Hey friend,

Alright, here’s the deal. I’m working on nailing down a new format for Signals In The Noise.

The new focus is about helping owner-led teams learn to use AI to automate repetitive, tedious work so their business can become leaner and more profitable.

That’s the aim. Find the good pieces of seemingly complicated technology and show you how your and your employees can use them to make your business lives less annoying and more efficient, all while adding zeroes to the profit line of your P&L.

It might take me a few issues here to find the best structure for this new version of Signals In The Noise, but I’m pretty clear on a few things already:

As you see me experiment with new structures for this newsletter, if you have any thoughts about how this newsletter could be better for you and your team, I’m all eyes. (Because I won’t actually hear your email with my ears, so I can’t be all ears.)

And now, it’s time for me to shut up with this preamble and get to some practical things you and your team can do with AI.

Things are about to get meta-Meta in here. It’s no longer enough to just get meta. I now have to get meta about creating Meta ads. I’m going to do a thing I thought I’d never do. I’m going to…gasp…build in public and show you how I’m actually building things in my quest to make Signals In The Noise the world’s best AI newsletter.

Let’s go.

(Want to read this tutorial in a better looking version on a webpage? I got you. Click here.)

Alright, let’s get practical tactical, but first, I want to do…

A brief aside to recommend a newsletter platform I love.

Give me one minute here. It won’t take long. Promise.

While I’m writing this issue of the newsletter for you, I’m also finishing the setup of my Beehiiv site, which is what I’m using to host the Signals In The Noise newsletter and website. I’ve used A LOT of software over the last 20 years of running various businesses for myself and my wife.

And I have to say, Beehiiv is a joy to use. It does what I want it to do, and they did a great job with the user interface. Everything is where I expected it to be. Well done, Beehiiv team. If you’re thinking of starting a free or paid newsletter and want a bunch of extra features, like being able to sell digital products, Beehiiv is waaaaaay better than Substack.

Yes, I can hear some grumbling from the back of the room about how Substack makes it easier for your newsletter to get traffic from new potential readers because of the social network and discovery algorithm they’ve built. I’m not denying that that part of Substack is better than Beehiiv’s version of that, but…and this is a big BUT…Substack takes 10% of your revenue from paid subscriptions.

That may not sound like much when you’re starting out, but, to take one example, if/when your paid newsletter starts making $10,000/mo, you’ll be giving Substack $1,000 of your revenue every. Single. Month. If your newsletter makes $20,000/mo, you’ll be giving them $2,000 per month. Their 10% fee is for newsletter software and a discovery algorithm that is supposed to make it easier for you to grow your newsletter subscriber list. 10% is steep.

Conceptually, I’ve never liked their business model. Maybe the math works out for some writers on Substack. Maybe the better discovery outweighs that enormous cut that Substack takes. Personally, I’d rather not play that game. I’d rather have a great newsletter app that gives me more flexibility than Substack, with more features that Substack doesn’t have, and that also has a low, flat monthly fee that doesn’t penalize you when your newsletter starts making decent money. That’s where Beehiiv comes in. So far, I’m loving it.

Okay. I’ll pause with my Beehiiv raving and my Substack ranting. If you want to give Beehiiv a whirl, you can get an extended 14 day trial and 20% off your first 3 months if you sign up via this link, which will cause me to get a small commission with which I shall buy an artisan hipster mocha drink with.

But back to the main story. While I’m writing this issue of SITN, I’m also creating ad content for ads that I’ll run on Meta properties (Facebook, Instagram, and Threads.)

I thought you might like to see the workflow and tools I’m using to do this, because it’s WAY faster than my old, pre-AI tool workflow for creating ad creative for Meta ads.

(Pre-S, which is like P.S., but before the main content: if you’d like these kind of tutorials in a downloadable PDF format that you can send to anyone on your team, just say the word and I’ll do that next time.)

Want a nice looking, easy-to-read, downloadable HTML file version of this tutorial that you can send to someone on your team? Click here to download it.

I gave Claude (and as of this writing, I’m using the Opus 4.8 model at max effort level, even though Opus 5 and Fable 5 are available. The latest models aren’t always the best for every task) a link to my Signals In The Noise subscription sales page, and gave it this prompt:

First thing to notice there: you do not need to be some kind of magical prompt engineer to get things done with AI chat tools. Talk to Claude or chatGPT or Gemini like you would if you were talking to a very smart co-worker or independent contractor/subject matter expert.

Second vital tip here: the more context you can give these models, the better. And by context, I mean things like any writing you’ve done about a topic, sales pages you’ve made, journal entries, executive briefs, slide decks, writing examples of yours, writing examples of others if you’re aiming for a particular kind of writing voice, and so on.

What you need to give for background context changes depending on what you’re trying to accomplish. In future issues, I’ll show you my workflow that generates amazing website and email copy that gets people to take actions, and when I do that, you’ll see the full extent of what I mean by “give the models A LOT of background context.”

Claude gave me a good response, and gave me some options for Meta ad copy, and suggested static image ideas. (Why am I doing only static image ads and not video ads despite everyone and their best friend in the marketing world saying that video ads are crushing it right now? Because I hate being on camera. If you ever see me show up in a video ad in one of your social media feed, that will be your sign that hell has frozen over.)

Claude read my sales page and grabbed 4 different concepts I wrote in that copy, and then used those concepts, along with most of my copy, in its Meta ad copy ideas. It shortened and tightened things up while maintaining my brand writing voice, which I’m thankful for, because even though, given enough time, I can edit my writing enough so that it looks like I can write short, writing short is not one of my strong suits. Those last two sentences are, in fact, great examples of my problem. Lol.

Claude also noticed that I put a photo of me with a baby goat on my back below my name at the bottom of my sales page, and suggested that one version of ad copy could work great with that image as the image for the ad.

I went back into my photo library, found the original photo of the baby goat standing on my back, and then started working on fixing it so I could use it in an ad with a bit of words on top of the photo.

Here’s the copy it suggested to put over the top of the photo:

Hero: Hate AI slop? Me too.
Subline: There's a good kind of AI, though. One that automates boring tasks.

But in order for those words to be visible somewhere in the photo, which you’re about to see 9 times below, I needed to remove some things from the background of the shot.

To remove some things from the background, I first tried the chatGPT image tool. It can sometimes do really well with this, but not today.

Here’s the prompt I used:

Remove the people behind me and behind the goat's legs, and also remove the power lines and fence in the background. Extend the same color of sky that's behind me and above the power lines and fence to the areas where you remove the people, power lines, and fence. Make sure to not remove the goat's leg or any part of his leg. There are people standing right behind the goat's legs so be careful to just remove the people and not the goat's legs. Leave everything else in the image untouched.

Here’s the raw, original photo:

Cute goat! He was very happy to be standing on my back and was there for a good 5 minutes. But…there’s people in the background on one side, and a not so great looking fence in the background on the other side.

After trying to get chatGPT’s image tool to remove those things, here’s what I got back:

It’s not bad, but it’s process introduced some not-so-great alterations to my face and the goat, because that’s how chatGPT’s image tool works. It rebuilds the entire image based off of your prompt, even if you tell it to not alter some portions of the image.

In this case, it removed objects from the background well, but ended up adding some weird artifacts, and changed the lighting and color.

So instead, I used an app called TouchRetouch, which did a good job at removing the background objects without rebuilding my face from scratch, and then I used some finer grained stamp clone tools inside TouchRetouch to fix some of its mistakes.

After 10 minutes of work, here’s the result:

Getting there. Could I have done this in Photoshop? Yes, but it would have taken me about an hour. And, side note, I did try Photoshop’s AI powered background remove tool and it’s generative fill tool as well. Both of them did a horrendous job. It added in all kinds of weird stone walls in the background. Not good and not useable.

TouchRetouch was much better for this job.

I particularly like the “Quick Repair” tool in TouchRetouch. You set the circle size of the tool, and then zoom in on the photo, click on any weird little bits you’d like fixed, and poof! They magically disappear.

Because I had goats climbing all over me before I took this selfie, I had all kinds of little specks of hay and dirt on my coat and face. 20 quick clicks with the Quick Repair tool and those all disappeared. Brilliant.

I then took that above version of this photo into Lightroom for a quick 1 minute edit to fix the lighting and colors:

That’s better. The colors and lighting are better now. It’s not perfect, but I’m not aiming for perfect. I’m aiming for good enough so that this still feels like a 100% authentic, free-range, organic selfie, which it is.

Next, I took that last version of this image of me and the goat over to Claude Design (if you’re using the Claude desktop app, Claude Design is accessible in the lower left. It’s just a click away and will open up a separate app window), and gave it this prompt:

It gave me back a few options, and it’s smart enough that it created them in the current size guidelines for Meta ads. Could I have done this in Photoshop? Sure. I would have had more fine grained control over exactly where the text would be, but using Claude Design is faster, and for what I’m doing with this static ad image, it’s plenty good. The 80/20 rule is in effect.

I wanted to see if it could fit the second line of text on that second version without putting it over my hair. I worked hard to get my hair right for this photo, so I don’t want text over it. Okay, I’m kidding, but having text over my hair does make it harder to read. (Don’t you hate it when you get text stuck in your hair?)

So, I took a screenshot of the part I wanted it to fix, then drew a red highlighted area on top of it with the screenshot tool on my Macbook so I could point to the area I want fixed. I uploaded this screenshot to Claude Design:

This may seem like a simple little thing, but it is SO FREAKING USEFUL. Seriously. Just take screenshots of anything and everything you have questions about when you’re working with Claude or Claude Design or Claude Code. Draw on the screenshots and then tell whichever version of Claude you’re working with what you need help with the thing you’ve taken a screenshot of and highlighted.

Do this often enough and I’m telling you, Claude will start to feel like a very patient, smart employee on your team, or, it can feel like a smart co-worker for every human employee on your team. It can see and fix and guide you if you give it enough context and images and links to things on the web.

I gave Claude Design that screenshot with the red circle and asked it to re-arrange the text so it didn’t cover my hair.

Claude Design took about 45 seconds and fixed it:

I probably won’t end up using this one for an ad, because once that image is shrunk to fit in the Facebook, Instagram, and Threads feeds on people’s phones, that second line of text is going to be too tiny to read easily while people are scrolling.

And making that text bigger would create other problems, because there just isn’t much room there for a second line of readable text.

Instead, I will use the version of this image with just the big text “Hate AI slop? Me too.”, on it and then move that second line of text to the description text of the ad, which will appear just below the image.