The CARE framework

Context. Audience. Role. Expectations. Four inputs that turn a generic AI answer into something you can actually use. The reason most owners conclude AI is not useful is almost always a CARE failure, not an AI failure. Here is how the discipline works.

Where to start

The CARE Framework

Roy Castleman's four-part prompt structure for UK owner-managers: Context, Audience, Role, Expectations. The discipline turns vague AI instructions into precise inputs that return usable outputs. Context explains the situation. Audience names who the output is for. Role sets the expertise the AI should bring. Expectations defines the shape, length, and tone of the answer. Sits inside the AI pillar alongside the EVOLVE Method, the 60/40 Principle, the Brain Dump Protocol, and the Business Brain. The communication layer of the method.

Most AI disappointment is an input problem

The single most common reason owner-managers conclude AI is not useful for their business is that they handed it a vague instruction and got a vague answer back. The instruction was something like, write me a proposal for this client. What came back was a generic proposal template with lorem-ipsum-flavoured paragraphs that could have been written for anyone. The owner read it, concluded AI cannot write proposals, and went back to doing it themselves for another six months.

The AI did not fail. The input failed. The same tool, given a different input, would have produced a draft the owner could actually send after ten minutes of editing. CARE is the discipline that reliably produces the better input. It is not a clever prompt trick. It is a structure for giving the AI everything a human expert would ask for if you handed them the same task.

Context. Audience. Role. Expectations. Four inputs, in that order, every time the task actually matters. What follows is what each one means in practice, why the order matters, and the shift it produces in the output.

The four inputs

C — Context

What is the situation? What has already happened? What is the background the answer needs to sit inside? Most AI instructions skip this entirely, which is why the answer floats free of anything real. Context is the difference between the AI drafting a proposal for an imaginary client and drafting a proposal for the specific client whose last meeting ended with an awkward silence when you mentioned budget.

Good context usually runs three or four sentences. Who is this for. What is the relationship so far. What constraints matter. What is already true that the answer has to respect. The fastest way to write this is to pretend you are briefing a new colleague who has just joined the call and needs to catch up before contributing.

A — Audience

Who is the output for? What do they care about? What do they already know? Audience is how the AI calibrates tone, vocabulary, and depth. An email to a long-standing client reads differently from an email to a finance director you have never met, and both read differently from an internal brief to your operations lead. Without an audience statement, the AI defaults to a neutral middle register that sounds slightly off to everyone.

The useful form is a single sentence. One-line audience description with the critical flag. "Long-standing client, direct communication style, values clarity over formality." "New finance director, formal register, wants numbers defended first."

R — Role

What expertise should the AI bring to the answer? This is the input most owners skip entirely, and it is the one that changes the quality fastest. "Act as a veteran commercial negotiator" produces a different answer from "act as a management consultant" or "act as a former client-services director who has run fifteen similar negotiations." The role activates the pattern-matching the AI does best.

Specific roles work better than generic ones. "Experienced estate agent who has sold prime central London properties for fifteen years" beats "real estate expert" by a margin that surprises people the first time they feel it. The AI is not inventing the expert. It is weighting the answer toward the patterns that expert would use.

E — Expectations

What should the answer look like when it arrives? Length. Format. Tone. Structure. Specific elements that must be included. Specific things that must be avoided. Without expectations, the AI defaults to whatever its training patterns say a typical answer looks like, which is often the wrong shape for your actual use.

Useful expectations are concrete. "Email, under two hundred words, opens with a personal line, closes with a clear next step, avoids any mention of pricing." "Strategy note, six to eight bullets, each bullet has a headline and two lines of reasoning, ends with the single recommendation I should take into the board meeting." The AI produces what you describe. Describe what you want.

Why the order matters

Context comes first because the rest of the instruction has nowhere to live without it. Audience comes second because it calibrates everything downstream. Role comes third because the expertise has to be applied to the situation and the audience, which have now been described. Expectations come last because they set the shape of the output after the thinking has been framed.

Owners who try to start with expectations (give me a 300-word email) without the other three inputs usually get a generic 300-word email. Owners who start with role alone (act as a negotiator) get a generic negotiator-voiced answer that does not match the situation. The sequence is the discipline. Running the four in order stops being effortful after about three weeks of practice, and at that point the framework has become how you think rather than what you consult.

This is the shift the Four Stages of AI Mastery describes moving from Conscious Competence to Unconscious Competence. The framework goes from checklist to instinct, and the time it takes to structure an instruction drops from three minutes to thirty seconds.

CARE plus Business Brain is most of the battle

CARE handles the instruction. It does not handle the context database. For a UK service business owner using AI seriously, CARE pairs with the Business Brain to close the input-quality loop. The Business Brain is where the persistent context about your business lives (voice, clients, offers, standards, frameworks) in a single Notion workspace any AI tool can read. CARE is the structure you use every time you ask the AI to do something specific with that context.

Together, the two eliminate roughly ninety per cent of the AI-is-not-useful complaints I hear on discovery calls. Context that lives in one place plus CARE-structured instructions produces output that actually matches the business it is for. The other ten per cent usually resolves once the EVOLVE Method is added on top: the process for moving through multiple AI conversations with intention rather than hoping a single prompt will land. CARE sits inside step two of EVOLVE (Voice the Problem) as the sentence-level discipline.

The short version

Context. Audience. Role. Expectations. Four inputs, in that order, every time the task matters. The framework is the difference between generic AI output and specific usable work. Thirty days of deliberate practice is usually enough to take CARE from checklist to instinct.

Pair CARE with a Business Brain and an owner-manager has everything they need to use AI as a thinking partner rather than a better search engine. Add the EVOLVE Method on top for the process-level discipline and the AI pillar is essentially complete as a working system.

AI for business owners The EVOLVE Method

Keep reading

AI Pillar · Live

The AI Pillar

The pillar hub. Where CARE sits inside the full AI method.

AI Pillar · Live

AI for Business Owners

The cornerstone that introduces CARE and the 60/40 Principle as the two core frameworks.

AI Pillar · Live

The EVOLVE Method

The process-level sibling. CARE lives inside step two of EVOLVE (Voice).

AI Pillar · Live

The Brain Dump Protocol

The starter practice. CARE is the discipline that makes every subsequent conversation more precise.

AI Pillar · Live

Business Brain

The context database. Context plus CARE is most of the battle for usable output.

Methodology · Live

Where to start

The Sequence Rule. Why reclaiming time through better AI conversations has to come first.

The CARE framework, answered

What is the CARE framework?+

A four-part prompt structure I teach to every owner-manager I coach: Context, Audience, Role, Expectations. The job of the framework is to take a vague AI instruction (the one that returns a generic, could-have-come-from-anyone answer) and turn it into a specific input that returns usable work. The reason most owners conclude AI is not useful is almost always a CARE failure, not an AI failure. You gave it nothing to work with, so it gave you nothing in return.

Why does CARE work better than clever prompts?+

Because it is not clever. It is complete. The problem with most prompting advice is that it teaches tricks, and tricks break the moment the task changes. CARE is a discipline that applies whether you are asking for a client proposal, a strategic analysis, a hiring brief, or a difficult email. The four inputs are the ones a human expert would ask for if you handed them the task. AI needs the same inputs, and it produces the same quality when it has them.

How long does it take to learn?+

The framework itself takes five minutes to understand. Using it deliberately in every conversation takes about thirty days to become the default. Most owners I coach use CARE as a checklist for the first three weeks, feel the difference in the output, and by week four the structure is informal: they simply think in terms of the four inputs before typing. At that point it has stopped being a framework and become how they communicate with AI. This is what the Conscious-Competence stage of the Four Stages of AI Mastery looks like in practice.

Should I use CARE every single time?+

For any conversation that matters, yes. For a quick factual lookup, the cost of full CARE outweighs the benefit. For a proposal, a strategy document, a delicate email, a hiring brief, or a tough client conversation, skipping CARE is almost always the move owners later regret. The rule of thumb I use: if I would brief a human colleague on it, I brief the AI with CARE. If I would Google it, I ask the AI directly.

How is CARE different from the EVOLVE Method?+

CARE is a sentence-level discipline. EVOLVE is a process-level one. CARE is how you structure a single instruction. EVOLVE is how you move from the problem in your head to the finished work through multiple AI conversations with intention. Most owners use CARE inside the Voice step of EVOLVE (step two). The two fit together: CARE builds the quality of each individual input, EVOLVE builds the quality of the overall thinking process. The dedicated EVOLVE cornerstone covers the full six steps.

Does CARE work for every AI tool?+

Yes, and this is part of why I teach it. ChatGPT, Claude, Gemini, Perplexity, Copilot, and every other general-purpose AI tool produces noticeably better output with CARE-structured input. The framework is not tool-specific because the problem it solves is not tool-specific. It solves the input quality problem, which exists whatever AI sits on the other side of the conversation. Owners I coach swap tools without re-learning how to prompt, because the thinking discipline transfers.

Where does CARE sit inside the AI pillar?+

CARE is one of the core method entities of the AI pillar, alongside the EVOLVE Method, the 60/40 Principle, the Brain Dump Protocol, and the Business Brain. It is the communication layer. It is usually the second thing I teach, after the Brain Dump Protocol, because CARE is the discipline that makes every subsequent AI conversation more precise. A Business Brain (context database) combined with CARE (instruction discipline) is what separates owners who use AI as a thinking partner from owners who use it as a better search engine.