The RTF Framework: A Simple Way to Get Better Results From AI Tools

I've run enough AI workshops at this point to notice the exact moment someone in the room goes from skeptical to genuinely interested, and it's almost never because of a fancy new model. It's because I showed them one small change to how they were asking for things.

Most first attempts at a prompt sound something like: "Write something about our new product launch." That's not really an instruction, it's a topic. The model has no idea who it's supposed to sound like, what the output is even for, or what "done" looks like — so it guesses at all three, and it usually guesses wrong.

Role, Task, Format

RTF is the simplest fix I've found for the single most common failure mode I see.

Role — who should the model act as? "You're an experienced customer support lead" produces a genuinely different answer than "you're a marketing copywriter," for the exact same question.

Task — what, specifically, do you want done? Not "write about the launch," but "write three subject lines for an email announcing the launch to our existing customers."

Format — what should the output actually look like? A bulleted list? A table? Under 100 words? A certain tone? Skip this and you'll often get a perfectly well-written answer to a slightly different question than the one you meant to ask.

Put together

"You're an experienced customer support lead (Role). Write three short email subject lines announcing our new product launch to existing customers (Task). Keep each under 8 words, no exclamation points, list them as a numbered list (Format)."

That's maybe fifteen extra seconds of typing, and it reliably saves two or three rounds of "no, more like this."

Why I bother training teams on this

The gap between someone who's internalized this and someone who hasn't is enormous, and it has nothing to do with technical skill. It's a five-minute habit. Which is exactly why workforce AI training pays for itself so quickly — most organizations I walk into don't have a technology problem, they have a "nobody ever actually showed us how to ask" problem.

RTF is a starting point, not the whole picture. Standardizing further with something like the CREATE framework, and building real guardrails around what's appropriate to use AI for in the first place, are the natural next steps for a team that wants this to stick instead of fading once the novelty wears off.