Why Your AI Output Is Inconsistent (And Nine Frameworks That Fix It)

A fast framework is the difference between the AI ​​output you can put in front of clients and the output you silently remove. Over the past year, almost every client engagement I’ve been involved in ended with the same question.

Whether it’s the launch of Microsoft Fabric, the Copilot adoption program, or an AI readiness assessment, someone is finally asking:

“How can we get consistent output from this model?”

That’s the right question. Most of the teams I work with have implemented Copilot or ChatGPT. The equipment is not the problem. But the results were inconsistent — sometimes brilliant, sometimes unusable. Same model, same data, very different results.

Nine times out of ten, that’s not the model. This is the prompt structure.

And the fix isn’t a “magic phrase” or clever trick from a LinkedIn carousel. It’s a small collection of repeatable frameworks that anyone on the team can implement, so that “good command” stops being a matter of personal style and starts to become a matter of process.

Three mistakes I see in 90% of enterprise requests

Before we discuss the framework, it is worth giving the pattern a name. When I audited commands within enterprise teams, the same three mistakes came up over and over again:

1. No defined roles. Without roles, the default model is a generic assistant voice. You get output that reads like it was written by a helpful scholar, regardless of whether you need a CFO or security architect perspective.

2. No output format specified. You ask for “analysis” and you get four paragraphs of prose when what you really want is a three-column table, bullet list, or structured memo.

3. There are no obstacles. There are no commands whatsoever telling the model what “good” looks like — no word limits, no note markers, no examples of the desired output. So the model is hallucinating sensibly and you spend 20 minutes fixing it.

You can patch any of these and see the improvements. Patch all three with a repeatable structure and you change the quality of output across teams.

Advice vs. fast engineering

There is a difference between encouragement — writing ad-hoc instructions and hoping for the best — and fast engineeringwhich is a discipline.

Rapid engineering as a discipline looks like this within a client organization:

• A framework libraryso there is no rediscovering control on each task.
• Fast versioningso you can A/B test and audit what is sent to the model.
• Evaluate outputso “good” can be measured, not based on vibrations.
• Reviewgateso that AI-assisted output reaches the same standards as human output before being sent to the client or board.

Most “quick engineering” content on LinkedIn is tactics — magic phrases, clever hacks. The tactic is entertaining. The system is billable. If you’re teaching an organization of 500 people to use AI responsibly, you need a system.

The framework library is the basis of that system. That’s what this post is about.

The eight frameworks I use most often

These are the eight frameworks I achieved in direct engagement. Each solves a different class of problem – the skill is in knowing which to reach for.

1. RTF — Roles, Tasks, Format

Best for: Fast transactional output where you need fast consistency.

The simplest framework and the one that fixes the most problems. You determine who the model should act on, what it should do, and how the output should be shaped.

Example: “You are a CFO preparing a board update. Summarize Q3 performance based on the three KPIs below. Results: five bullets, 15 words max each, plain English, no jargon.”

RTF is my default starting point for any new prompt. Maybe 60% of everyday commands don’t require more than this.

2. CHAPTER — Before, After, Bridge

Best for: transformation and persuasion.

Describe the current state (Before), the desired state (After), and the bridge between the two. Very useful for internal change communications, sales collateral, and anything else you need to move readers from one position to another.

Example: You’re drafting communications for the launch of Copilot. Previous = “we waste two hours a week on status updates.” After = “status update took ten minutes.” Bridge = “here’s how Copilot does it.”

3. CARE — Context, Action, Outcome, Example

Best for: results and case studies.

Very useful when you need a model to replicate a particular shade or pattern of results. Include previous work as an example and the model will calibrate it. This is the framework I use most often for client case studies and testimonial drafting.

4. CRIT – Context, Role, Interview, Assignment

Best for: complex strategic brief.

The “interview” element is what sets CRIT apart – you ask the model to interrogate you before it is conceptualized. This brings up assumptions you haven’t stated, and this is where most complex summaries fall apart. Heavier to use than RTF, but worth it if the output must withstand scrutiny.

5. RISE — Roles, Input, Steps, Expectations

Best for: results and phased plans.

The “step” element makes RISE especially useful for multi-stage work — project plans, staged migrations, and training programs. This model produces outputs that you can incorporate directly into your delivery plan rather than having to restructure it.

6. CO-STAR — Context, Purpose, Style, Tone, Audience, Response

Best for: voice-only content.

The toughest framework on the list and the one to reach for when voice and tone are more important than content. Style, tone, and audience are intentionally separated — you often want a formal style, a direct tone, and a non-technical audience, and most frameworks combine them into one instruction.

Use CO-STAR for executive communications, papers, and anything that needs to be read in a specific person’s voice.

7. RODES – Roles, Goals, Details, Examples, Taste Check

Best for: content is critical to accuracy.

This “sense-check” element is what makes RODES different. You ask the model to flag anything that seems questionable before you receive the results. It is valuable for orchestrated content, technical documentation, and any output that causes hallucinations to have a real impact.

8. APE — Actions, Goals, Expectations

Best for: Very sleek and fast task.

Minimum viable structure. Use it when you need something quickly but still want more than a three-word command. APE is what I reach for when I’m working on something else and just need a quick hand — not when the results have to go somewhere important.

Before and after examples

Here’s what it looks like in practice.

Previous (without outline):

“Write something about our Q3 results on the whiteboard.”

After (RTF applied):

Role: You are the CFO preparing board updates. Task: Summarize Q3 performance against the three KPIs below. Format: Five items, maximum 15 words each, simple English, no jargon.

Same model. Same basic data. Very different output. The first version produces a general press release. The second produces something you can place in front of the board.

That’s the essence of a framework. It’s not about finding the perfect sentence. It’s about structuring the instructions so that the results are always good, not just on lucky outcomes.

Real client examples

A finance team I work with spends two hours every Monday putting together an executive summary for their weekly trading package. They’ve tried ChatGPT — it makes things worse. Generic copy, wrong notes, numbers the MD doesn’t trust.

We didn’t change the model. We changed the prompt structure using CARE:

• Context: trading package objectives, audience, three necessary parts.
• Action: prepare an executive summary based on the data provided.
• Results: 250 word summary, four parts, direct tone, no hedging.
• Example: summary of the previous week, which has been signed by the MD.

The result: fifteen minutes instead of two hours. The MD couldn’t tell which summaries were AI-assisted and which weren’t.

That’s the difference structure makes. And the scale increases — the same framework can be applied across teams, with the same results, by people who didn’t design it.

Which framework should you start with?

Eight frameworks are enough to handle most real-world scenarios, but they can also be quite crippling if you try to adopt them all at once. Practical sequence:

Start with RTF for any transaction output — board summaries, meeting notes, quick drafts. This is the 80/20 framework.

Add CARE when you need to match an existing sound or tone. The example component does most of the work here.

Reach out to CO-STAR or RODES when the stakes rise — voice-first executive communications (CO-STAR), or accuracy-first regulated/technical output (RODES).

Use BAB, CRIT, RISE, and APE situationally. CHAPTER for a narrative of change. CRIT for summaries that you don’t fully know the summaries of. RISE for multi-stage plans. APE when you just need a quick win.

Don’t try to adopt all eight things at once. Pick two, use them until they become muscle memory, then layer on the rest.

How to use this in your organization

If you’re responsible for rolling out AI within a real organization — whether it’s a Copilot adoption, a Fabric engagement, or an internal AI enablement program — there are three practical steps:

1. Choose two or three frameworks of the eight above and standardize them. RTF and CARE cover most of the daily work. Add CO-STAR or RODES for more sensitive cases.
2. Build a shared prompt library. Even a simple document sharing “here’s our RTF template for board updates” eliminates the per-person variations that break consistency.
3. Evaluate the results. Select two or three quality dimensions (accuracy, format compliance, tone compatibility) and assess the output based on those dimensions. If a command doesn’t consistently score well across ten runs, then it needs more structure — not a different model.

Try the framework

I’ve packaged all eight frameworks into two free resources.

The first is interactive tool. Choose your scenario, choose a framework, describe your task, and it will return a fully structured prompt that you can paste into Copilot or ChatGPT. Less than a minute, no need to register.

On the same page, there is also a PDF cheat sheet You can download — each framework, when to use them, worked examples, and failure modes to avoid. It’s designed to be displayed on a second monitor or shared to a team channel.

→ Access quick framework tools and PDFs

Both are taken directly from direct client work. They’ll land hardest if you’re in the position of “we’ve deployed Copilot, now what?” stage — but this model will be useful for anyone who wants to get more consistent output from the model they are already using.

---

Gethyn Ellis is a UK-based Microsoft data and AI consultant working with enterprise teams on Fabric, Copilot adoption and AI readiness. Contact us to discuss rapid system engagement for your organization.