10 Best Prompts to Maximize Claude Fable 5 Performance by 107x

The availability of Claude Fable 5 has been extended until July 12th.

However, I can say this for certain: most people are using this historically powerful AI as just a "smart chat" and aren't even drawing out half of its performance.

I didn't want to waste the incredible performance of Fable 5 either.

With that in mind, I scoured every X article and read through primary information on English-speaking overseas sites. I thoroughly investigated what people who "really get it" are using Fable 5 for.

I arrived at the 10 use cases I'm introducing today. And I've actually tried all of them in my own work.

A major cleanup of 95 misplaced files in work folders. Extracting 69 improvement points for video editing manuals. Compressing accumulated revision feedback by 94%. These are all my actual results from using these 10 methods.

Let me say this upfront: this isn't about programming.

In this article, I'll leave you with those 10 use cases and the prompts you can copy and paste to reproduce them.

It's going to be long, so I recommend saving it if you want to look back later.

Conclusion: Don't Entrust Fable 5 with "Tasks"

Using Fable 5 for mass-producing articles or small daily tasks is a waste. You can do that with regular models.

What only Fable 5 can do is "brain work" like design, auditing, and distillation.

Designing a system from scratch. Thoroughly inspecting massive amounts of files over a long period. Compressing scattered records into a single rule. The more a job requires long, deep thinking, the more the difference shows.

I chose based on data, not feelings. In coding ability tests, the commonly used Opus 4.8 scored 69.2%, while Fable 5 scored 80.3%. This difference isn't visible in simple tasks. It only becomes apparent when you entrust it with a complex job as a whole.

What's interesting is that whether I read X articles or dug into overseas primary info, everything led back here. Though expressions varied, everyone who understood it said, "Use Fable 5 for thinking work."

Here are the 10 use cases I reached after thorough research and testing in my own work.

1: Cleaning Up Entire Work Folders (95 Misplacements)

I had it read a massive folder for 2 businesses and 5 departments, identifying and moving 95 files that were in the wrong place. I had it keep a restoration log of "Original Location -> New Location" for every move, so I could revert if anything went wrong. The key was not letting it move things automatically; I always had to approve it.

A job that would take a human two full days was finished in a few hours through conversation.

1You are the audit manager for my work folders.
2Please read this entire folder and perform a "comprehensive structural audit" in the following order:
3
41. First, list the current folder structure and the role of each folder.
52. List all "files whose location does not match their role," providing a suggested destination and reason for each.
63. Move only the items I approve (do not move anything automatically).
74. Once moved, record all instances in a single file as a restoration log of "Original Location -> New Location."
85. Finally, summarize the "rules for deciding what to put where" in a single md file so files don't get lost in the future.
9
10Do not delete anything. For anything you are unsure about, separate it into a "Pending List."
11Before starting the work, please ask me any questions you have for confirmation.

2: Auditing the Most Complex Work Manuals (69 Improvements)

I had it thoroughly audit a set of video editing procedures, which is the most complex part of my work. From the perspective of someone actually doing the work, I had it list "ambiguous instructions, contradictions, omissions, and error-prone areas," and it found 69 improvement points. These were for something I thought was already perfect.

This is where I established the division of labor: "Audit and planning for the strongest model, implementation for the regular model." Use the smart model's time only for the stages that require intelligence.

1Please thoroughly audit my "[Task Name] Work Procedures (Manual/Template)."
2
3Method:
41. Read the manual from beginning to end and, from the perspective of someone actually performing the task, list all "ambiguous instructions," "contradictions," "missing steps," and "error-prone areas."
52. For each issue raised, provide a set of "Why it's a problem / What happens if left alone / Suggested fix."
63. Sort them by severity (High, Medium, Low) and summarize them as an improvement plan in an md file.
74. Do not implement the fixes yet. First, just provide the plan and wait for my approval.
8
9No need to hold back. Don't inflate the numbers just for the sake of it, but if there are truly problems, list even 100 of them.
10Before starting the work, please ask me any questions you have for confirmation.

3: Batch Quality Checking Self-Made Systems and Templates

I had it check the quality of all the procedures, templates, and prompt collections I've built up. Outdated information, references to non-existent files, and duplicate management. An inspection that would take weeks if done one by one was finished in a day.

Self-made systems start becoming obsolete the moment they are created. Only those who don't know this continue to waste time using broken systems.

1Please find all "self-made systems" (procedures, templates, prompt collections, scripts) in my folder and perform a batch quality check.
2
3Checkpoints:
4- Is there outdated information left as is (ended service names, changed prices, old names)?
5- Does it refer to non-existent files or folders?
6- Are there omissions or contradictions in the procedures?
7- Is the same content being managed twice across multiple files?
8
9Output:
101. A list of all files and their respective points of concern.
112. Categorize them into three levels: "Fix immediately," "Better to fix," and "No problem."
123. Once I give the OK, execute the fixes starting from the top.
13
14Before starting the work, please ask me any questions you have for confirmation.

4: Distilling Accumulated Feedback into a "Fixed Rule Set" (94% Compression)

I had it read the mountain of "records of my revision instructions" accumulated for newsletters, SNS posts, and reels, and distill them into a single "Fixed Rule Set." This resulted in a 94% compression.

The reason this is effective is simple: the files you have the AI read each time become lighter and stronger. Knowledge is king. Before polishing your prompts, organize your knowledge.

1Please read all the "records of my revision feedback" accumulated in this folder and distill them into a single "Fixed Rule Set."
2
3Method:
41. Read all feedback and turn recurring points into rules.
52. Pick up points that only appeared once as "exception rules" (do not discard them).
63. If there is conflicting feedback, adopt the one with the newer date and keep the unadopted one as "change history."
74. The final form should be a state where "reading only this rule set allows you to follow all past feedback."
85. Attach one NG example and one OK example to each rule.
9
10The goal is compression. Reduce the volume to one-tenth without losing the meaning of the original records.
11Before starting the work, please ask me any questions you have for confirmation.

5: Verbalizing the "My Style" of Course Scripts

I had it analyze 29 revisions I made to course video scripts and verbalize "my script style" into a single rule set. How I grab attention at the beginning, the order of explanation, and words I never use. Habits I couldn't explain myself were all put into words.

From next time, I just need to have the AI writing the new script read this one file. The accuracy of the first draft becomes something else entirely.

1I will provide scripts for course videos (or seminars/content) I've made in the past, along with the records of revisions I made to them.
2
3Please analyze all these differences (AI's first draft -> my final draft) and verbalize "my script style" into a single rule set.
4
5What to analyze:
6- Habits in the opening hook (how many seconds to get to the main point, the pattern of the first sentence)
7- Habits in the order of explanation (conclusion first, or story first)
8- Habits in phrasing (frequently used conjunctions, words never used)
9- Length per script and how paragraphs are broken up
10- Tone of addressing and encouraging students
11
12The completed rule set will be used as the "master copy" for the AI to read first when writing new scripts. Please write it with that in mind.
13Before starting the work, please ask me any questions you have for confirmation.

6: Turning Meeting Post-Processing into a "One-Word Task"

After a client meeting, there were 6 post-processing steps I always did: saving the transcript, adding to the Q&A, updating knowledge, recording in the management table, adding minutes to the chart, and storing the recording. It was an hour of work.

As a result of having Fable 5 systematize this, it now ends with a single word: "Process yesterday's meeting." It's like stopping the practice of writing the same handover notes for an assistant every time and making the handover notes themselves the system.

1After a meeting with a client, I perform these post-processing steps every time:
2(Please rewrite these to match your process)
3- Save the recording transcript
4- Add questions and answers to the "Q&A Collection" file
5- Record the number of meetings in the management table for each client
6- Add minutes to the client chart
7
8Please turn this series of steps into a "system where everything finishes just by telling you the date and the person I met with."
9
10Process:
111. First, interview me to identify all steps of my post-processing.
122. Determine if each step can be automated, semi-automated, or if manual work remains.
133. Build the system and test it with one recent meeting.
144. Save it as a procedure manual so it can be reused from next time.
15
16Before starting the work, please ask me any questions you have for confirmation.

7: Creating a Content "Repurposing Factory" and "Inspection Officer"

I had it create a "Repurposing Factory" that takes one piece of content and outputs it with different styles, lengths, and structures for multiple media outlets. I also had it create an "Inspection Officer" to check drafts against my NG rules before publishing.

The Inspection Officer is designed so that the checklist automatically grows every time I give a new revision instruction. The more I use it, the closer it gets to my standards.

1Please create two systems.
2
3[System 1: Repurposing Factory]
4A system where, when I provide one piece of content (newsletter, learnings from a meeting, script for an audio broadcast, etc.), it outputs the content rewritten for multiple media outlets I use (e.g., X, Instagram, newsletter, blog) according to their respective styles, lengths, and structures. It should also automatically maintain a ledger of what content has been repurposed for what.
5
6[System 2: Pre-publication Inspection Officer]
7A system where, when I provide a draft before publication, it inspects it against my NG rules (words not to use, fact-checking, broken links, typos, prohibitions for each medium) using a checklist and points out problematic areas. Every time I provide a new revision instruction, it should automatically add to and grow the checklist.
8
9First, interview me about my media outlets and style rules before creating these.
10Before starting the work, please ask me any questions you have for confirmation.

8: Systematizing Management with "Morning Orders" and "Weekly Reports"

I had it create a system that reads tasks and deadlines for all departments and issues a single "Order Sheet" when I ask "What should I do today?" every morning. A report to look back on the week at the weekend. A directory that automatically updates the list of systems I've created. I had it make this set of three.

The biggest change was that morning hesitation disappeared. Decision-making changed from "feeling" to "system."

1After reading my work folder, please create three "Business OS" components.
2
3[1. Morning Order Sheet]
4A system where, if I ask "What should I do today?" every morning, it reads the task lists and deadlines for each department and issues a single order sheet for "things to do today, things that can be postponed, and things to confirm with me."
5
6[2. Weekly Report]
7A system where, if I say "Review this week" at the end of the week, it summarizes what was done this week, numerical movements, and priorities for next week in one page.
8
9[3. System Directory]
10A system that automatically indexes the list of self-made systems and templates I've created so far, allowing me to see at a glance "what systems exist." It should update automatically every time I create a new one.
11
12Save all three as procedures so they can be output with the same quality every time.
13Before starting the work, please ask me any questions you have for confirmation.

9: Inventorying All Work as a Business Improvement Architect

I gave Fable 5 the role of "Business Improvement Architect and QA Lead" and had it analyze all my work. It listed all tasks, categorized them into "fully automatable, semi-automatable, or should be done by a human," and actually built the systems starting from those with the largest time-saving impact.

To be honest, several of the systems I've introduced so far were born from this inventory. You can even leave it to the AI to think about what to leave to the AI.

1You are an AI that combines the roles of Business Improvement Architect, AI Automation Engineer, System Designer, and QA Lead.
2
3The goal is to analyze my current work and use your long-term thinking and implementation capabilities to significantly reduce my future work time.
4
5Process:
61. First, read my entire work folder and list all the tasks I usually perform.
72. Categorize each task into "fully automatable / semi-automatable / should be done by a human."
83. Prioritize them based on "time that can be saved x ease of implementation."
94. Starting from the top tasks, actually create the specific systems (procedures, templates, prompts).
10
11Do not end with just light brainstorming. Today, right here, finish creating at least the top three in a working state.
12Before starting the work, please ask me any questions you have for confirmation.

10: Having Fable 5 Create Content Explaining the Trending Fable 5

When Fable 5 became a hot topic, I had Fable 5 itself research the official announcements for fact-checking and create a whole set of content, from draft outlines for explanatory articles to a collection of prompts readers can copy and paste, and even scripts for explanatory videos. By adding my own thoughts to that, a high-precision article was completed.

And that article hit 500,000 impressions.

https://x.com/masaki_aihack/status/2072227526223729020

1Please create a set of explanatory content for my followers about the currently trending AI "[AI Name]."
2
3Process:
41. First, research official announcements and reliable reports online to fact-check (clearly separate speculation from facts. Include reference links).
52. Make it so you can explain "what's actually amazing about it" in 3 lines without using technical jargon.
63. Based on that, create the following:
7　- An outline for an explanatory article (with 3 title suggestions)
8　- A collection of practical prompts that readers can copy and paste to try
9　- A script for an explanatory video
10
11The readers are "people interested in AI but not experts." Be sure to include specific examples that convey the greatness.
12Before starting the work, please ask me any questions you have for confirmation.

The One Tip More Important Than Prompts

Did you think, "The prompts are surprisingly short"?

Actually, that's fine.

Smart AI works better with a "definition of goals and authority" than a "work instruction manual." Rather than writing 100 lines of detailed steps, you get better results by just giving it the role, purpose, things not to do, and completion conditions, and letting the AI think for itself.

I've built more components into those 10 prompts than meet the eye. An approval gate like "do not move automatically." Insurance like "record all instances in a restoration log." A prohibition like "do not delete anything." A completion condition like "a state where reading only this one allows you to follow everything." What beginner prompts lack isn't length; it's these guardrails.

However, there is just one tip.

After pasting, answer all the questions that come from the AI.

That's why I've included "please ask me any questions you have for confirmation" at the end of every prompt. A smart AI will ask rather than filling in missing information on its own. That's where you talk about the whole plan in your head. This is where a world of difference is made.

Think back to when you ask an external contractor to do a job. If you just say "make it look good," you'll never get something great. The work of a client who answers questions carefully turns out better. AI is exactly the same.

Summary: There Are Only 3 Types of Use Cases

To organize them:

1. Design: Meeting post-processing (6), Repurposing factory and inspection officer (7), Business OS (8), Work inventory (9)
2. Audit: Folder cleanup (1), Manual audit (2), Batch system check (3)
3. Distillation: Turning feedback into rules (4), Verbalizing script styles (5), Fact-checking and converting new info (10)

What they all have in common is that I didn't have it do a one-off task, but rather had it create the "system side."

If you have it write one article, the result is one article. But if you have it turn the way of writing articles into a rule, it works for all subsequent articles. I call this the "compound interest of systematization."

If you use a limited-time smart model for one-off tasks, it's over once the task is done. If you turn it into a system, it continues to remain as an asset even after the model's provision ends.

It's cruel, but this is reality. In the same week, people are divided into those who consume and those who create assets.

You can do it too. Starting with the folder cleanup in #1 is more than enough. There's still time until the 12th.

I hope you found this useful.

Finally, One Last Thing.

I've put all 10 of these prompts in this article without holding anything back. Copy and paste them and start using them today.

And for a limited time only.

I'm giving away a bundle of "20 bonuses" for mastering AI at work.

First: 11 complete guides for Claude / Codex / ChatGPT / Gemini.

Second: "100 God Prompts" that work just by copying and pasting.

Third: 6 practical AI tools you can use as they are.

Fourth: The entire process of launching an AI business and generating 3.27 million yen in the first month.

A total of 20 items. And they are all free. You can get them all without participating in seminars or free individual consultations. It sounds like a lie, doesn't it?

The Slide Creation GPTs among them are quite popular, and anyone can easily create slides like the one below.

Receiving them is simple. It starts with joining the LINE Open Chat below.

You can enter from here.

I'll be honest. None of the systems introduced in this article were completed in one go. I answered questions many times, corrected them many times, and nurtured them. Prompts are the entrance to that.

Let me say it just one more time.

What's needed isn't technology. It's just the ingenuity to not let the AI get lost.

Why not end the exhaustion of giving instructions to AI today?