Building ad variations by hand is slow and produces inconsistency that makes the results hard to read. This guide is for whoever produces and ships creative. It covers the copy bank, the queue, and pushing to Meta, and why building variations systematically makes your testing conclusions trustworthy.
The copy bank
The copy bank stores reusable copy in groups rather than as one off ad text.
A group holds up to five headlines and up to five primary text variations, together with a call to action, the landing page it points at and its destination URL. Notes sit alongside for context that is not part of the ad.
Storing copy this way rather than per ad is what makes bulk building possible. Five headlines and five primary texts produce twenty five combinations from ten pieces of writing, and every one is consistent with the others because they came from the same set.
Keep groups meaningful rather than large. A group should represent one angle or one product, so that when it performs you know what performed. A group that mixes several angles produces results you cannot attribute to anything.
Why systematic beats ad hoc
The reason to build this way is not only speed.
When variations are written independently, each differs from the others in several ways at once: different hook, different length, different tone, different call to action. When one wins you cannot tell which difference caused it, so you learn nothing transferable and the next round starts from scratch.
When variations are combinations from a controlled set, the differences are known. A headline that wins across several primary texts is a headline that works. A primary text that lifts every headline it is paired with is a genuine finding. That is how a creative programme compounds instead of restarting.
The queue
Built creatives go to a queue rather than straight to the platform.
The queue is a review step, and it is worth using properly. Bulk generation produces combinations that are grammatically fine and occasionally nonsense, and the cost of one embarrassing live ad is far higher than the minute it takes to scan the list.
It also gives you a staging point. You can build a large batch when you have time and release it in waves, which is usually better than launching everything at once, because launching twenty variations simultaneously splits budget so thinly that none of them gathers enough data to be judged.
Pushing to Meta
Pushing sends the approved items to Meta through the connection you set up when you connected the account. The same permissions that let Tatheon read performance are what allow this, which is why this is the one place Tatheon writes rather than only reads.
Push in batches small enough that each variation can accumulate meaningful spend. The scale, wait or kill verdict needs enough spend behind an ad before it can say anything, and a batch that spreads budget across too many ads leaves every one of them permanently in wait.
Closing the loop
The point of building systematically is what happens afterwards. Performance flows back into the creative library, tagged by hook, angle and format, which is what lets you see patterns rather than individual results.
So the cycle is: read the library for what has worked, build a copy bank group around that angle, generate variations, review, push a batch, wait for enough spend, and read the results back in the library. Each pass narrows what you know works.
What to do first
Build one copy bank group for your best performing angle from the library, with three headlines and three primary texts rather than five and five. Nine variations is enough to learn from and small enough that each gets real spend.