How much creative an automated account actually needs

Ad accounts used to be won on structure. Tight ad groups, clean match types, disciplined bid rules. Google has taken almost all of that back. Matching is automated, bidding is automated, and from September 2026 a large share of accounts get moved onto AI Max whether they choose it or not.

What is left that you fully control is the creative and the feed. So the question stops being how to build the campaign and becomes how much creative to feed it, and how to tell which piece worked.

The number is concepts, not assets

Teams answer the volume question with permutations. Fifteen headlines for a responsive search ad, four descriptions, ten images, done. That is a checklist, not a creative strategy, because the fifteen headlines are usually the same claim written fifteen ways.

The unit that matters is the distinct concept. A concept is a different reason to buy, a different objection handled, a different audience named, or a different proof shown. Three genuinely different concepts beat thirty rewrites of one.

A working cadence for a mid-sized account: one new concept per asset group per fortnight, tested properly, with the losing concept retired rather than left to rot. That is roughly two per month per asset group. For an account with six asset groups, twelve new concepts a month. That is a real production load, and it is why creative capacity, not media management, is now the constraint on most accounts.

Fill the slots, but fill them with difference

Responsive search ads still take fifteen headlines and four descriptions, and you should still supply all of them, because pinning and short lists cut the number of combinations Google can build. Just make sure the fifteen cover different ground: the outcome, the proof, the specific service, the objection, the geography, the guarantee terms you actually offer, the credential you actually hold.

Asset groups in Performance Max want images, logos, video and text. If you do not supply video, Google will generate one from your other assets, and it will look like it was generated. Supply your own.

Asset experiments are the part that changed

Until April 2026 the honest answer to which asset moved it in Performance Max was that nobody knew. Asset-level reporting gave you performance labels, not causal results, and the labels are relative to the other assets in the same group, not to a control.

Asset experiments split traffic between two versions of an asset set and report the difference. That is a real test, and it changes how you should plan creative.

Three rules for running them.

1. 1.Test one variable. Swap the image set or the headline set, not both. If you change everything, a win tells you nothing you can reuse.
2. 2.Give it enough conversions. A test that resolves on thirty conversions is not a result. If your campaign produces twenty conversions a week, a two-week test is not going to separate a 10% effect from noise, and most creative effects are smaller than you hope.
3. 3.Decide the rule before you start. Write down what result makes you keep the challenger. Otherwise you will find a reason to keep whichever one you made.

What to produce, in order

If you are building a creative pipeline from nothing, this is the sequence that gets results fastest.

First, proof. Real photography of real work, real people, real premises. Nothing beats it and almost nobody has enough of it. A single well-shot job site or product session produces months of assets.

Second, specificity. Copy that names the thing. The suburb, the product code, the timeframe, the material, the certification you hold. Generic copy is what the automation already writes for itself.

Third, format coverage. Vertical video for the surfaces that want it, square and landscape for the rest. One shoot, cut three ways, is cheaper than three shoots.

Fourth, volume. Only after the first three. Scaling weak creative just spends more on the wrong thing faster.

The honest limits

Creative volume has a ceiling. Past a certain point you are diluting the data across too many variants for the system to learn anything about any of them, and your fortnightly review becomes unreadable. If you cannot review every active asset in twenty minutes, you have too many.

Creative also cannot fix a poor offer, a slow site or a sales team that takes two days to return a call. Those show up as a creative problem because that is where you are looking, and they are not.

The account work is nearly finished. The studio work is just starting.