From niche research to a validated ad: running a real Meta campaign with AI
Every other project in this log points the same method — research, plan, produce, verify — at games and 3D tools. Could it run a marketing engagement end to end? Not a mock-up: a real local business, a real audience, real videos, real money in Meta Ads Manager, and a measurable answer about which creative earns attention cheapest.
Two things. First, the sequence beats the spend: a month of organic posting is a free A/B test lab, and by the time money enters, you're only paying to amplify content that already proved itself. Second, Ads Manager itself is the hardest terrain in the whole pipeline — the traps we hit and documented (drafts that exist for one login and not another, a hidden pre-selected post that nearly attached an outdated promo to a live ad) are now the most valuable pages of the playbook.
Not stopped — productized. The whole arc got distilled into a versioned playbook: six phases, copy conventions, the access checklist, the Ads Manager traps, real hours per phase. The next engagement starts from page one of that document instead of from zero.
The arc, start to finish
The engagement ran the full distance — no phase skipped, no phase faked. Names withheld; the client is a local practice in a regulated sector, which made everything harder in instructive ways.
1 — Segment first. Before any content: who is the ideal customer, what's the entry product, who are the three direct competitors in the same city, and what does the sector's regulation forbid you to say in an ad. That last one shaped every script that followed.
2 — Editorial plan. A dated calendar built on three layers — universal commercial dates, niche dates, the business's own — plus a hook system and one DM trigger word per piece, in caps, so replies become countable leads instead of vibes.
3 — Production, designed for a non-actor. The client isn't a creator, so the system adapts to them: one card per video with hook, message and CTA; cue cards in large type used as a teleprompter; recording rules (no brand names, no prices on camera) that keep every clip compliant before it exists, instead of filtering it after.
4 — Post with the plan's language. Captions written from the editorial plan's research — bilingual, five hashtags maximum, the trigger word, a fixed signature — at a sustainable three pieces a week. This month of organic posting is the quiet genius of the sequence: it's creative testing with a budget of zero.
5 — Ads, finally. Only then Meta Ads: a visibility campaign around the business's actual location, built from existing posts that had already earned engagement organically. Two different creative formats went head to head as separate ads under one campaign budget, and Meta's delivery plus our checkpoint reviews did the voting.
6 — The verdict. Thirty days in, the two formats stopped being a matter of opinion. The straight service reel — one treatment, presented directly — delivered profile visits at $0.56 each. The educational step-by-step walkthrough cost $0.84 for the same result: a full 33% more expensive per visit. And because the budget lived at campaign level, Meta's own delivery had quietly reached the same conclusion, steering nearly twice the reach toward the winner before we ever read the numbers. That figure is the whole point of the sequence: a data-backed answer to "what should this business's next dollar buy?" — and a benchmark the next month's creative has to beat.
What actually almost went wrong
The production phases went smoothly. The near-misses all lived inside Meta's tooling, and they're worth publishing because nobody warns you:
- Drafts are per-user. A campaign built under one login doesn't exist for another.
- The "use existing post" picker can pre-select a post you can't see — an old promotion with expired pricing was silently attached until a final review caught it. The Add button adds; it does not replace. Read the review tab every time.
- The real bottleneck is access, not production. Pages, ad accounts and Instagram profiles scattered across business portfolios of owners, relatives and former vendors. Untangling permissions took longer than making the videos.
Why this entry is here
Because the method transferred. Research the niche before touching the tools; make the plan a file, not a vibe; let cheap organic reality filter the creatives before money amplifies them; verify in the hostile environment (Ads Manager) with a checklist written in blood. That's the same loop that builds my game prototypes — pointed at marketing, and it held. It now lives as a playbook under my consulting label, SparkTeam, and every new engagement is required to make it better.