13 August 2026
Why I built Marketing Frameworks
I reviewed 68 marketing frameworks: methods for planning campaigns and writing copy. For each one, I looked for a documented example with a named company, a published result, and a source I could find. I wanted to know what I could rely on when recommending an approach.
For PAS, the copywriting formula Problem - Agitate - Solve, I couldn’t find a reliably documented author or a published comparative test. The method itself is straightforward: name a problem, explain its impact, and offer a solution.
I matched 68 frameworks against 7 product types, giving me 476 combinations. I found published support for 116 of them, roughly 24%. The rest remained undocumented in my research, which doesn’t by itself mean those approaches don’t work.
A table wasn’t enough to choose an approach
The first version was a 68 × 7 table. I still had to stare at it and work out what it meant for my brief. So I turned it into Marketing Frameworks, a tool for choosing methods and preparing an AI brief. It’s currently in Czech.
You select what you sell, what content you’re creating, what your audience knows about the product, and how many similar promises the market has heard. Add a few sentences about your brief. The app suggests a combination of frameworks, shows the level of supporting evidence, and gives you a prompt to paste into your AI chat.
The prompt asks the model to clarify missing information and propose a concept first. It explains who the campaign should reach and which argument it should use. Then it waits for your approval. Only after that should it write the copy for your chosen channels.
The prompt also tells it to cite sources for claims and leave out anything it can’t support. You still need to check the result; an instruction alone doesn’t guarantee accuracy.
The first prompt only looked finished
The app was already live when a user walkthrough showed that the prompt ended with a “FILL THIS IN” placeholder and didn’t explain its scales. Pasted into another AI chat, it lacked the information needed to work. Since then, I’ve tested the whole path through to the resulting copy.
I also started with two prompts: one for the concept and one for the copy. People had to return to the app between them. I dropped that because they could approve the concept in the chat itself. One prompt with an instruction to stop at the right point was enough.
Caffeine gum exposed the limits of a claim
I tested a made-up product with an actual brief and study. The study compared caffeine absorption from gum with a capsule. Twice, the model refused to write “faster than an energy drink” because the study hadn’t tested a drink.
Instead, it built the campaign around using gum where drinking isn’t an option. At the concept checkpoint, it suggested a shelf-edge card to reach customers at the petrol station checkout.
Once, it was too cautious: it refused to divide the pack price of 129 CZK by ten pieces because it “must not invent numbers”. But the brief supplied everything needed to calculate a unit price. I changed the rule to allow calculations from supplied figures, as long as they were labelled.
A brief should set the boundaries of a claim
With marketing copy, I need to know what supports each claim. If the model leaves “TO ADD” in place of a recommended daily maximum, I can see what I need to look up before publishing.
I’d add that to any AI writing brief: use the supplied sources, flag missing evidence, and let me approve the main argument before writing a whole campaign around it.