AI creative tools moved the bottleneck. Volume is now cheap, and winning creatives peak within 2 to 4 weeks, which means the thing deciding your CPA is no longer how many ads you can make. It is how you test them.

This is the testing discipline we run around the Vibelets MCP, and every example comes from a real session, including a weak claim the pre-generate check caught before it cost anything. The other posts in this series cover teardowns, UGC video, and image ads. This one covers the rails under all of them.

Key takeaways

• Everything in the connector edits 1 stateful draft. Starting a new creative resets it, generate executes it, and nothing spends until generate.

• Run review_plan before every generate and actually read it: aspect ratio against placement, claims against what you can substantiate, hook against sound-off viewing.

• Plan the spread before the volume. 4 angles times 5 hooks with varied presenters is 20 tests; the same idea 20 times is 1.

• The kill line after real spend: hook rate at or above 30%, hold at or above 25%, CTR at or above 1.0%. Under all 3 is a loser, not a slow starter.

1 draft, 1 state, 1 moment of spend

Every tool in the pipeline mutates the same session draft, which is why revision is free: swap the product mid-session, re-rank the templates, regenerate the hooks, and nothing costs anything until generate executes whatever is composed. We swapped demo products mid-session while writing this series and the draft simply absorbed it.

The corollary: starting a new creative resets everything. Fresh creative, fresh draft, gone selections. Treat the draft like a cart, not a library.

The check that caught a bad claim

Here's why the pre-generate ritual isn't theater. In our session, the hook generator returned a social proof archetype claiming thousands had already switched to the product. The product had zero customers. Nothing was wrong with the tool, that's simply what the social proof archetype does, but shipping that line buys you a compliance problem and an honesty one.

So the ritual, every single time: run review_plan, confirm the ratio matches the placement you are buying, flag any claim you cannot substantiate, and check the hook still lands with sound off. 30 seconds against a 20 generation mistake.

Spread beats volume

20 variants of 1 idea is 1 test. Before any fan-out above 6, write the grid: which angles, which hooks, which presenters, and make sure no 2 variants test the same thing. The mechanics make the spread cheap: several hooks is 1 video per hook, avatars times scripts multiplies, and image counts apply per template.

Name it before you make it

A 20 way fan-out is untrackable as "final video 3". The convention we use across this series is [product]-[angle]-[hook#]-[format]-[v#], set before generating, because every segment survives as a filter in Ads Manager.

Then the read after launch. Give everything meaningful spend, then apply the kill line: hook rate at or above 30%, hold at or above 25%, CTR at or above 1.0%. Anything under all 3 isn't a slow starter, it's a loser, and killing it fast is what pays for the volume.

Test like the volume is free

Because with the connector, it nearly is. Connect and plan your first 12 variant grid.

Start free with Vibelets

3 prompts to paste

Variant matrix planning

Before we generate anything, plan a 12 variant spread for [PRODUCT]. Give me the angle, hook, and presenter for each one, and show me the grid so I can see the coverage. No 2 variants should be testing the same thing.

Pre-generate check

Run review_plan and walk me through it. Confirm the aspect ratio matches the placement, flag any claim in the copy that needs substantiation, and tell me whether the hook still lands with sound off. Don't generate until I say go.

Naming before the fan-out

Name all 12 variants using [product]-[angle]-[hook#]-[format]-[v#] and give me the list as a table I can paste into a sheet before we generate.

Conclusion

Volume is cheap now, but learning is not automatic. Learning comes from the spread, the check, the names, and the kill line, in that order. Run those 4 and the 20 variants buy you 20 answers.

FAQ

Only starting a new creative. Everything else, source, product, specs, hooks, scripts, copy, edits the live draft, which is why revision stays free right up until generate.

Judge against the 30 / 25 / 1.0 thresholds only after meaningful spend, and remember decisions get cleaner as a campaign approaches the roughly 50 conversions in 7 days that exit the learning phase.

Steal them from ads with staying power. The teardown path returns which dimensions are carrying a winner, and those become the angles your grid varies. The teardown post in this series covers it end to end.

No. Nothing in the MCP touches the Meta Ads API. Finished assets land in your Vibelets library, and you launch from Ads Manager or the app, which is also where the kill line gets applied.

Sources

1. Best DTC Meta ads examples 2026, for creative fatigue windows and problem-aware hook patterns

2. Top 25 D2C brands in India 2026, Meta strategy analysis, for creative refresh cadence at scale