own product, rather than proving the answers were good.

Those ordinary jobs offer somewhere to start: explain something, organise something, or help put it into words.

​Read OpenAI's research summary.

The job matters as much as the tool

A study involving 758 consultants found that GPT-4 improved performance on some tasks and reduced accuracy on another. The final paper appeared in March 2026; it concerns an older model, rather than ranking today's apps.

Choose a trial you can judge. A suggested plan lets you check whether the answer fits your circumstances before acting on it.

​Read the consulting study (PDF).

These findings are background for today's lesson.

Understand it: the missing bit is often context

Compare these requests:

Help me prepare for a market.

I'm selling handmade greeting cards at a Saturday market, 10am to 2pm. They're £4 each or three for £10. I have an hour on Friday to get ready, and I'm doing this on my own. I'm tempted to spend it making an Instagram post, but I don't know whether that's the most useful job. Help me decide what to do first.

The second request supplies a goal, a time limit and the actual decision. You can use the same approach for a meeting, a household job or a piece of writing: say what you're trying to achieve, what limits you have, and where you're stuck.

The following walkthrough is fictional teaching material, written for this article. It isn't a transcript of a live AI test.

Before asking for a plan, add:

Ask up to three questions that would change your recommendation. Don't make the plan until I've answered. If you still need to assume something, say what it is.

Useful questions here concern whether the cards are ready, whether the seller has the equipment, and whether the arrival instructions are clear. Their answers would change the plan.

Suppose the cards and equipment are ready, but nothing is packed and the arrival instructions haven't been checked. An illustrative plan could be:

• 30 minutes: count and pack the cards and existing equipment.

• 10 minutes: make a price sign showing £4 each or three for £10.

• 5 minutes: check the organiser's arrival instructions.

That leaves 15 minutes for anything forgotten. The first action is clear: get the cards and packing box together. The Instagram post can wait until the preparation is done. Adjust the time allowances to the real task.

If the tool returns a long marketing strategy instead, you can correct it:

This is too much for the hour I have. Keep the preparation jobs that must happen before Saturday. Remove anything that needs new equipment or a campaign. Give me the first job to start now.

You don't have to accept the first answer as the finished product. Tell it where the answer failed.

Try it with one thing you're putting off

Choose something ordinary that you can judge yourself. Getting ready for a meeting, organising a weekend job or planning a short piece of writing will do. Use fictional details if the real task involves somebody else's private information.

Paste this into an AI chatbot you have access to:

Here is the messy version of what I need to do: [your notes].

A useful result would be: [what you want to have when you're finished].

My limits are: [time, budget, equipment or other constraints].

Ask me up to three questions that would materially change your advice. Wait for my answers. Then suggest a short plan and identify the first action. Mark any assumptions. Use plain UK English.

When the plan comes back, check whether it fits the time and resources you've given it. Notice any job it added without a reason. Change it until the first action is something you could actually do.

Without a chatbot, you can still use those headings to put your own notes in order. They help define what you need before you choose a tool.

What is worth keeping

A useful answer should make the next step easier to see. For the market seller, that's a manageable preparation job. For someone preparing a meeting, it might be a short list of decisions the meeting needs to resolve.

Judge the result by the task in front of you. A long answer can still leave you wondering where to start.

What have you been putting off because you don't know where to begin? Reply to Millie with the type of task. Leave out private details.

Millie

Someone forwarded this to you? Subscribe to One Eye on AI.

Sources and corrections

Sources are linked beside the findings. Reply if something needs correcting.