A walkthrough for Mekdi
You asked me to walk you through everything since I took over: the website experiments that made the cut and the ones that didn't, the Facebook ads, and the prompts I've been feeding the AI to make all of it. That's what this is.
One
I wrote everything down on July 10 before touching anything, because I wanted a record of what was true on day one. Here is what I found.
Three things surprised me.
The first: the old Facebook ad account was not empty. Back in 2024 somebody had run waitlist campaigns for this exact planner and spent about $4,736, which collected 1,159 email addresses from people who raised their hand for it. Nobody has ever emailed that list. Those people are almost certainly cheaper to reach than strangers, and they are still sitting there.
The second: the website had a slow leak that nothing else could fix. People were arriving, looking, and leaving. Not because of the price. Most of them never scrolled far enough to see it.
The third: the store's legal terms named the wrong company. The Terms of Service and the contact page said MY SUCCESS LLC, which is not the business. I corrected that on the live site on July 29 to One Percent Planner LLC, and renamed the ad account to match.
I didn't start by making things. I started by writing down what was true. Inventory, price, reviews, what the site said, what the ad account had already done. That way nothing I built later would rest on a guess.
Every claim in an ad or on the page has to be something I can point at. $36 is real. "16 reviews, all five stars" is real, and I re-check it before I reuse it. Hardcover, undated, lays flat, four tabs, ships in a box. All real, all checked against the actual product.
Things I will not say: risk-free (the buyer pays return shipping), money-back guarantee (same reason), scientifically proven, thousands of happy customers, or any number I can't source. It costs a little persuasive power. It's worth it. A false claim on a live storefront is a legal problem, not a stylistic one.
Two
A prompt is just the instruction I type to an AI. If I want a photograph, I describe the photograph. If I want a piece of writing, I describe the writing. The AI reads it and produces something.
The catch, and it's the whole game: the AI has never seen your planner. It knows what "planner" means in general, spiral notebook and tabs and pages, so unless I stop it, it invents one. It will give you a black spiral instead of white. It will make the tabs grey and rectangular instead of black with one rounded corner. It will print gibberish onto your pages that looks like words until you zoom in.
So every prompt I write has two halves.
The mood. The light. The room. Who's in the picture, what they're wearing, where the camera sits. This part I write like a photographer's brief, and I let the AI interpret it.
The product. I paste the same block of physical facts into every single prompt. Word for word, this is it:
And immediately after it, the list of things that get an image thrown out:
I let the AI be creative about the light and the room. I never let it be creative about your planner. Those two sentences are most of what I've learned in a month.
Three
We had a handful of real product photos and no budget for a shoot. So I built a library. Today it holds 36 approved images, plus 28 square crops sized for the store and 16 real photos of the actual product that I keep as the ground truth everything else is measured against.
Here are real prompts, straight out of the scripts.
This one is a lifestyle photo, a woman using the planner in a home office. Notice how much of it is about where the camera stands and what her hands are doing, and how little is left to chance:
Attached to that, invisibly, are three more paragraphs I reuse on every lifestyle shot. One of them is oddly specific and it exists because the AI kept getting it backwards:
The AI kept producing beautiful photos where the planner was upside down to the woman holding it, or where the morning page was on the wrong side. Nobody notices consciously. Everybody feels that something is off. So I wrote the rule down and now it ships with every prompt.
An image is not finished when it looks good. It's finished when it survives an inspection. I crop into three places at full resolution and look:
| Where I look | What it has to be |
|---|---|
| The binding | White paired twin-loop wire, both strands through one square hole |
| The tabs | Black, correct four words, rounded on the outer corner only |
| The cover | Clean "The 1% Planner" lockup, no melted letters |
| Any printed text | Either correct, or blurred past reading. Never legible-and-wrong. |
That last line is the one that fails the most images. AI writes text that looks like text from a distance and turns into nonsense up close.
A single batch on August 8. Five images, two attempts each. This is a fair picture of the hit rate:
Four out of six, and only one of those four passed on the first try. That ratio is normal. It's why there's a folder of rejects for every folder of finals.
The single most useful trick I found: when the AI mangles your logo, hand it a tight close-up crop of just the logo as the first reference image, before anything else. It fixed the lockup on both AI tools we use. It took an embarrassing number of attempts to work that out.
Four
This is the part you asked about most, so I'll go in order.
Microsoft's session recording tool showed people tapping the main photo to see the planner more closely, and the page would go, in your words, out of whack. They'd zoom, the layout would break, and you could watch them dragging the screen around trying to get back.
Three problems stacked on top of each other. The photo had a "grow on hover" effect, which on a phone sticks. One tap and the photo is jammed at 1.3× with no way out. So people pinch-zoomed the whole page instead, which detaches the floating Add to Cart bar on iPhones. And there was no swipe at all; you had to tap tiny arrows.
I studied how Steven Bartlett's diary does it and copied the shape of it: swipe to browse, a magnifier button that opens a proper full-screen zoom you can pinch and pan inside, and the page itself no longer scales at all.
The old return policy said items had to come back unused and in original condition, with the buyer paying return shipping. That's a fine policy and it's the exact opposite of what somebody needs to hear when their worry is "I'll stop using it after two weeks."
So: 30 days, keep the planner, refund from the email alone. No return, no box, no shipping label. I called it the Drawer-Proof Guarantee because the fear it answers is the drawer.
I published the policy first, then turned on the badge that advertises it. A guarantee the written policy contradicts isn't a guarantee, it's a liability. The code was built so the badge literally can't appear until the policy is live.
Planner $36 + digital edition $16 + the free hardcover $29.99 = $81.99, and you pay $36. Every one of those numbers is a real published price. I don't invent "values." The AI mockups I was working from wanted to call things a "$14.99 value" and a "$9.99 value" for products that don't exist, and I threw all of that out.
One consequence worth knowing, because it's the kind of thing that bites six months later: the digital planner has to stay listed on the site. If it comes down, it stops being separately purchasable, and the moment that happens the $16 and the $81.99 stop being true. The page would still look perfect while quietly making a false claim. I unpublished it for about half an hour on August 12 and put it straight back.
The old page led with the offer: the price, the bundle, buy now. The ads were doing their job. Cheap clicks, lots of people arriving, and almost nobody buying. 973 people landed on the page, 2 added to cart, 0 checked out.
My read: they weren't hesitating over $36. They were hesitating over themselves. They've bought a planner before. It's in a drawer. The question in their head isn't "is this good?", it's "why will I actually use this one?"
So I built a second version of the page that answers that question before it mentions the price. Same product, same $36, completely different order of argument. It went live August 9 as a parallel page while the old one kept running, so we could compare rather than guess.
Two days later I promoted it to the entire store. Homepage, product page, everything. There's no old version left.
Five questions, under a minute, no email required. Pick an area: health, wealth, relationships. Pick a goal. How much time do you honestly have most days? What's stopped you before? When would it happen?
Then it thinks for half a second and hands you one specific action:
Then: "That's your first 1%. The planner is 30 more of these."
There are 36 of those lines and I wrote every one by hand, because this is the one place on the site where an AI-generated sentence would be dangerous. The wealth answers never give financial advice. They're all "write this down," "make a list." The health answers are gentle and never medical.
If you come back later, it remembers, and opens with "Your 1% from last time — did it happen?" Yes gets you the next one. Not yet gets you "Today counts."
On August 11 I sat and watched the session recordings. It was uncomfortable.
The single most-clicked thing on the page was the photo carousel. People were flicking through pictures, 47 clicks then 42 then 28 then 22 on the four thumbnails, and then leaving. The carousel was a false bottom: it looked like the end of the page, so they never learned there was a page underneath it.
The demo video I'd been careful about was buried below the buy box, the guarantee, the book offer and the whole reviews section. One in ten visitors ever reached it. Nobody clicked it. Not once.
So I rebuilt the page the same day. The carousel is gone, replaced by one still photo. The buy button now sits above the fold on an iPhone. And the first headline is deliberately cut off by the bottom of the screen. A half-visible line is the oldest trick there is for making somebody scroll.
Somebody tapped the search icon in the header, searched, and landed on a stock template page that had nothing to do with us. I audited all 20 addresses the store can serve. The worst find wasn't the search page.
"Your Privacy Choices," the legally required page linked from the footer, was still the old pre-launch landing page. It said COMING SOON, Reserve Today, Deposit $1 Now for Early Bird Discounts, and listed six products called Example product title. Live, in public, while we were paying for traffic.
A shopper can now reach exactly four pages: the store, the product, contact, and the privacy page. Everything else redirects home.
Five
You're right that Marketplace was the original plan. The whole July 10 brief was written to prepare one. Two things moved it down the list.
Marketplace is built for local, one-off, second-hand selling. It's excellent for "I have a couch, come get it." It is not built for shipping 445 identical new units nationwide, its checkout and shipping tools aren't available to every seller, and a listing is public in a way that makes the address question awkward.
And it doesn't teach you anything. A listing sits there. Paid ads tell you, within days and for real money, which sentence makes a stranger stop scrolling. That was worth more to us than a few local sales.
So the money went to paid Facebook and Instagram ads. Same company, completely different machine. Marketplace isn't dead; it's just below churches, coaching groups, corporate gifting and independent bookshops on the list of places these 445 planners are most likely to go. That list is a separate piece of work and it's written up.
Marketplace is a yard sale. Ads are a laboratory. We had 445 planners and no idea which words worked, so I bought the answer to that question first.
One useful thing I found early: no other planner brand is running Facebook ads at all. Empty territory. It also means no one else's homework to copy.
Every ad on Facebook is really testing one thing: the angle. What do you lead with? I picked four:
| Ad | Leads with | Opening line |
|---|---|---|
| 1 · Drawer | The problem | "You've bought a planner before. It's in a drawer somewhere." |
| 2 · Why $36 | The objection | "Thirty-six dollars for a paper planner. Here's where it goes." |
| 3 · Free book | The offer | "There's a goal you keep pushing to next month." |
| 4 · The method | The problem, at length | "What if the goal you've been putting off had a date on it by tonight?" |
| 5 · Still thinking | People who already visited | "Still thinking about the planner?" |
Ad 1 is the one I'm proudest of, and it's the one still running today. The full text:
It works because it doesn't argue with the reader. It agrees with them, tells them the failure wasn't theirs, and then names the exact feature that fixes it. Nothing in it is a claim I can't back.
Four ads went live on July 26. Within fourteen hours, Ad 1 had taken 100% of the impressions and the other three had zero.
Nothing was broken. Facebook tries the creatives, picks a winner fast, and then spends everything on it. At $12 a day across four ads there simply isn't enough money for it to explore properly. A few hundred impressions and it's made up its mind. Video needs more exposure to prove itself than the budget ever allowed.
So I got a confident answer about Ad 1 and no answer at all to the question I'd actually asked. I left it alone rather than interrupt it, because Ad 1's numbers were genuinely good and restarting would have cost real money to un-learn a real result.
The thing I'm careful not to conclude: this does not mean static beats video. The videos were never shown. All it proves is that Ad 1 won an auction early.
Two campaigns ran. Here's the whole thing.
| Old page (Jul 26 – Aug 11) | New page (Aug 9 – 11) | |
|---|---|---|
| Spent | $317 | $49.89 |
| People who arrived | 1,634 | 102 |
| Added to cart | 9 | 20 |
| Started checkout | 1 | 4 |
| Bought | 0 | 1 |
The advertising itself was never the problem. 6.12% of people who saw the ad clicked it, at about 16 cents a click. For context, that's a very good click-through rate. We were buying attention cheaply and losing it on the page.
The first sale came in on August 11 at 6:14pm. $36. It came through the new page, from the new campaign. The web address the buyer arrived on carries the campaign and the ad stamped into it, so there's no guessing about where it came from.
I set up a second advertising channel on OpenAI on July 29, and then didn't spend a dollar on it. The reason is a single number: OpenAI requires a $3.00 minimum per click. Facebook is delivering clicks at 14 to 23 cents.
The same $300 buys roughly 100 visitors there, or about 1,400 on Facebook. It's set up and paused, ready if the economics change. I turned off the automated budget alarm on August 10 since there was no budget to guard.
Advertising accounts are one careless click away from spending money you didn't intend. Two things I did about that.
The paused campaign was more dangerous than it looked. The campaign was paused, but the four ads underneath it were all still switched on, held back only by the parent. Un-pausing one thing would have restarted four ads at $15 a day. Everything is now switched off independently, at three levels.
And I made a real mistake here. Setting the account's total spending ceiling, I read the existing value in cents and wrote the new one in cents to match. That particular setting reads in cents and writes in dollars. I set a $100,000 ceiling instead of $1,000. Facebook returned "success". I caught it on the read-back and fixed it within about a minute; the actual overspend was 25 cents, and only because the daily budget capped it. The lesson is now written into my notes in capital letters: after any money change, read the number back and check it before saying it's done.
Six
There are 44 finished video files in the folder. The demo film alone went through twelve numbered versions before I was happy with it. A few of these are worth telling you about.
In July I built an AI presenter. A synthetic person called Maya, with a face and a voice, talking to camera about the planner. It's an impressive piece of technology and it took a while to make.
Then I watched it back. The planner looked low quality in every shot and was never centred in frame. But the deeper problem was the format itself: a presenter ad borrows its power from a real person vouching for something. An avatar delivers the format and none of the substance.
I retired it, and wrote a standing rule into the project notes: no presenter avatars on this project. If we ever want a face in an ad, it should be yours, on a real camera.
On August 8 I tried a tool that assembles a whole video for you. I attached four real photographs of the planner and asked it to use them. It largely ignored them and generated its own AI planner instead, including a cover with a giant "1%" mark that isn't your logo. Rejected.
On August 5 I tested whether Google's video model could shoot the ad for us. Five eight-second clips: a drawer of abandoned planners, a hand writing the goal, the tabs, and so on. Here's the actual prompt for the opening shot, so you can see the shape of it:
Note the last instruction: no on-screen text. Any words in these ads get added afterwards, by hand, where I control the spelling. And note the small warning I wrote to myself underneath that prompt: the old planner in the drawer must NOT be green and must NOT carry the lockup. It's the generic competitor. Miss that and you've made an ad where your product is the one being thrown away.
The finished ads aren't AI video at all. They're built from real photographs with typography animated over them, assembled in a tool that renders web pages into video. Four of them, 18 to 25 seconds each, in square for the feed and vertical for Stories and Reels.
They're not glamorous. They're honest, they're on-brand, they're re-renderable in ten minutes if a fact changes, and none of them will ever invent a word on your product.
One of them shipped with a real defect I caught later: on two shots the camera drifted past the edge of the photograph and let a strip of green background show down the side of the frame. 189 pixels wide, for nearly three seconds. It was invisible in a thumbnail. I've since written a script that scans every frame edge of every render for exactly that, because eyeballing a contact sheet missed it completely.
Seven
A month of this, in the order it hurt.
I read the contradictory copy on the website and concluded shipping wasn't free. I should have opened the shipping settings, where it plainly says $0.00, no minimum. I banned a true, useful selling point for weeks because I checked the wrong place.
The dashboard was filtered to a period with no online orders in it. The number was right for that window and wrong as a statement about the business.
Covered above. Cost: 25 cents and a small heart attack.
For ten minutes I was convinced a website change hadn't deployed, because the tool I was checking with was serving me a cached copy from ten minutes earlier. The real browser had the new version the whole time. Twice now I've chased a bug that didn't exist because I was looking at the wrong copy of the page.
A photograph sat 15 pixels wider than the page on iOS and nowhere else. The two browsers genuinely disagree about that rule. I now check on a simulated iPhone rather than by narrowing my own window.
The planner is 30 days, not 90. That got corrected on August 6, but by then it had been baked into a graphic on the site, where fixing it meant regenerating the picture rather than editing a line of text.
Every one of these is written up in the project notes with the reason it happened, so the next person, or the next me, doesn't repeat it.
Eight
As of today, August 12, the whole operation is deliberately small:
| One page | The method-first page, now the entire store |
| One campaign | Optimised for people who add to cart, not people who click |
| One ad | The drawer ad, $38 a day |
| One sale | August 11. The first from advertising. |
| $696 | Left under the $1,000 ceiling, about 18 more days |
Everything else is switched off on purpose. One page, one ad, one number to watch is the only configuration where you can actually tell what caused what.
I spent about $370 to find out that the advertising was never the problem. Attention was cheap; we were losing people on the page. So I rebuilt the page around the question they were actually asking, which is "why will I use this one?" rather than "is this worth $36?" Two days later it made its first sale. That's one sale, not a business. But it's the first honest evidence of what's broken, and it's the first thing that's moved.
Nine
| Prompt | The instruction you type to an AI. Everything in the boxes above. |
| Landing page view | Somebody clicked the ad and the page actually finished loading. Stricter than a click. |
| Add to cart | They put it in the basket. Didn't necessarily buy. |
| Click-through rate | Of everyone who saw the ad, the share who clicked. Ours is 6.12%, which is good. |
| Above the fold | What you see before scrolling. On a phone, not very much. |
| False bottom | Something on a page that looks like the end, so people stop. Our photo carousel was one. |
| Pixel | The small piece of code that tells Facebook what happened after the click. |
| Learning phase | Facebook's first few days of figuring out who to show an ad to. Big changes restart it. |
| A/B test | Two versions running at once so you can tell which is better. Only works if one thing differs. |
| Undated | Your planner's actual best feature, and the thing every ad leads on. |