What you'll do: use the RTCF formula on a real business task and run your first iteration to improve the result.
⏱ 20 minutesAnthropic's own research into how people use AI found one behaviour that separates people who get great results from people who get average ones. They followed up. They didn't start over. They didn't give up. They sent a second message.
The skill isn't writing the perfect first prompt. It's knowing what to add when the answer isn't right. That's what this module is about.
Tell Claude what kind of expert to be.
State exactly what you want done.
Add the details that make the answer specific to you.
Tell Claude how to structure the output.
"Write me an email to a client who hasn't paid."
Result: Generic, formal, could be from any business.
"You're a friendly but firm bookkeeper (Role). Write a payment reminder email (Task). The client is 3 weeks late on a $600 invoice for a brand photography shoot. We have a good relationship and this is the first time it's happened (Context). Keep it under 4 sentences, warm tone, end with a clear next step (Format)."
Result: Sounds like a real business owner. Specific situation. Clear tone.
When the first answer isn't right, you don't start over. You add one instruction and send it again. Here are five follow-ups that fix most results:
One more thing about iteration: Each part of RTCF gets better the more you refine it. Your first Role is too generic. Iteration: "Actually, I'm specifically a B2B marketer." Your first Task was vague. Iteration: "Make this sound like an email I'd actually send." Your first Format got too long. Iteration: "Just 2 sentences." Each iteration teaches you what to include next time. By your third attempt at the same task, you're usually not iterating anymore. You're just running the prompt. That's the point.
Chat is forgiving, you can nudge it message by message. Cowork and Claude Code are different, they go off and do the work, so the brief you give upfront matters far more. Front-load it: say what you want, where it should go, what the finished result looks like, and any limits, all before they start.
Think of it like texting a contractor. "Fix the thing" gets you questions and delays. "Replace the kitchen tap with this model, it's under the sink, do it Tuesday morning" gets it done. Same with these tools: the fuller the brief, the fewer the back-and-forth rounds.
Each prompt is built on RTCF. The more Context you add, the more specific the answer.
You're a professional email assistant. Draft a reply to this email: [paste email]. My tone is [friendly/formal/firm]. Keep it under 3 sentences.
Write 3 Instagram captions for a [type of business] post about [topic]. Tone: [conversational/inspiring/educational]. Include a question at the end.
Add a Role to get stronger results: "You're a social media copywriter who specialises in [your industry]."
Research [competitor name]. Summarize: what they offer, who they target, and 2 things I could do differently. Bullet points.
Add: "I run [my type of business]" to get gaps that are relevant to you.
Create a 30-minute meeting agenda for [type of person]. Goal: [goal]. Include time allocations.
Add Role + Format: "You're an experienced facilitator. Format as a numbered list."
Create an outline for a proposal to [type of client] for [your service]. Include: problem, solution, deliverables, timeline, investment.
Add Role: "You're a senior consultant writing for a small business owner."
Prompt #2 (the social caption one) runs every week in my business. I paste a voice memo transcript, add 3 tone words from my Project, and get 3 options back in under 30 seconds. I pick the one that sounds most like me and post it. The first time I did it I spent 20 minutes tweaking the result. Now I barely touch it. That's what consistent context does.
Run the email reply prompt above. When you get the draft, send this follow-up instead of starting over:
Make it sound more like me. Here's an example of how I write: [paste one real sentence]. Use that tone.
Watch the second version. It should sound noticeably more like you.
Read the result out loud. Does it sound like you? Would your client know it wasn't written by you? If the answer is "not quite," that's your cue to iterate once more, not to accept it because the AI wrote it.
You're missing the Context part of RTCF. Your context should say things like: "I write in short sentences. I use the word 'honestly.' I start emails with a question when possible. Here's an example of an email I wrote..." Without context, Claude defaults to generic business voice. Add examples of YOUR voice, and the output shifts immediately.
That's a Format problem. Don't just say "make it shorter." Say "Give me exactly 2 sentences" or "Under 100 words" or "5 bullet points, 1 sentence each." Numbers in the Format section change everything.
Same reason your quiz score varied. Claude is thinking, not looking things up. If you need consistency, be more specific. Add "structure it like:" or "follow this pattern:" or "compare it to:" in your Comparison section. That narrows down the possibilities.
You probably didn't include all the context the second time. Go back to the prompt that worked. Copy it exactly. If it fails again, something changed. Claude's knowledge cutoff, your input data, your expectations. Identify what and adjust the prompt.
The Iteration Habit: RTCF gives you a framework, but frameworks need iteration. Your first prompt teaches you what to change. Your second prompt teaches you more. By attempt 3 or 4, you're usually getting what you want. That's not failure. That's how it works.
3 questions
I share what I'm learning in real time, new workflows, honest reviews, what worked and what didn't.
#BuiltwithSally