Kashi Keefe← All resources
Guides·Claude·6 min read

Getting the Most From Claude Fable 5

What Fable 5 actually changes and how to put its new strengths to work.

Anthropic ships new Claude models often enough that it's easy to tune them out. Fable 5 is one worth paying attention to. It's the creative-leaning member of the family, and if you write, sell, or build with words, it changes what you can hand off.

This is a practical walkthrough, not a spec sheet. The goal is simple: show you what actually shifts with Fable 5 and how to prompt it so you get the upside instead of fighting it.

What Fable 5 is actually for

Every Claude model can write. Fable 5 leans into it. Think long-form drafting, voice and tone control, narrative structure, and idea generation. It holds a thread across a long piece without going generic, and it's better at sounding like a specific person instead of a polished, anonymous brand voice.

That makes it a strong fit for the work that usually eats your week: sales pages, email sequences, scripts, long posts, course outlines, naming, and first-draft concepting. It's not a different kind of tool. It's the same conversation you already know, pointed at the creative end of the dial.

Where it earns its keep

Reach for Fable 5 when the output is judged on feel, not just correctness. Here's where it tends to pull ahead:

  • Long-form copy that has to stay sharp from the headline to the close, not fade into filler halfway down
  • Matching a specific voice — yours, a client's, a character's — when you feed it real examples to study
  • Turning one rough idea into ten angles you can actually choose between, instead of three obvious ones
  • Story-shaped content: case studies, founder narratives, VSL scripts, anything that needs a beginning, tension, and a turn
  • Rewrites and tone shifts — same message, dialed warmer, punchier, plainer, or more editorial on demand

For pure data extraction, classification, or tight structured output, you don't need the creative tier. Match the model to the job.

How to prompt it well

The biggest mistake people make with a stronger model is over-directing it. Old prompts written to drag mediocre output up to acceptable are often too prescriptive here — they cap the ceiling. Say what you want and who it's for, then get out of the way.

Give it the reason behind the ask, not just the task. "Write a welcome email" gets you a welcome email. "Write the first email new buyers see right after a $27 purchase — they're nervous they wasted money, and I need them to feel like they made a smart call and open email two" gets you something that actually does a job.

And feed it voice samples. If you want it to sound like you, paste two or three things you've actually written and tell it to study the rhythm, not copy the words. Voice control is one of its real strengths, but it can only match a voice it's seen.

A prompt you can paste

Here's a starting template for voice-matched long-form copy. Swap in your specifics and let it run.

You're writing as me. Study the voice in these samples — match the rhythm, sentence length, and attitude, not the exact words:

[paste 2-3 samples of your own writing]

Now write: [the piece — e.g. a 600-word sales email]

Context:
- Who it's for: [audience, and what they're feeling right now]
- What I need them to do or feel by the end: [the goal]
- What to avoid: hype, fake urgency, guru-speak, empty adjectives

Write the full draft first. Don't explain your choices or give me options — just the piece. I'll ask for revisions after.

Working with it on bigger jobs

On long or multi-step work, treat it less like a vending machine and more like a sharp writer you've handed a brief. Give it the full picture up front in one well-specified pass instead of dribbling instructions in over ten messages — it does better with the whole goal than with a slow reveal.

Then iterate in passes. Get the draft, react to it, ask for the specific shift. "Tighten the open," "make the third section land harder," "cut a third without losing the argument." It responds well to direction once something concrete is on the page, and that loop is faster than trying to specify perfection in advance.

Rule of thumb: prescribe the outcome, not every step. Stronger models reward a clear goal and punish a rigid script.

Start with one real project this week — a sales page, a launch sequence, a batch of scripts — and run it through with voice samples and a real brief. You'll learn more from one honest pass than from any overview, including this one.