You can't say AI is bad if you're not getting good results. That's on you. The models improve every month. The people using them mostly don't.

You know the feeling. You open a chat, type what you want, and get back something almost right that needs three more rounds to fix. You've had the brain fry that comes from wrestling a tool that should be saving you time. Maybe you bought a course to sort it out and hit the same wall. The tool got smarter. Your instructions stayed the same.

The good news is the answer is written down. Anthropic publishes its own advice on getting the best from Claude Fable, its most capable model. OpenAI does the same for its ChatGPT models. Learning this has a high return on your time. I run an AI company, and the founders who get the most from these tools treat prompting as a skill worth practising. Fewer rounds with the machine. More time away from your laptop.

Claude Fable is Anthropic’s most powerful tool. It goes way beyond being a content assistant. Fable is capable of building tools, producing workbooks, and creating entire funnels. You can give it extensive instructions. Walk away and come back to a completed workflow. But only if you prompt it right.

Here are seven lines to put in your prompts. Steal them, adapt them, and watch the output change.

How to prompt Claude Fable for expert-level results

Let it act when it has what it needs

Claude Fable can spend too long planning. Give it a task with some room for judgement and it may map out options, weigh choices you never raised or ask for details it could reasonably decide for itself. The result is more discussion when you wanted a completed piece of work.

Tell it when to stop planning and start producing. Add: “Once you have enough information to complete the task, proceed with the build. Make reasonable decisions where needed, use your recommended approach, and explain the important choices alongside the completed item. Only ask me a question if you are missing information that is essential to producing a usable result.”

That will not stop every clarification question. It tells Claude to proceed when the missing details are optional and ask only when the task cannot be completed properly without them.

Match the effort to the job

Claude Fable has an effort setting that controls how much thinking it applies to your request. Higher effort produces more thorough answers, but takes longer and uses more of your allowance. High works for most jobs. The hardest problems want the top setting. Routine work wants less, and even the low setting on Fable holds up well.

Choose the effort level before you begin, then reinforce it in your prompt when the task needs special care. For quick jobs, tell it to “keep things fast and skip the deep thinking.” For hard ones, tell it to take its time, gather what it needs, and check its work before handing it back.

Say it once and keep it short

Fable follows instructions well enough that you can stop writing essays. Older models needed you to spell out every behaviour you wanted gone, being explicit with your instruction and repeating it often. This one takes a single clear line and applies it across the whole task.

Send the instruction once, and it remembers. Something like: “lead with the outcome, tell me what happened first, then the detail, and keep it readable rather than crushed into fragments and shorthand.” That one line does the job of a paragraph of rules. Your prompts get shorter and your output gets cleaner.

Make it show its evidence

Setting Fable a task that will take it a long time comes with a risk. Deep into a task, a model can tell you it finished something it never checked, or report progress that reads well and means nothing. Fable does this less than older models, and you can shut it down completely.

Make it prove the claim. Tell it that before it reports anything as done, it must “show a result from the session, and if a step failed or got skipped, include this information with the output.” Anthropic found this nearly wiped out invented status reports in testing. You see what happened, backed by the result that proves it.

Draw the lines it can't cross

Sometimes the model does more than you asked. It drafts the email you were only thinking about. It creates a backup you never wanted. Helpful in theory, a mess when you wanted an answer and got an action.

Set the limits up front. Explain that when you are describing a problem or thinking out loud, its job is to assess and report back, and to change nothing until you ask. Before it runs anything that alters your files or your setup, it should check that the evidence supports that exact step. Claude Fable is eager to please you. But you decide what it starts.

Tell it why you're asking

This is the one that changes the most. Claude Fable does better work when it knows why you want something. Hand it a bare instruction and it guesses at the reason. Hand it the reason and it finds routes to your goal you would never have spelled out.

Tell it the whole picture. Who the output is for, what it needs to do once it exists, where it fits in your business. Give it a prompt like: “I'm writing this for founders who have revenue but loose business systems. My previous writing has not held their attention. This assessment tool needs to be accessible and practical for them, as well as compelling enough to finish.” That gives the model something to aim at. Give it the why, and it builds toward the goal you named.

Give it a memory to build on

The best operators give the model somewhere to recall information from. Fable performs well when it can record what it learned on one run and use it on the next. Without that, every session starts cold and repeats old mistakes.

Build it a memory, and store it where the model can find it. If you use Claude Cowork, point it at a folder and let it write notes there. If you work in Projects, compile the lessons into the project library. During the session, tell it to “document the lessons with a one line summary at the top, record what worked alongside what you corrected, and update an old note rather than writing a new one.” Over weeks the model gets better at your work because it stops forgetting what you taught it.

Turn better prompts into expert-level results with Claude Fable

When the AI models are improving fast, make sure you are too. These tips help you get more from every prompt without all the back and forth. Seven lines, dropped into the prompts you already write. Use them and the same model hands you decisions faster, output you can trust, and a memory that builds week on week. Say what you want with more care and Claude Fable does more of your work for you.

Get my free playbook for founders who want more from AI with fewer hours at the laptop.