Give ten people in your firm the same AI tool and you’ll get ten different levels of output quality. Half will treat it like a search bar and get generic answers. A quarter will write elaborate prompts that take longer than doing the work themselves. The rest will land somewhere in between, and nobody’s results will be consistent enough to rely on. The problem isn’t the tool. It’s that most firms roll out AI without teaching people how to write a prompt that actually works.
The CRIT framework solves that. It’s a four-part structure for writing prompts that produce consistent, usable results: Context, Role, Interview, and Task. Firms that train their teams to use CRIT stop getting random one-off answers and start getting output they can actually use in client work.
What does Context mean in a prompt?
Context is the background, constraints, and success criteria the AI needs before it can give you a useful answer. Without it, the AI guesses at what you want, and the guess is usually wrong.
Picture a six-attorney estate planning firm that asks an AI to “draft an engagement letter.” The AI will produce something, but it won’t know whether the letter is for a $2,500 simple will or a $75,000 complex trust, whether the firm’s standard payment terms are flat fee or hourly, or whether the letter needs to comply with a specific state bar format. The output will be generic, and the attorney will spend 30 minutes rewriting it instead of five minutes editing a solid first draft.
Context tells the AI what the situation is, who the audience is, what constraints apply, and what good output looks like. For the engagement letter, that means specifying the service type, fee structure, jurisdiction, and tone. The more specific the context, the less time you spend fixing the result.
What does Role do in a prompt?
Role tells the AI what expertise to apply. “Act as a financial analyst” produces different output than “act as a copywriter” or “act as an executive coach,” even if the task and context are identical.
A CPA firm drafting a client advisory email about tax planning might ask the AI to act as a tax advisor writing to a business owner with no accounting background. That role instruction shapes the vocabulary, the level of detail, and the structure of the explanation. Without it, the AI defaults to a tone that’s either too technical or too casual, and the email doesn’t land the way the firm needs it to.
Role works because it activates a specific pattern of language and reasoning in the AI’s training data. The more precisely you define the role, the more consistent the output becomes across different people using the same prompt.
Why does Interview improve the output?
Interview means asking the AI to ask you a few clarifying questions before it answers. It’s the fastest way to catch gaps in your prompt without writing a 300-word instruction set upfront.
A litigation firm drafting a discovery response might tell the AI, “Before you draft this, ask me three questions about the case facts, the response deadline, and any sensitive issues we need to avoid.” The AI will ask, the attorney answers in two sentences each, and the draft that comes back is dramatically closer to usable than it would have been if the AI had guessed.
Interview also trains your team to think about what the AI actually needs to know. After running a handful of prompts with the Interview step, people start including that information in the Context section automatically, which makes the whole process faster.
What does Task define in a prompt?
Task is the definition of done: format, length, structure, and deliverable. It’s the difference between getting three paragraphs of general advice and getting a formatted, ready-to-send engagement letter.
For the estate planning engagement letter example, Task might be: “Draft a two-page engagement letter in plain language, structured as an introduction, scope of work, fee schedule, client responsibilities, and signature block. Use bullet points for the scope section. Keep the tone professional but approachable.” That level of specificity means the output doesn’t need to be rebuilt from scratch.
Without Task, the AI picks its own format and length, and the result rarely matches what the firm actually needs. With it, the output is usable in the first pass, not the third.
What does a full CRIT prompt look like in practice?
Here’s a worked example for a law firm drafting an engagement letter for a new estate planning client.
Context: We are a six-attorney estate planning firm in California. This engagement letter is for a married couple in their early 60s who need a revocable living trust, pour-over wills, powers of attorney, and healthcare directives. Our flat fee for this package is $4,500, with half due at signing and half due at document execution. The clients have no prior estate planning documents. The letter needs to comply with California State Bar guidelines on engagement letters.
Role: Act as an estate planning attorney writing to clients who are not familiar with legal process or terminology.
Interview: Before drafting, ask me three questions to clarify anything about the clients’ situation, our firm’s process, or the tone we want to strike.
Task: Draft a two-page engagement letter in plain language. Structure it as: introduction, scope of services (bulleted list), fee and payment terms, client responsibilities, timeline, and signature block. Keep the tone professional but warm. Avoid legalese where plain language works just as well.
That prompt produces a draft the attorney can review, edit lightly, and send. Without CRIT, the same request produces something generic that needs 20 minutes of rewriting.
The firms that get consistent results from AI are the ones that train their teams to write prompts the same way, every time. If you want to implement CRIT in your firm or need help adapting it to your workflows, reach out.
Aspen Management Group works with boutique advisory firms to clarify key workflows, improve efficiency, layer in AI where it adds value, and build governance and training around that change.
Scott spent 20 years running a managed IT services practice serving professional services firms across the DC Metro area, and has worked in technology for 35 years. AMG helps boutique professional services firms get practical value out of AI.