Building Automation Tools For Paid Ads

[eliquid]

I've spent a large part of my life trying to automate paid ads in 1 way or another.

While I know we have things like:

1. Google Gems
2. Claude Projects
3. ChatGPT Projects and 3rd party GPTs
4. Tools specifically like Arcads, HeyGen, etc...

Lets face it, they all suck or get things wrong.

They def don't compare to someone doing things for 27+ years.

And other tools people make, are just wrappers to Claude or Gemini or DeepSeek or whoever... and they just push the entire request to the Ai and let Ai DO ALL THE WORK with no additional context.

I got tired of that.

So what's better?

If you have a ton of domain knowledge ( you are an expert or close to it ), replicating yourself and giving that to Ai is so much better than the above options.

I'm talking about PKM ( personal knowledge management ) and THEN Ai/LLM.

And this doesn't have to be YOUR personal knowledge, but it shouldnt be Ai dumps either ( sorry no Perplexity crap here ).

---
So I started to use Obsidian on my Mac. Think of it like Evernote, Notion, Apple Notes - but 1000x better.

And in Obsidian I realized it's just markdown files in folders. But you can add in plugins.

And once I entered in all my knowledge about paid ads - because I didn't want to keep this in memory and I didn't want scattered notes in paper or on Google Drive - I realized I could use plugins on Obsidian to do even more.

I started with adding Ai tools to it.

To chat with my notes ( but this is way better than NotebookLM ), to give me advice , to plan out tasks, and then finally to create paid ads.

That extended to me getting Cursor ( or pick any IDE ) and adding Claude Code, Gemini CLI, and Codex to Cursor and then using Cursor ( with Ai ) to chat with the same markdown file notes in Obsidian.

To finally moving everything over to Cursor.

And what does that give me?

Perfectly made paid ads, exactly targeted to my customer demographics/Audiences ( based on tons of research ).. at the rhetorical appeal ( Logo, Ethos, Pathos ) I want.. at the emotional avatar I want.. at the right market awareness stage I want.. with trustpilot testimonials pulled in.. and more.

It's writing video scripts and creating image ads better than any other tool out there.

Wanna see an example?

This is an "benefits call out" ad I made for a skin care company for:
Targeting Summary:
- Customer: Established Professional Women (Ages 45-60)
- ABC: [Hidden so you can't copy]
- DEF: [Hidden so you can't copy]
- Awareness: Solution Aware

If you know deeply about any of the above, you know there has to be a specific ad made to appeal to this person.

And Ai right now ( other tools, options ) just don't cut it.

You may give them the same prompt with specifics above, but you want get back a true Solution Aware ad. Or a true ad for Ethos - even though you told it to.

And you for sure wont get back legit real TrustPilot testimonials. It will shortened or cut off.

And it also pulls the style guideline of the brand automatically. So every ad is "on brand" with the font choice, color palette, and more.

And it learns what you like, and don't like so that future iterations and generations are more accurate each pull.

And this is the image ad.

Here's another, but Im not crazy about it. Need to clean it up.

It's for a different demo, pain point, awareness level, etc:

These ads end up being automatically generated for the different persona's, avatars, demos, audiences, pain points, objections, desires, etc - your market is after.

Other tools just pick 1 or 2 things and never deepdive.

They never triple check and learn.

And this is where building tools for paids ads ( or any subject ) using your PKM will outdo regular Ai all day.

And static ads arent the only output. I havent even shown you the video scripts yet that cross all types of story arcs, frameworks, and more.

Questions on how to implement your own PKM with Ai?

Hit me up here.

 

[Ryuzaki]

Very cool. As I was reading and thinking, the first thought that came to mind was "YouTube thumbnails". The second, back-on-track, thought was "I wonder if there's a way to get these ads to be delivered to you in Photoshop layers."

Is that possible? Seems like that would make being able to make slight variations easier, unless these image generators are stable enough now to not introduce slight variations on each generation. When I was doing this before you couldn't do "in-painting" so it would dream up entirely new images each time. But times have changed. With in-painting, could you select one word to change and have it nail the alignment and font and kerning and everything, or do you have to generate an entirely new version and hope it holds true to the rest of the image as it was and as you like?

That's not a big deal either way. It can nail the design and then you could recreate it yourself in Photoshop if desired. What I think is impressive is the copywriting aspect of it and really defining customer profiles and understanding the problem-solution aspect of it. If it can understand the insecurities of the buyer and really poke those pain points, that's money in the bag.

 

[eliquid]

Very cool. As I was reading and thinking, the first thought that came to mind was "YouTube thumbnails". The second, back-on-track, thought was "I wonder if there's a way to get these ads to be delivered to you in Photoshop layers."

Is that possible? Seems like that would make being able to make slight variations easier, unless these image generators are stable enough now to not introduce slight variations on each generation. When I was doing this before you couldn't do "in-painting" so it would dream up entirely new images each time. But times have changed. With in-painting, could you select one word to change and have it nail the alignment and font and kerning and everything, or do you have to generate an entirely new version and hope it holds true to the rest of the image as it was and as you like?

That's not a big deal either way. It can nail the design and then you could recreate it yourself in Photoshop if desired. What I think is impressive is the copywriting aspect of it and really defining customer profiles and understanding the problem-solution aspect of it. If it can understand the insecurities of the buyer and really poke those pain points, that's money in the bag.

For the layers/PS question - Not able to generate a PSD or similar type of file. The outputs in Ai ( seems all models and LLMs ) is jpg, png, gif. At the moment at least.

But my system actually labels the areas on the image and writes what needs to go there.

So the headline area is labeled as Headline 1 and Headline 2 ( second line ).

Same with the call outs around the hero product. All labeled.

So what I want it to say, where it is aligned, the font, the color, etc - is all dictated before the image generation happens.

Which, is what people tried to do before, right? Before you would say to the Ai "I want this image and this background and then have it say ABC XYZ" and it would generate all new and most times was shit.

Kinda the same system now, but Im doing a large part of the prompt in such detail and in a fashion that the ONLY thing I have to worry about it messing up right now is the "hero" product part of this.

The background, the copy, the color, the font, spacing, etc is all nailed down. The product hero shot itself is also nailed down a lot, but for some reason Ai will sometimes do something odd ( maybe 1 out of 9 runs ) to the product. If that happens, I just regenerate and consider that the cost of "doing business".

I haven't had to edit an image yet, but my back up is Canva and Lightroom with their Ai features built in the tools to "edit" or correct something. But it's so fast ( and somewhat cheap ) to just regenerate that if I get a product it twisted to the right too much in the shot, I just regenerate.

All other properties stay great atm.