Create Your Visual Brand Identity

Before AI, I didn’t update my website except, like, once every decade.

Really. I was too frugal to hire a designer, and too lazy to do the branding work. So I left it alone, for years at a time.

What that looks like after seventeen years

Here are my own course covers. Twenty-two of them.

• Roughly ten different visual treatments, and six or seven different typefaces.
• Each one made with care at the time. Together they look like ten different businesses.

That’s my “before” picture, and I’d guess it’s a familiar starting point for anyone who’s been at this a while.

Why now, and not in any of the last seventeen years

I’ve been avoiding this work my whole business life. What changed is the price of it. Brand work used to be expensive, or slow, or both — a designer, a discovery process, weeks of back and forth. Now it’s an afternoon with an AI, and you can run it again whenever you’ve changed.

AI put this course together for me, and then I ran it on myself. I’m doing every exercise here alongside you, for the first time.

It’s worth doing because coherence reduces friction for the people who already want to believe you. Every mismatched cover and off-key color spends a little of the trust you’d otherwise have. I’ve been paying that quietly since 2009.

Showing up as you are is a legitimate choice

Someone wrote in to my website’s AI assistant a while back, and said it better than I could:

“They all feel a little intimidating because I’m not Internet shiny... I don’t have the Internet aesthetic. I look like I’m low income... There’s a part of me that just wants to show up as is anyway, because I think more people need to see what regular people look like.”

She’s right, and I’m not going to talk anyone out of that. Showing up as you are is a real position, held on purpose by people I respect.

This course is for a different moment: when you’d like your look to stop working against what you’re offering. Someone else described that side well — “it needs to convey the mastery that is behind all this... so it doesn’t look cheap, if you know what I mean?”

The permission, same as the previous two courses

• Try on a look the way you try on clothing. Nothing here locks you in.
• Your visual identity is a snapshot, not a vow. You’ll revise it as you grow.
• Nothing goes public until you want it to. Most of what we make is for your own clarity first.

Where we’re headed

By the end of the second module you’ll have a small brand kit that’s genuinely yours:

1. A visual direction you can describe in a sentence
2. A color palette with actual hex codes
3. A font pairing
4. An imagery style
5. A wordmark
6. A one-page style guide that holds it all, plus three ready-to-use assets

This module builds your visual vocabulary — direction, palette. The next one makes the assets.

One practical thing before you start: keep a single AI chat open for the whole course. Every lesson hands the next one its material, so if you stay in the same session you never re-explain yourself — it already has your reactions, your brand kin, your not-me list. One long conversation, six lessons.

Share in the Student Discussion doc: where are you with visual branding right now — avoiding it, halfway through a redesign, somewhere in between? 🙏🏼

Your aesthetic is already implied by what your clients feel in your presence and by what you stand for. We’re translating something that exists, not inventing from nothing.

Bring your Evidence file, if you have one

If you built one in Discover Your Brand Strengths, open it now and skim for the sensory words your clients used about you — calm, warm, grounded, spacious, no-nonsense.

Those are already visual instructions, and they’ll sharpen everything in this lesson and the next four.

And if you don’t have it, nothing here depends on it — the exercise below works from scratch.

The thing that stops most of us

Almost nobody gets stuck because they lack taste. They get stuck because there are too many possible answers.

Someone asked my AI assistant this, and it’s the clearest description of the block I’ve read:

“I am struggling with my brand colors, how did you find yours? There are so many possibilities that I am confused. Should the colors be what I like, or what fits with the targeted audience... I just had the thought of having one neutral...”

And from someone else, in five words — “I am afraid to make the wrong decisions.”

That’s a decision problem rather than a talent problem, which is genuinely good news. Decision problems have methods, and we’re about to use one.

You’re already good at this, in one direction

You know within about two seconds when a website feels off. Cluttered, dated, trying too hard, not quite trustworthy.

• That’s taste, running in recognition mode instead of production mode.
• Almost nobody can generate a visual direction from a blank page. Almost everybody can react to one.
• Reacting needs something in front of you to react to — which used to mean hiring someone, and now takes an AI minutes to deliver a draft to you.

👀 The exercise: react to three real pages

You don’t need design vocabulary for this. You need opinions, and you already have those.

Open three things:

1. Something you made. Your website if you have one — otherwise your most recent flyer, slide deck, social post, book cover, or email header.
2. A nichemate’s website. Someone doing work close to yours.
3. One more of either kind, picked because looking at it gives you an opinion.

For each one: what here feels like me, and what doesn’t? What do I like about how it looks, and what bothers me?

Plain words are the right words. “Too busy.” “Feels cold.” “That photo doesn’t look like anyone I’d call.” “I like how much empty space there is.” “The green is nice, the gold is trying too hard.”

Judging your own work is hard; judging someone else’s is easy — and the reactions you have to theirs are still yours.

If you use something you made with AI, reacting against it is the point. Getting in the habit of pushing back on what an AI hands you is how your own discernment stays sharp — how your energy signature stays yours, instead of drifting toward whatever the model happens to like.

🎥 Talk it, don’t type it

The fastest way through this, and the one I’d recommend: record a two- or three-minute screencast of yourself scrolling those pages and saying what you think, moving your cursor to whatever you’re talking about.

Mac: Shift-Cmd-5. Windows: Win-Alt-R. On a phone, your screen-record button — and pointing with your finger works as well as a cursor.

A third way, if neither of those is familiar: start a Zoom meeting with yourself, share your browser screen, and hit record. Most of us already know how to do that one.

Nobody sees this recording. It exists so your AI can read your taste back to you — it is never published, and there is nothing to make it presentable for.

Then hand the recording to your AI:

You’ll say more out loud in two minutes than you’d type in twenty.

On video: Gemini takes video files directly — up to five minutes on the free plan, an hour on a paid one, so two or three minutes is safe. Other tools vary and keep changing.

If yours won’t take the file, screenshot each page instead, attach the images, and type your reactions into the chat. Same three requests, same result.

What you keep: two short lists — what feels like me, and what doesn’t. That’s the first draft of your visual direction.

Go ahead and do this now, before the next lesson 🙏🏼

Did you run the brand-kin prompt from the previous course’s lesson on the one trait worth borrowing?

• If you did — attach what came back to the prompt below, and tell your AI to skip the questions and synthesize straight from those answers.
• If you didn’t — that’s the common case, and we’ll get you there in about three minutes right now. This is the main road through the lesson.

The three-minute self-interview

Brands you love already carry a full visual vocabulary. Borrowing the feeling of one isn’t copying — it’s the fastest way to give an AI something it can render.

“Brands” here means any kind of product or place you know well. Asked in the abstract, most of us go blank — so pick from what you actually use and enjoy:

• something you buy regularly: food, drink, household, skincare
• a shop, café, restaurant or hotel you like walking into
• a tool or app you open every day
• a car, a bicycle, a kitchen thing, a piece of furniture
• a magazine, newspaper, publisher or podcast
• a clothing label, if clothes are something you pay attention to
• a musician or a band whose albums have a look
• a public figure or influencer whose whole presentation hangs together

And if no brand comes to mind at all, use places instead. A particular café, a library you love, your grandmother’s kitchen — they each have a look, and you already know how they feel.

Paste this into your AI:

And if you have your Brand Compass or brand brief, attach it. The AI will interview you far better with your own words on the screen.

One extra line, and it’s worth it

Before it writes the two or three directions, add one instruction:

An AI’s first few ideas sit close together, and they’re the same first few ideas everyone else gets. A wide, shallow list costs nothing — and your two-line reaction to it (what you pictured, what felt tacky, what’s missing) is a better brief than the interview was.

Then:

You’re not repeating the previous course

That prompt unpacked each brand individually, one trait at a time. This one synthesizes them into a visual direction — a different output, built from the same material. If you did the earlier work, you’re spending it here rather than doing it again.

Two things that go wrong here

• You get six options instead of two or three. Reply:

  Cut this to your best two, fully described.

• The directions all sound the same. Usually means it has too little to work with. Give it your two lists from the previous lesson — what feels like you and what doesn’t — plus one sentence about who your clients are.

Mine, as an example

My four, one trait each:

• Mister Rogers — calm that regulates, rather than calm as a manner
• Bob Ross — hands over the exact technique, and turns the mistakes into part of it
• Khan Academy — ungated, and it expects you to leave
• Muji — utility with nothing wasted

It was five until recently. I cut Julia Child once I noticed Bob Ross already covered that ground and covered it better, which is what a kin list is supposed to do over time — get shorter and sharper.

One caveat worth saying out loud. Muji is internal scaffolding for me, not public copy. I chose it over IKEA knowing that most Americans wouldn’t recognize the name. Fine for briefing an AI, wrong for a sales page — borrowed clarity only works publicly with brands people already know.

So you get two kinds of kin: the ones you say out loud, and the ones that are just for you and your AI.

You’ve just seen directions you might like. Next we find your dealbreakers — by looking at one you won’t.

Comment in the Student Discussion doc with the name of one visual direction your AI gave you, and whether it felt close. 🙏🏼

Two lessons, two opposite jobs. This one finds your no. The next one finds your yes.

And if you couldn’t name your taste directly if I asked you to — good. This lesson doesn’t need you to. It’s the one that breaks the decision paralysis we named in 1.1: too many possible answers, and no way to choose between them.

Starting with the no isn’t a detour. What repels you narrows the field far faster than what attracts you, and it’s much easier to feel — the no arrives in about a second, while the yes usually takes a few tries.

Why we generate images here, and don’t just describe them

In lesson 1.1 I said that almost nobody can generate a visual direction from a blank page, and almost everybody can react to one.

That applies to this lesson too. If an AI hands you a paragraph describing a visual direction, you’re still the one doing the imagining — which is the exact thing 1.1 says you can’t easily do. So we make the pictures, and you react to them.

I learned this the hard way while building this lesson. The first version of it was text only. It didn’t work, and it didn’t work for a reason my own earlier lesson had already explained.

The move

You’re going to ask an AI to pick a visual direction for you — confidently, with no options and no hedging — and to render it as an actual image. Then you do it a second time, with a deliberately different direction.

• Some of it you’ll like. Some of it will produce a small, immediate “no, that’s not me.”
• That “no” is what we’re here for. Dissonance is data.
• Two generations, and then stop. Don’t start improving either one — refining is the next lesson’s whole job, and doing it here is how you run out of afternoon.

If you’ve done Your Energy Signature, you’ve felt this before. Same contrast principle, pointed at the visual layer this time.

React the moment it appears. Give it thirty seconds and you start getting used to it — the image stops being new and the no goes quiet. Whatever comes up in the first few seconds is the honest answer.

🎨 The prompt

This lesson needs an AI that can make pictures — ChatGPT, Gemini, or whichever image tool you already use. If your usual assistant only writes text, open a second tab for this one; you’ll want it again in the next lesson.

Fill in what you have. Blank lines are fine — it works with less than you’d think. And if you have your Brand Compass or brand brief, attach it instead of filling in the fields...

Then, for round two:

Four notes on the prompt

• Ask for the paragraph as well as the picture. The image is what your body reacts to. The paragraph is what makes the reaction useful, because it argues the case from your own words — and a picture can’t tell you it chose warmth because three people who never bought anything from you described you as calm.
• About the last line — do not hedge, do not offer alternatives — I tested this fairly carefully across several AI models. They commit to one direction almost every time, even with that sentence deleted. So it’s insurance, and it costs you one line. Keep it.
• What actually breaks this exercise is your own wording. Ask for ONE direction. If you ask for “a few directions so I can compare them,” you’ll get a menu — and a menu is the previous lesson again, not this one. If it hedges anyway, reply: “Commit to one.” That rescues it reliably.
• Make the two directions genuinely far apart. For round two, add the line below. The test is simple: if you can’t say in one phrase what separates the two pictures, they’re neighbors rather than opposites, and you’ve only found one no. Ask for a wider second.

  as different from the first as you can make it — even one that sounds like it couldn’t work for me.

What you keep

Your “not me” list — three to five short phrases naming what you now know isn’t you.

Not “I don’t like brown.” More like: “warm but not sleepy,” “no hand-lettered script,” “nothing that looks like a wellness stock photo.”

Name the part, not the whole picture. An entire image is almost never wrong — usually one or two things in it are. “The colors are right, the clutter isn’t” is a useful note. “Not this one” throws away the parts that were working.

If you go blank, run down the list: shapes, colors, structure, the light, and the juxtapositions — which two things are sitting next to each other and shouldn’t be. A few words per item is plenty. AI reads a short note better than you’d expect.

This list is as useful as your yes-list, and it goes into every visual brief you write from here on.

And one more thing to carry forward: whichever of the two came closer — or whichever parts of either came closer — write that down in a sentence or two of your own. That’s your starting direction for the next lesson.

If both of them repelled you, that’s a normal result of an exercise built to find the no. Your sentence is then simply the opposite of your not-me list, and the next lesson works fine from there.

Mine, since I’ve already run this

During the previous course I looked at Wikipedia as a possible brand kin and winced twice — “quite left-leaning” and “loud fundraising.” I almost skipped past both. That wince produced the sharpest line in my whole positioning: the ask stays quieter than the gift. I didn’t reason my way there. I reacted badly to a fundraising banner, and then asked myself why.

Then I ran this exact exercise on myself while building the lesson, and it did something I didn’t expect.

• Round one came back as a warm, near-black room with everything resting on one floating panel of cream paper. I liked it.
• Round two was the opposite — bright near-white, hard-edged blocks of cobalt blue. Most of it wasn’t me.
• But I couldn’t let go of one small thing in the picture I’d rejected: a piece of clear glass. It felt fresh, and refreshing turns out to be one of the words my own clients reach for most consistently — ten different people, on ten separate occasions.

So the no was easy, and the surprise was a yes hiding inside the no. That is exactly why you generate two and not one.

Comment in the Student Discussion doc with one line from your “not me” list — and if something you liked turned up inside the direction you rejected, say that too. They’re often the most fun thing anyone shares here. 🙏🏼

Now we converge.

In the last lesson you generated twice and stopped, on purpose. Here you generate once and stay with it, changing one thing at a time until it’s yours.

Start from the sentence you wrote at the end of the last lesson — whichever direction, or whichever parts of one, came closer. It doesn’t need to be good. It needs to exist, so there’s something to change.

Create a “moodboard”

You may have heard the word before — it means a collage that shows the feel of a direction before anything gets designed. That’s what we’re making, and you’ll have one by the end of this lesson.

Until recently this step cost real money. A designer charges in the low thousands to bring you a moodboard, and if it misses you pay again for the next one. Here it costs a prompt, so you can iterate until it’s right — and keep iterating later, as you change.

The only difference is its shape. An image AI hands you one composed picture instead of a grid of separate clippings you can rearrange. Same job, one frame — and for our purposes that’s better, because you can regenerate the whole thing in seconds.

It’s a fitting, not the garment. Tailors sew a cheap draft in muslin first, on purpose, so the expensive fabric only gets cut once. This image is your muslin.

And if an arrangement of objects feels too abstract to react to, ask for a real room instead — “show me a room that feels like this direction.” Some people find a scene much easier to have an opinion about. Just remember you’re reading it for its colors, materials and light, not deciding whether you’d want to live there.

What this is actually for

Fair thing to ask before you spend an afternoon on it. A moodboard has four jobs, and none of them is “be a nice picture.”

• It’s a brief you can hand over — to an AI, to a designer, to yourself in six months. “Make it calm” gets you anything. The picture gets you the thing.
• It’s where your colors come from. That’s the next lesson. You don’t pick a palette off a color wheel; you pull it out of this image.
• It’s your tie-breaker. Stuck between two versions of a graphic? Don’t argue about it in the abstract — hold both against this, and one of them is suddenly obviously wrong.
• It’s your consistency check. A year from now something you’ve made will feel off and you won’t be able to say why. This is what you hold it up against.

And one thing it isn’t: a picture of your office, your life, or anything you’re going to build. Nobody but you ever sees it.

If you catch yourself judging whether you’d want to sit in that room, that’s the commonest way this goes sideways — you’ve slipped from reading it as a reference into reading it as a photograph. I did exactly that while building this lesson.

🖼️ The prompt

Use whichever image AI you already have open — ChatGPT, Gemini, whatever you’re comfortable with. I’ll be demonstrating in ChatGPT so we’re all looking at the same buttons.

If one of the three pages you looked at in lesson 1.1 had a look you liked, attach that screenshot as a reference too. And if you have a brand brief from the first course, attach that as well.

Why the prompt says both: I ran it with and without the “one composed image” clause and got a single composed frame either way. The clause stays in as a guard against a labelled grid.

Your first one will probably be wrong

That’s the process working, and it happens to everyone including me.

Change one thing at a time. “Same image, but half as much color.” “Same image, colder light.” If you change five things at once and the next image is better, you’ve learned nothing you can repeat, because you don’t know which of the five did the work. One change at a time is how you find out what your own preferences actually are.

Keep the rejects, and name what was wrong with each one. A whole image is almost never wrong. Usually one or two things in it are.

So don’t tell your AI “not this” about the picture. Tell it which part:

• “The palette is right. The clutter isn’t.”
• “The mood is close, but the photography looks too corporate.”
• “Keep the empty space, lose the gold.”

If you hand back whole images as rejections, the AI drops the parts you liked along with the parts you didn’t.

And carry across whatever you liked inside a direction you rejected. If one element from the picture you said no to keeps coming back to you, put it in this prompt by name. Mine was a piece of clear glass, and it changed the whole direction.

A second pair of eyes that knows nothing

When you have one you almost like, show it to a chat that has seen none of this — a fresh conversation, the image alone, none of your prompts or your reasons. Ask:

Don’t tell it what you wanted or how many tries it took. A critic that knows the story goes easy on it. Take the one gap, change that one thing, and stop when the answers stop improving.

The last move is subtraction, and it’s yours rather than the AI’s. An AI adds; it almost never removes. Ask of each element, what breaks if this is gone? — and take out whatever nothing misses. Restraint is most of what reads as expensive.

If everything comes back soft and warm

Check whether that’s all of you.

Warm is easy for an AI to render, so it tends to over-serve warmth. If your work is also rigorous, or unsentimental, or a bit fierce, and none of that made it into the picture, say so directly:

I have this problem myself, and I’ll show you where I found mine. When I polled my own audience on how my work feels, the gentle option won the vote — and then several of the people who’d invested most in me described me in words the gentle option couldn’t hold: energizing, no-nonsense, fiercely passionate. One of them rejected my whole set of options as “a little bland, which doesn’t represent you at all.” If I’d read only the tally, I’d have built my look on the half the tally could measure.

What you keepOne image you don’t hate, and a sentence describing the direction in your own words. That’s what the color lesson works from next, and what you’ll bring into Module 2.

It doesn’t have to be right. It has to be close enough to react to — which, by now, you know how to do.

Comment in the Student Discussion doc with your direction image once you have one you don’t hate. Seeing everyone’s side by side is one of the better parts of this course. 🙏🏼

Don’t start from colors you like. Start from the feeling you want a client to have in your presence, and let the colors be chosen to produce that feeling.

That one reordering is most of what separates a palette that holds together from a palette that’s your favorites in a row.

You may already have your feeling words

If you did the first course, the non-negotiables in your Strengths doc are exactly this list. Mine are:

calm over hype · lightness over intimidation · dignity over condescension · presence over performative energy · simplicity over unnecessary elaboration · sincerity over persuasion.

Six words is plenty. If you don’t have that doc, write three to five now — how you want someone to feel after an hour with you.

🎨 The palette prompt

Six or fewer. A palette with eleven colors is a decision you haven’t made yet.

AI has favorite colors — check yours against your niche

Image and text models both drift toward a handful of default palettes. You’ll start noticing it once you look: the same sage, the same terracotta, the same dusty blue, across businesses with nothing in common.

So hold your palette up against two or three of your nichemates’. If it matches people whose work is nothing like yours, that’s the default talking, and it’s worth adjusting.

Matching your own field is a different matter, and often correct. Visitors read your category off your colors in about a second, and some industries have a convention that’s doing real work for you. Different from everyone isn’t the goal. Different from the model’s defaults is.

✅ Then check it can be read

Paste your text color and your background color into the WebAIM Contrast Checker — free, one page, no account.

• 4.5:1 or better for body text.
• 3:1 or better for large headings — and “large” has a real size behind it: about 24 pixels, or about 19 if it’s bold. Anything smaller needs the 4.5.

Don’t ask your AI for the ratios. It’ll give you numbers that look right and sometimes aren’t. The checker takes ten seconds and isn’t guessing.

One exception: if your AI can browse the web, hand it your hex codes and have it run the checker itself and report back. That’s the checker doing the arithmetic, which is the whole point — it’s only guessing when it answers from memory.

My own palette fails this, and I’ll show you live

When I recorded this, my live site ran on five colors — cream #f6f3ee, sage #6faf9c, slate #3d474d, navy #07273F, coral #f19766.

Run through the checker:

• slate on cream — 8.60:1, fine
• navy on cream — 13.84:1, fine
• sage on cream — 2.29:1. Fails.
• coral on cream — 2.03:1. Fails badly.

So two of my five colors are unreadable as text on my own background color. I’ve been running this business since 2009.

The fix isn’t to throw them out. Sage on navy is 6.05:1, which passes comfortably. So those two colors work fine as fills, borders and small accents, and fine as text on the dark ground. They’re only wrong as body text on cream.

The verdict is never on a color. It’s on a color in a role.

One reason this is worth ten seconds

I don’t treat this as an accessibility checkbox. Some of the people reading you have low vision, or are on a phone in bright sun, or are simply tired at the end of a day. A beautiful pastel palette that nobody can read serves nobody, including you.

Two quick sanity checks

1. Put your palette next to a real photograph of you. Does it fight your skin tone, your usual clothes, your rooms?
2. Put it next to a paragraph of your own writing. Does the palette read as the same person who wrote those words?

If either one is off, adjust now — much cheaper today than after a whole website is sitting on top of the palette.

Comment in the Student Discussion doc with your hex codes, and which pair failed the contrast check. Somebody’s will surprise you. 🙏🏼

🍷 The vertical tasting

Wine people taste several vintages of the same wine side by side, because differences you’d never notice alone become obvious in a row.

Then just look. Notice which one resonates more with you.

Sometimes your own palette wins and you stop second-guessing yourself. Sometimes a neighbor wins, and you’ve saved yourself a year of low-level dissatisfaction. Bring what you notice to the next module.

🗳️ Optional: ask your audience

A bonus, and the one that tends to surprise people.

Post your top two direction images and ask:

Four steps, and steps 2 and 3 are the ones people skip:

1. Post it where your people already are. Two images, a forced choice, and always ask for the why — the reasons are the data, the votes are just fun.
2. Two reasons beyond the answers: it’s a genuinely fun thing to do with your people, and it leaves a visual impression on them — they see your direction before your website does.
3. Then send the link privately to three to five people who represent your ideal client. Ask kindly, give a gentle due date, and say there’s no pressure at all if now isn’t a good time. A handful of direct asks beats one public post, at any audience size.
4. Share your poll link in the Student Discussion doc, and answer a couple of other students’. We’re often each other’s ideal clients here.
5. Report back to whoever answered, telling them what you learned. That turns an ask into a relationship.

Let your AI go first at every step — have it propose the two images to use, draft the post, and draft the private note in your own voice. Then you edit.

If this is new to you, Authentic Market Discovery teaches the whole method properly. No need to re-derive it here.

And if you do neither

Carry on anyway. Nothing in the next module depends on having done the tasting or the poll — we’ll be making type, imagery, a wordmark and your one-page style guide, and a direction plus a palette is all you need to walk in with.

Nobody is behind. Polls trickle in for weeks, and your kit updates easily.

And there’s a real reason to skip it. Some of us already know what we want and would rather not be talked out of it by a vote. That’s a legitimate call — a poll this early can dilute a direction you’re genuinely sure of.

You can also just poll later. Once you have two finished web-page looks to choose between, the question is sharper and the answers are worth more, because by then you’ve settled the things you don’t want reopened.

Comment in the Student Discussion doc with what your vertical tasting showed you — or with your poll link, so another student can answer. 🙏🏼