Prompt Template Design Guide Part 4: "Outputting Deliverables" - Specifying Formats with "Plating Instructions"

Hello, I am Pokego, and I am researching how to write high-quality articles using generative AI.

As the fourth (and final) installment of the prompt template design guide distributed in my magazine, today I would like to talk about the "Outputting Deliverables" section.

If you haven't read the third explanation on "Work Procedures" yet, please read that first.

Now, through the explanations so far, we have treated the AI as a skilled chef, and have:

• Shared our vision in the "Kick-off Meeting (= Prompt Explanation)"

• Provided ingredients in the form of "Fresh Ingredients (= Provided Information)"

• Instructed the cooking process with "Secret Recipes (= Work Procedures)"

We have completed all the preparations up to this point.

The collaboration with the AI is finally in its final stage!

Now that the best dish is complete, there is one final, important step remaining.

That is the theme of this article, "Outputting Deliverables," or in other words, "Plating Instructions filled with hospitality."

No matter how delicious a dish is, it would be ruined if it were sloppily thrown onto a plate, right?

The same applies to output from AI.

If you understand the role and design of these "Plating Instructions," you will be able to receive AI deliverables in a format that is easy for anyone to use and can be utilized in the next step!

Let's take a closer look.

Why do "plating instructions" determine the quality of AI output?

You may have been using prompt templates and wondered,

"Why is the output format specified in such detail at the end?"

Some of you might have had this question.

In fact, this "specification of output format" is the "spirit of hospitality"that makes AI deliverables easy to use and increases the productivity of the entire task.

Preventing "whimsical plating" by AI and ensuring reproducibility

AI is an excellent chef, but it actually has a slightly chatty and overly eager-to-please personality.

Without clear instructions, it might add unnecessary explanations out of good intentions, or output in a different format every time.

"I only want the JSON data, but it comes with polite explanations before and after..."
"I want it in a table format, but the layout is broken..."

Have you ever had an experience like this?

This isn't because the "AI is being whimsical," but because our instructions were ambiguous.

Clear "plating instructions" eliminate variations in AI interpretation and can be called the "lifeline"for obtaining deliverables of the same quality every time.

Once reproducibility is ensured, there is no need to repeat the same work over and over, and work efficiency improves dramatically.

This is exactly the value of the templates provided in this community!

Hospitality for the next step: Making the information relay run smoothly

Structured output is actually a form of consideration for your "future self" or the "next AI."

In a previous explanation, we talked about the "information relay." didn't we?

Imagine a scenario where you process information generated by an AI using another tool or handle it in a subsequent prompt.

If it is output in an organized format, the next step can proceed smoothly, but if the format is inconsistent, you will end up having to reformat it manually.

This is the same as a chef beautifully plating a finished dish to serve to a customer.

If you hand it over in a well-organized form, it is easier for the next person to receive and enjoy.

Especially when linking multiple templates, this spirit of "hospitality" significantly impacts the productivity of the entire workflow.

Organizing AI thinking and improving output consistency

Clearly specifying the output format has another important effect.

That is the effect of organizing the AI's own thought process.

When there is a constraint to "ultimately output in this format," the AI proceeds with its thinking while organizing information to fit that format.

As a result, a virtuous cycle is created where logical consistency is maintained and information omissions are reduced.

This is similar to when humans create documents; if you decide that you will "ultimately summarize this into a presentation," you can collect information with the key points in mind from the very beginning.

Specifying the output format is not just a matter of "appearance"; it is a crucial element that improves the quality of the AI's thinking itself.

A deep dive into the three design elements of "Plating Instructions"!

Now, let's dissect the specific structure of the "Output" section, which serves as the plating instructions.

In our prompt templates, we use a simple, minimal structure so that even those who have just started using AI can understand it.

By simply understanding the roles of these three design elements, anyone can freely control AI output.

The "Plate Layout" that directly shows the finished form: Markdown templates

The first design element is the "Plate Layout".

This is like the "dividers on a plate" or a "layout diagram" for arranging food.

In the template, we show the finished form we want the AI to output directly in Markdown format.

For example, if you want the output in a table format, you write the actual skeleton of the table directly into the prompt.

By showing the real thing and saying, "Please create it according to this sample," you eliminate any room for the AI to get confused and ensure it can always generate output with the same structure.

Showing the finished template directly is overwhelmingly more reliable than abstractly telling it to "use a table format"!

At first, I also tried using abstract instructions like "output in a table format," but the format often broke. Since switching to this template method, stability has improved dramatically.

"Sticky Notes" that Connect Thought and Output: The Role of Placeholders

The next design element is the "sticky note."

This is like a "sticky note" placed on a layout diagram that says,"Put the grilled fish here" or "Put the simmered dish here."

In the template,"[Describe the 'XX' thought of in Step 1]" serves this role.

This allows you to accurately instruct where to place the parts (fragments of thought) created in each step of the "work procedure" recipe on the final plate.

Placeholders play an important role in bridging the AI's thought process and the final deliverable.

Especially when organizing complex information, using this "sticky note" method ensures the AI places information accurately without getting lost.

By logically connecting the AI's thoughts and output, the consistency and quality of the output are guaranteed.

This is the true value of the template!

"Wrapping" the Deliverable: The Convenience of Code Blocks

The final design element is "wrapping."

This is like the "special tiered food box" or "beautiful wrapping" that a chef uses to deliver a finished dish to a customer in the best possible condition.

In the template, the final deliverable is enclosed with the symbol"```markdown".

Example of a prompt template for search keyword analysis:

```markdown
## 「[検索キーワード]」検索キーワード分析結果

### 1. このキーワードで検索する人の主な目的は？（検索意図）
- **一番の目的:** [Step1で考えた「検索の目的」]
- **検索するときの気持ち（推測）:** [Step1で考えた「気持ち」]
- **どれくらい急いでる？（緊急度）:** [Step1で考えた「緊急度（高/中/低）」] （理由: [緊急度の簡単な理由]）

### 2. どんな情報を求めている？（ユーザーニーズ）
[Step2で考えた「ユーザーニーズ」を1〜2文で]

### 3. 記事に入れると良さそうな内容は？（主要トピック案）
[ここにStep3で考えた「主要トピック案」を箇条書きで3〜5個記述]
- [トピック1]
- [トピック2]
- [トピック3]
- [トピック4（必要に応じて）]
- [トピック5（必要に応じて）]

### 4. どんな人向けの記事にする？（想定読者像）
[Step4で考えた「想定読者像」を一文で]

```

Why do we use this kind of 'wrapping'?

Actually, there are two major benefits to this.

🟢 Ease of retrieval (Convenience)

Just like quickly taking home a wrapped item, there is the convenience of being able to easily copy the entire deliverable with a single click.

You can extract exactly what you need without including unnecessary preambles or postscripts.

In my experience, by instructing the AI to enclose the output in code blocks, its 'chatter' is dramatically reduced, allowing you to receive only the pure deliverable.

🟢 Content assurance (Visibility)

There is a sense of security in being able to see at a glance where one finished product begins and ends.

Within a chat conversation, you can clearly distinguish the AI's formal response (deliverable) so there is no confusion.

Also, when processing text in Markdown editors or programs, text within code blocks is easier to handle, which significantly increases the efficiency of subsequent processes.

Summary: Ensure reproducibility by providing 'samples' for output format specifications.

So far, as a design guide for the 'Work Procedure' section of a prompt template, we have covered:

• Why specifying the output format influences AI quality

• The concept of 'presentation instructions' as the basis for instructions

• Three Design Elements That Make Up Plating Instructions

Based on my research into prompts for writing high-quality articles with generative AI, I have shared the above with you using easy-to-understand examples.

Specifying an output format for AI is the same as "plating instructions" in cooking. "Plating Instructions" is the same as that.

By concretely showing a "sample" of the finished form, you can eliminate AI interpretation errors and obtain highly reproducible output.

Specifying the output format is not just a technique, but something like a spirit of "hospitality" toward the AI. "Hospitality" toward the AI is something like that.

Please make this design philosophy your own and use it to level up your AI utilization!

Over the past four parts, I have explained the four major structures of prompt templates.

Once you understand the components of a prompt, the next step is practice!

Please read other articles as well and let's deepen our relationship with our best partner, AI.

Thank you for reading until the end.

✅ Click here for Poke-Go's content