Back to blog

More Meta ad creatives: how to keep test records that actually help

Recent discussion of Andromeda has brought creative diversity back into focus, but uploading more versions does not mean running more meaningful tests. This article explains how to distinguish concepts from editing variations, record budgets and attribution conditions, identify results that need retesting, and keep assets clearly linked to the right accounts when managing multiple brands.

A team uploads twenty ads in a week, yet cannot explain at the weekly meeting which selling point worked. The problem may be the records rather than output: changing subtitles on the same video counts as a new concept; results from different countries are compared directly; and assets that barely received any impressions are marked as failures.

In October 2026, discussion of Meta Andromeda continues. AdManage's recent analysis emphasizes separating creative concepts from small variations, while warning that its account sample cannot represent every advertiser or prove that one algorithm change caused every performance shift. That is more useful to take back to a team than a fixed weekly creative quota: first clarify what is being tested and under which conditions.

A new file is not necessarily a new hypothesis

Defining a creative by a hypothesis makes reviews easier than counting files. For travel organizers, for example, helping families spend less time searching for things and helping business travelers pack faster address different situations. Cutting the first video from thirty seconds to fifteen is usually an execution variation of the same concept.

This does not make editing changes worthless. A shorter version may be easier to finish, and different subtitles may be clearer. The distinction is in the level of recording: assign a concept ID first, then record versions of its opening, duration, subtitles and aspect ratio underneath it. Reviews can then compare execution while also revealing whether the team keeps circling around just one reason to buy.

Connect assets to delivery conditions in the record

You do not need to build a huge asset library first. These field groups support the most basic judgments:

Record levelInformation to retainQuestion it helps answer
Creative conceptAudience, use case, main promise, supporting evidenceWhich reason to buy are we testing?
File versionOriginal file, edited version, subtitle language, landing pageWhich exact version appeared in the ad?
Delivery conditionsBrand, market, objective, dates, budget and attribution basisAre the two results comparable?
ObservationsImpressions, spend, clicks, conversions and anomaliesIs there enough evidence to make a judgment?

An ID is only an entry point for retrieval; do not make the filename carry everything. A name with a dozen abbreviations is difficult to maintain. You still need a table that opens the original asset and ad record.

Also save a preview or backend link for the ad actually used. Otherwise, a designer may hold an updated local file while the media buyer analyzes the earlier ad. Both can discuss the same ID while looking at different content.

Unspent budget is not automatically evidence of failure

When a new asset receives few impressions, it has not had enough opportunity to be tested under those delivery conditions. That does not directly prove the concept is ineffective. A review sheet can distinguish pending observation, insufficient evidence, continue testing and stop using, instead of keeping only a simplistic good-or-bad label.

Agree on evaluation conditions before testing: acceptable cost ranges, observation periods, unusual events to exclude and results requiring further validation. Appropriate criteria vary with order value, buying cycle and budget. Experience-based numbers in a service provider's article are not universal platform standards.

When returns decline, do not immediately record creative fatigue as the cause. Stock shortages, ended promotions, website problems and changes in attribution conditions can all affect results. Darkroom's discussion also argues against replacing account data with a completely fixed refresh calendar. Examine trends across versions within the same concept, then check traffic and business conditions to reach a more specific conclusion.

For multiple brands, verify the ad account before acting

Agencies and brand groups often maintain several authorized accounts at once. If a similarly named asset goes to the wrong brand, Page or market, even perfect test records cannot undo the mistake.

In PurpleMark, you can group browser environments by client and brand, keeping the ad backend, brand asset sheet and authorization document links in the same work context. After opening an environment, verify the ad account, Page and landing page before uploading the correct version. Add the environment name and ad record to the asset sheet so the owner can trace it. Allocate member permissions by actual work responsibilities rather than opening every client environment to everyone for convenience.

Environment management helps reduce context-switching mistakes. It does not increase creative conversion rates or replace formal ad-account authorization. Budgets, publishing and pausing ads must still follow client agreements; do not simply synchronize advertising actions across clients.

Weekly meetings should leave a question for the next test

An effective review does more than pick the ad with the highest return. It should answer: which reason to buy deserves further investment? Which versions merely lack evidence? Which result may be affected by a promotion or landing page? Should the next change concern the audience situation, supporting evidence or execution?

For example, “The gift-giving scenario attracted more clicks but lacks purchase evidence” is more useful than “The gift creative is great.” In the next round, check the information on the gift landing page before deciding whether to change the creative. Retaining decision criteria and unresolved questions lets the team build experience instead of naming a new batch of files every week.