How to actually prompt Claude and what changes with Fable 5

I learned prompting the hard way.

Just doing. A lot of doing. Testing what worked, reading everything I could find, learning from others who were further ahead.

And somewhere along the way, the difference between an okay prompt and a great one became obvious.

Here's what actually moves the needle:

Give Claude a role and a reason

Don't just ask. Tell it who it is and why it matters.

"You are a conversion copywriter. I'm writing a LinkedIn post for recruitment managers who hate slow hiring processes. Help me write an opening line that stops the scroll."

Context changes everything. The model reads intent, so give it something to work with.

Use examples, not just instructions

Telling Claude what you want is fine. Showing it is better.

If you want a specific tone, paste a sentence that has that tone. If you want a specific format, show the format. One good example outperforms three paragraphs of description.

Structure complex prompts with XML tags

When a prompt has multiple parts — instructions, context, examples, input — separate them clearly.

`<instructions> <context> <example> <input>`

It sounds technical. It isn't. It just removes ambiguity, and ambiguity is where output quality dies.

Put the long content at the top

If you're feeding Claude a document, a transcript, or a data set, put it at the top of your prompt, before your instructions. Not after.

This one change alone can improve response quality by a significant margin on complex inputs. Most people do it the other way around.

Ask Claude to think before it answers

For anything complex, add: "Think through this step by step before responding."

It sounds almost too simple. But it forces the model to reason rather than react and the output difference is real.

What changes with Claude Fable 5

Now, with Claude Fable 5, Anthropic's newest model, one thing shifts: these principles still hold. But the overcorrecting stops.

Earlier models needed pushing. "ALWAYS do this." "CRITICAL: use this tool when..." That kind of scaffolding was sometimes necessary to get reliable output.

Fable 5 reads intent well enough that over-instructing now actively hurts quality. The model is built for long, complex, autonomous tasks — multi-hour runs, hard problems, workflows that would have taken days of iteration before. Give it the destination. It figures out the route.

What's also worth knowing: Fable 5 is not where the difference shows up for everyday tasks like writing, summarizing, or building content. For that, the gap between models is small. The real shift is in agentic, technical work — long coding runs, multi-step research, complex document workflows. That's where the ceiling moved.

AI moves fast. The way we use it moves slower. But the fundamentals don't change as fast as the models do — and that's actually reassuring.

👇 How did you learn prompting — by doing, by reading, or did someone show you the way?