The problem this solves
Most creative tests are compromised before they launch. The plan calls for a clean comparison, but production pressure means one angle gets the polished asset and the other gets whatever was left over, so the test measures production quality instead of the angle. Or both variants ship in a single format, and the result confounds message with placement. Getting a genuinely clean two-by-two - two angles, two formats, consistent execution - is a production discipline problem more than a media one.
Then there is the staging gap. Even with four clean assets in hand, someone has to build the campaign structure, attach the right asset to the right ad, and resist the temptation to tweak copy mid-setup. That last mile is where tests wait for days, because the person with ad-account access is rarely the person who made the creative. Every day the matrix sits unstaged is a day the account keeps spending on the old creative it was supposed to challenge.
How the mission runs
- Sharpen the two angles. The agent takes your two strongest angles and tightens each into a testable claim with matched copy discipline: comparable hook strength, comparable specificity, comparable calls to action. A fair test starts here, because an angle handicapped by weaker copy produces a result that reads clean and means nothing.
- Check the field first. A quick pass through Ad Libraries confirms neither angle collides head-on with a message a competitor is currently saturating. If one does, the agent flags it with the evidence and proposes a sharpened framing, so the test measures your angles against each other on ground you can win.
- Generate the two-by-two. Image Generation produces four assets: each angle rendered as a square and a story-format image, with execution quality held deliberately even across all four. Concepts iterate at draft quality until the pair of directions is approved; the four test assets then render at final quality.
- Stage the draft, paused. The agent builds the campaign in Meta Ads as a paused draft: the two-by-two structure laid out, each asset attached to its correct ad, copy in place. Paused is the contract - the structure exists, the assets are attached, and spend is impossible until a human flips the switch.
- Hand over the launch checklist. The mission closes with a summary in the thread: the matrix logic, each cell's asset and copy, and what to look for when results arrive - which cell winning would mean the angle carried it, and which would mean the format did. You review the draft and launch when ready.
The prompt
This is the exact objective the agent receives. Swap the obvious placeholders for your own domain, segment or channel and run it as-is from the console, Slack, or the API.
What comes back
A paused Meta Ads draft containing a complete two-by-two creative test: two angles, each executed as a square and a story-format image at final quality, four assets attached to their correct ads with copy in place. Alongside it, a thread summary explaining the matrix logic and how to read each possible outcome. Launching remains a single human action in your own ad account.
Make it yours
- Widen the matrix to two-by-three by adding a landscape format per angle, when the media plan includes placements where landscape historically earns cheaper reach.
- Run it as a challenger cadence: each mission tests the account's current control angle against one new contender, keeping a permanent testing lane without ballooning spend.
- Stage the same matrix twice with different audience definitions attached, so the follow-up question - does the winning angle hold across segments - is already built.
Frequently asked questions
Why does the campaign stage paused instead of launching?
Because spend is the one action that should always be human. Paused staging gives you the entire structure - assets attached, copy placed, matrix intact - with zero risk of an unreviewed campaign spending money. You inspect everything in your own ad account and launch with one click when satisfied.
Can I adjust copy or swap an asset before launch?
Yes, either way: edit directly in the draft, or ask in the thread and the agent regenerates the asset or revises the copy and re-stages it. If you change one cell, the agent can rebalance the matching cell so the matrix stays a fair comparison.
Will all four assets really look consistent?
Consistency is the point of generating them in one mission. All four come from the same approved visual system at the same final quality tier, so no cell wins on production polish. The brand kit governs palette and type across the set, keeping every variant unmistakably yours.