The 2-Rule Prompting Technique I Use With Every AI Tool

Use the exact term. Ask for one contained result. Save the cheat sheet for your next prompt.

A useful prompting technique can fit into 2 rules: use the shortest established term that still says exactly what you mean, then make one single-purpose, self-contained request. These checks work across the AI tools you already use because they make your meaning and your requested result easier to identify.

I can read a prompt guide, nod along, and understand every section. Then I close the tab and most of it disappears. The next blank prompt still expects me to remember a page of advice.

These are the 2 rules I can remember.

First, I will show you how one exact term can replace a long improvised description. Second, we will turn a message with several possible jobs into one contained request. At the end, you can copy the complete check into your notes.

Rule 1: Use the Shortest Established Term

Use the shortest established term that still says exactly what you mean.

Suppose you want AI to tell someone how to cook onions over low heat until they become sweet and deep brown. You can write that entire description every time. You can also use one established word: caramelize.

That word already carries the heat, time, and result. AI has a recognized cooking term to work from instead of several improvised phrases for the same job.

Sometimes you know the idea but do not know the term. Describe what you mean, then ask:

Is there an established term for this?

I used that question while editing video. I needed my lip movements, audio, captions, and on-screen visuals to stay lined up. AI gave me the term multimodal synchronization, and I have used it as my broad phrase for that relationship ever since.

Scope still matters. If the problem is only the mouth and speech, lip sync is more precise. If the captions appear too early, caption synchronization names the smaller problem.

Length comes second. Meaning comes first.

Use the shortest term that preserves the full job. When AI gives you a candidate term, check its scope before you paste it back into your prompt.

Rule 2: Give AI One Contained Request

Make one single-purpose, self-contained request.

The laundry example makes this visible. Compare these 2 messages:

Honey, could you help with the laundry tonight, please?

Honey, could you collect all the laundry? I'm so tired. But I'm wondering maybe if we could have the washing machine run at some point, maybe tonight. What do you think?

Which message gives you one obvious job?

The first asks for one action with one time boundary. The second asks you to collect the laundry, respond to the tiredness, decide when to run the machine, and give an opinion about the timing.

AI faces the same structural ambiguity. It may answer the last question, react to the tiredness, act on the first request it noticed, or attempt all 4 jobs in one result.

Single-purpose means the request has one action or decision. Self-contained means the information needed to complete that action is included, attached, or available at an exact location the AI tool can access.

A difficult request can still be long. One deliverable can still contain several parts. A project update with a summary, risks, and next actions is one project update.

The request becomes mixed when you also ask AI to decide whether the launch should move and tell you whether your plan makes sense. Keep the result you need first. Move the other purposes into follow-up requests.

One message. One purpose. Enough information to complete it.

Copy This 2-Rule Prompting Cheat Sheet

Copy this block into your notes and run it before you send a prompt:

THE 2-RULE PROMPTING CHECK

1. Use the shortest established term that still says exactly what you mean.

My improvised description:
[paste the phrase or sentence]

The established term:
[add the shortest accurate term]

If I do not know it, ask:
"Is there an established term for this?"

Check: Does the term preserve my full meaning and the right scope?

2. Make one single-purpose, self-contained request.

The one action or decision:
[name it]

The information AI cannot know:
[include it, attach it, or give the exact accessible location]

The intended result:
[name what should exist when the work is done]

The boundaries that matter:
[state only the limits whose absence could make the result wrong]

Final request:
[paste the revised request]

Run the checks in order. Fix the term first. Then make sure the request asks for one result and carries what that result needs.

If you want to test these rules on one real request and turn it into a reusable prompt, open the members-only Prompt Engineering: The Only 2 Rules You Need course.

It continues from this cheat sheet with the full eight-lesson method.

Open the last prompt you sent. Run these 2 checks before you send it again.