Many CEOs and business owners are wondering if they should rewrite their prompts now that Fable 5.1 is out. Since our company runs nearly 60 automations using Claude, I conducted a check on this yesterday.
The answers are all in Anthropic's official guide. It contains 14 recommended prompts, 7 of which are applicable to the work of executives and business owners.
All 7 of these prompts share a surprising commonality: they don't specify the content of the work at all. Instead, they focus entirely on the process of how to work.
I work on the front lines of marketing support for various companies and have conducted AI training for over 5,000 people. In this article, I will translate those 7 official prompts into Japanese and explain what they fix. I've also prepared a copy-pasteable summary at the end, so please read until the end.
I post every morning about management, AI, and marketing. If you find this helpful, please follow me.
Chapter 1: The Identity of the Official Guide
First, let's clarify where these 7 prompts came from. It's dangerous to copy and use information without knowing the source.
There are 14 Recommended Prompts in Total
Anthropic has prepared a dedicated page for Fable 5.1, which includes specific text for "how to write to fix issues." These are 14 ready-to-use prompts, not just summaries.
The breakdown is 7 for developers and 7 for general work. We will only cover the latter here. The former deals with code editing and testing, which isn't relevant to management or administrative tasks.
Why They Aren't Circulating in Japanese
Simply because they are provided in English. While the page is accessible to anyone, it's located within developer documentation, so it's not somewhere executives or business owners usually look.
I actually researched nearly 90 posts about Fable 5.1 in Japanese on X, but I couldn't find any that covered these 14 prompts. Most posts were biased toward speed and pricing, leaving a clear gap in information.
Source for This Chapter
The original text for the recommended prompts is on this page: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5-1
Discussions on reasoning and boundaries are also on the page for the previous version. If you want to read the original, check here: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5
Chapter 2: Two Prompts to Fix the Habit of Working in Silence
Now for the main topic. Let's start with the most common problem when using Fable 5.1.
Making It Report Progress
Fable 5.1 doesn't report much during long tasks. This is a habit mentioned officially; it might look like it's stopped, but it's actually working. It makes the user anxious, but it's not a bug.
The fix is this: ask for one line on what it will do before starting. Provide short updates at each milestone. End with a summary that makes sense on its own. Just asking for these three things changes the rhythm of reporting.
Preventing It from Stopping Midway
The other issue is the opposite habit: stopping in the middle of a long task to ask, "Can I continue?" There have been cases where someone thought they delegated a task, only to find it hadn't moved for an hour because it was waiting for confirmation.
The official guide provides a solution: pre-determine that it should only stop in three cases: when deleting something, when an action is irreversible, or when the scope of the request changes. Telling it to proceed without asking otherwise significantly reduces these stops.
Chapter 3: Three Prompts to Improve Writing Quality
Next are three prompts that will resonate with CEOs and business owners who have AI write for them. The more you find yourself making the same manual edits to proposals or email drafts, the bigger the impact will be.
Eliminating Pretentious Phrasing
This is the core of this article.
The official guide explicitly names and bans a certain style of writing: replacing direct statements with metaphors. If someone says "dials that need to be turned" instead of being direct, the reader has to translate it in their head.
The official short version is surprisingly simple: just one line saying, "Remove all pretentious phrasing." Adding this at the end has visibly reduced the manual editing needed for proposals.
Stopping Excessive Bullet Points
Fable 5.1 uses bullet points less than before, but internal documents can still come back entirely as lists. When a consultation for a single reader is lined with bullet points, it looks cold.
The official instruction is to "use bullet points only when requested or when the content is so diverse that it's easier to understand." Specifying that it should use normal sentences for consultations or personal interactions keeps things stable.
Preventing Verbatim Copying of Source Text
When summarizing materials, the AI sometimes mixes in the original text as is. The official guide notes this habit has increased in Fable 5.1. This can lead to issues when using other companies' text.
The instruction is simple: have it rephrase in its own words and use short quotes with quotation marks only where the original text is absolutely necessary. Including this one sentence almost entirely eliminates verbatim copying.
Chapter 4: Two Prompts to Protect Scope and Freshness
The final two prompts stop the most costly errors in Fable 5.1 as tasks get longer.
Preventing Unauthorized Scope Creep
Sometimes the AI reaches beyond the requested scope or, conversely, only finishes a part of it. The longer the task, the more these differences add up at the end.
The official text is quite long, but there are three key points: the requested scope is the final product; if something is unclear, finish all unrelated parts first; and report unrequested improvements instead of implementing them. Providing these three points reduces misunderstandings in the final output.
Making It Research Familiar Names
With shallow settings, the AI tends to answer from memory for things it "knows." Since the status of AI and tools changes in months, being "slightly familiar" is the most dangerous state.
The official instruction is precise: it notes that being slightly familiar makes the AI give old answers with confidence. It's safer to include an instruction to always research before answering for questions centered on specific names.
Chapter 5: Summary of the 7 Prompts for Copy-Pasting
I've summarized the 7 prompts so you can paste them directly. You don't need to use all of them; just pick the ones you're struggling with.
The 7 Prompts
Add these to the end of your request when you open Claude. If you use them every time, it's easier to put them in the system instructions.
11. Before starting, write one line about what you are going to do. Provide short updates at each milestone during the process. End with a summary that is understandable on its own.232. Please proceed using your own judgment. I am not watching over you, so if you ask if you can proceed, the work will stop. Only stop in these three cases: when deleting, when an action is irreversible, or when the scope of the request changes.453. Remove all pretentious phrasing. Do not replace things with metaphors; write using direct language.674. Use bullet points only when requested or when the content is so diverse that it is easier to understand that way. For consultations or personal interactions, write in normal sentences.895. When summarizing materials, do not copy the original text as is. Rephrase it in your own words. Only use short quotes with quotation marks where the original text is absolutely necessary.10116. The requested scope is the final product. Do not narrow or expand it on your own. If something becomes unclear midway, proceed with all unrelated parts first, then summarize your assumptions or questions at the end. If you find improvements I didn't ask for, write them in the report instead of fixing them.12137. For questions centered on names where you are not confident, or in fast-moving fields like AI and tools, always research before answering. A state of being 'slightly familiar' leads to giving old answers with confidence.
What the 7 Prompts Have in Common
If you look at them side-by-side, you'll see that none of the 7 prompts specify the content of the work. There isn't a single line about what to create or what structure to use. They only describe the process.
I believe this is the biggest change this time. Leave the content to the AI and just decide on the process. It's like the way you delegate work to a highly capable person. Once you understand this, you no longer need to over-engineer your prompts.
Chapter 6: Precautions for Use
I've provided them in a copy-pasteable format, but they won't work if you put them in the wrong way. I'll mention two things to avoid losing out.
Don't Put Everything in at Once
If you paste all 7, the instructions become too long, and accuracy actually drops. The official guide also states that quality decreases as you pile on detailed instructions. Choose only one or two that address the habits you're struggling with. Putting them all in at once is the most common mistake.
In my case, I often use #1 for progress and #6 for scope. Since I delegate many long tasks, these two have almost entirely stopped accidents.
The Result Changes Depending on Where You Paste
The official guide mentions something interesting: the result changes depending on where you paste the same text. Specifically, progress instructions worked more strongly when added to the end of each specific request rather than in the system settings.
Conversely, writing style instructions last longer when placed in the very first interaction of a conversation. It's safer to try adding them to the end of a request first, and if it's the same every time, move it to the settings. It was quite surprising that the same text could differ based on location.
Conclusion
Since you've read this far, here is one final conclusion.
Pick just one of the 7 prompts that you are currently struggling with most and paste it at the end of your request tomorrow. You don't have to do everything. If one makes a difference, that's enough to then add a second one.
Reading the official guide made me realize that what we should give to AI has completely changed. Before, we wrote in detail what to make and how. Now, we decide the process and leave the content to the AI.
This is almost the same as the order of improvement when delegating work to people. That's why I believe that from now on, the people who use AI well won't be those good at "prompting," but those who are good at delegating work.
I write daily about how to use AI and marketing in management, including stories of failure. If you want to read tomorrow's post, please follow me → @Charlie_no_site
