Let AI do 100 percent of it and you make zero dollars. Here is which 40 percent has to stay yours, and the skill that runs the rest.
AI can build the middle of a digital product very well. It cannot build the ends. And the ends are the entire reason anybody buys.
This is the 20/60/20 version of the AI Sandwich Method: you direct the first fifth, AI executes the middle three fifths, and you take the last fifth back to make it something a person would actually pay for. The skill at the bottom runs the whole thing and refuses to let you skip the parts that matter.
Before anything else, here is the finished product, free. It is a field guide for e-commerce brands scaling Meta ads from seven to eight figures. The economics to lock before you raise a budget, how to build one clean signal, the creative portfolio framework, the testing order that buys you a decision, how to scale without the death loop, and a 30 day operating rhythm.
Ten pages, and it is genuinely the thing I would charge for. Take it as a reference if you run paid ads, and if you want to go deeper on that, here are the three lessons behind it plus the analyst skill that reads your ad account.
Everything below is the exact process I used to build it. The ebook is what that process produces.
This is research, not prompting, and it is the part that decides whether the thing sells. Telling AI "make me an ebook about Meta ads" gets you the most generic product on the internet.
Step 1
Pick something you can fact-check
Not just something you find interesting. If you cannot tell whether the finished draft is true, you cannot defend it and you cannot fix it. Real expertise is what makes the last 20 percent possible at all.
Step 2
Answer four questions before AI writes a word
Who exactly it is for. What problem you are solving. What gap nobody else is closing. What promise you are making to the person who buys. Vague answers here produce a vague product. "Helps with ads" is not a promise. "Know within 48 hours whether a campaign is worth scaling" is.
Step 3
Dump in everything you already know
Old reports, strategy docs, notes, tests, spreadsheets, transcripts. Zip it and hand the whole thing over unorganized. This is what stops the product from being a rewrite of what is already on Google.
Step 4
Show it what good looks like
Go through Pinterest and Canva, save five to ten layouts you actually love, and hand AI the images. Showing beats describing by a mile. Then lock three colors and two fonts before a single page gets built.
Most AI-built products look like five different products stapled together, because every page was generated fresh. The fix is one sentence.
Perfect page one. Then for every page after it, tell AI to match page one exactly and only change the content. Same layout, same type sizes, same spacing, same rhythm. That single instruction is the difference between something that looks designed and something that looks generated.
What AI's 60 percent actually covers
Researching the gaps in your material, flagging what has changed since you learned it, organizing everything into a real structure, and writing the first full version. That is the grunt work. It is genuinely most of the hours, which is exactly why it is worth handing over.
Do not skip this. Most people ship the second AI hands them something, and that is the version nobody buys twice. Work this checklist:
This runs the whole process. It interrogates your expertise before it agrees to start, refuses to move past the four decisions until you answer them specifically, marks the places it needs your real example instead of inventing one, and hands the product back with the review checklist attached.
The one rule underneath all of it
People do not buy from AI. They buy from people. AI is allowed to do the middle, and it should, because that is where the hours are. The direction and the judgment stay yours or there is nothing to pay for.