Don't Let AI Misunderstand You
The simple technique that gets results to better match your intent.
If AI still feels unpredictable, you might be missing the one ingredient that will make its answers sound better.
My previous post on roles showed how including a role in your prompt can get you better AI responses. But a role only gets you halfway there — because it still leaves the model to decide how that role should act. That’s what this post is about: helping you close the gap between do the task and understand my intent.
With this approach, you’ll get the results you want faster, with fewer iterations, and more aligned to your specific needs.
🎧 Prefer to listen? Here is the 9-minute audio version, also on Spotify, and Apple.
Roles Contain Multitudes
Here’s the secret: you want to create a character, not just name a role.
There’s a scene in Good Will Hunting where Robin Williams’ character shares his perspective on love, learning, and life. Here is a link to the full monologue — one of my favorites in any movie. It’s emotional, cinematic, and full of amazing dichotomies. And it perfectly illustrates the difference between a role — Williams playing a psychologist — and a character — Williams playing a lonely psychologist, stuck in the past, and still grieving over his beloved wife who he lost to cancer years ago.
It’s the character, not the role, that determines the performance.
The script’s words may have been the same no matter which actor played the part, but Williams’ delivery and character choices change everything. He could have played that monologue with frustration or authority, but he chose empathy, gentleness, and love. And the entire meaning shifted.
A role tells the model who to be.
A character tells the model how to behave.
And once you see that distinction, the next leap becomes obvious. Roles shape identity. Character shapes interpretation. Try this: Imagine a gardener. Don’t think too long — what’s the first thing that came to mind?
Your grandmother spraying blue Miracle Gro on her summer tomatoes?
Alain Baraton1, the head gardener at Versailles since 1982?
A biotech researcher studying plant pathogens?
They’re all gardeners. But if you asked each of them to, “describe the perfect garden,” you’d get three entirely different — and entirely valid — answers.
If a gardener contains this much variation, imagine the variance inside roles like CISO or CEO — where interpretation isn’t aesthetic, it’s operational.
And this is the heart of the problem: a single role can hold wildly different lived experiences, instincts, and values. Walt Whitman captured that truth in 1855, long before large language models existed:
Do I contradict myself?
Very well then I contradict myself,
(I am large, I contain multitudes.)
— Leaves of Grass, Project Gutenberg Release #1322
Roles contain multitudes. And when you’re leading teams or making decisions, “multitudes” is another word for “variance”, and variance slows everything down. Your job is to reduce variance, not manufacture it.
Why Defining the Right Character Matters
If you’re leading teams, programs, or product decisions, the “average” version of a role is never the one you want. But before you can nudge the model beyond the average, you need to think about which direction you want. Professional? Creative? Warm? Urgent? Critical? Here are a few examples in some professional roles:
A program manager might sound operational and disciplined, or diplomatic and coalition-building, or ruthlessly focused on delivering outcomes that matter.
A product manager could be visionary and expansive, or analytical and exacting, or a cool-eyed realist who surfaces the risks no one else wants to say out loud.
A compliance officer could offer feedback as calm reassurance, or as firm guardrails, or as a direct, sleeves-rolled partnership that keeps you out of trouble.
A CISO might lead with policy and governance, or with threat modeling and live-fire scenarios, or with the clarity of what keeps the company off the front page.
A regulatory advisor could be cautious, or pragmatic, or forward-leaning in a way that balances innovation and scrutiny without decreasing quality or risk posture.
A CEO might be a bold storyteller shaping hearts and minds, or a steady operator, or a strategist who cuts straight to risk, ROI, and leverage.
Once you’ve decided what kind of behavior you need, you can tell the model how to interpret your specific role. And you can get much better results for your specific task.
Character Notes: How You Direct the Role
That’s where character notes come in: they narrow the interpretation so the model behaves like the specific version of the role you intended. They fill in the human texture that roles alone can’t supply — the judgment, taste, and emotional intelligence that make the response feel crafted instead of computed. Here are a few examples:
Character notes help you steer the response towards your intent.
Without character notes:
You are a creative director.
Review this product launch email.With character notes:
You are the creative director for the most successful marketing company in the country, and a former New York Times copy editor.
You’re expressive and artistic, known for award-winning campaigns, but you’re also a perfectionist who notices rhythm, tone, and nuance. You communicate with the precision and tone that senior stakeholders admire.
Review this product launch email.Same role. Same task. Completely different result.
How to Add Character Notes that Actually Work
You don’t need much. A couple of sentences is usually enough. As long as what you’re providing is directionally aligned to your intent, it should be helpful. There are essentially three types of things you can include. Think of these as character cues:
Background – What is this character’s history? Where were they before?
Temperament – How do they approach the world? How should they engage?
Values – What does “good” look like? What other qualities are important?
Example for product work:
You are a senior PM who started your career in operations. You’re direct, structured, and allergic to fluff. You care about clarity, feasibility, and real-world constraints.
Rewrite this roadmap summary for executive review.Example for strategy work:
You are a CEO who built and sold two companies. You value fact-based decisions, crisp thinking, and an understanding of tradeoffs. You're framing a decision memo for the board.
Review this memo for narrative, risk, and executive usefulness.The Bottom Line
Character notes affect the tone, perspective, and emotional calibration of the result. They make AI feel directed instead of instructed, and shift the model from completing a task to inhabiting a point of view. And that’s where the quality jump happens.
This is especially useful if your work requires clarity, speed, or alignment across teams, which is most modern leadership work.
Once you start adding character notes, you’ll notice it immediately. The answers will sound sharper, more confident, and more aligned with your intent.
Before you close this tab, try one line. Use whatever you’re working on right now:
You are a [role] with [character traits]. Here’s what we’re working on …One sentence. Ten seconds. A noticeably better response — multitudes better. And once you feel the difference, you won’t go back.
This is the third post in my AI Superpowers series, where you’ll find some techniques that make AI feel a little more human, and a lot more useful.
Next in the series: Using roles to build a creative brainstorming buddy.
Rob Allegar
I’m a lifelong builder and advisor exploring what happens when technology stops behaving like a tool and starts acting like a collaborator. In this newsletter, I explore the space between ideas and execution, and help people build things that matter. roballegar.com
Please share this with someone who contradicts themselves and contains multitudes.
Here is an interview (English language available) with Alain Baraton. His job is essentially to keep the Versailles gardens looking the same as they did in ~1700 or so when Louis XIV first moved the seat of the French government there.