i've been automating our ENTIRE google ads research flows inside claude code…
and the results have been INSANE
so i decided to document the ENTIRE system..
covering the setup, workflow, and ready-to-run prompt templates to research winning google and youtube ad in SECONDS inside ONE chat
here's what's included inside the guide:
> the ICP deep-research master prompt
(mines reddit, quora, amazon and forums for the exact language customers use, plus 2 follow-ups that go deeper)
> the competitor teardown
(homepage, ads and review-sentiment, compiled into the conquest weakness file)
> google ads angle extraction
(research turned into ToF, MoF and conquest keyword clusters with the copy written)
> the voice-of-customer doc
(87 verbatim phrases off one brand, ranked by frequency, every swipeable hook lives here)
> the GMC feed rewrite
(titles that match how people actually search instead of "supplement 60ct")
> niche templates, the research cadence, funnel architecture, and the compounding research library
(everything past phase 5, including the page-type map and the weekly re-pull)
all backed by everything i learned after generating +$20M for ecom brands at my google/youtube ads agency
like + comment "GUIDE" and i'll send it over
(must be following + RT for priority access)
turning json prompts into high roas ads is mostly a translation problem
models perform best with structure and constraints, but most tools cant accept raw json
when people rewrite prompts manually
they flatten hierarchy and lose intent
and performance drops
this guide shows how to convert json logic into plain language
while keeping structure tone and creative control intact
so ads keep converting even without native json support
rt + comment “json” and i’ll send the guide
(follow for dm)