You cannot steer Performance Max, but you control the menu it picks from. Most accounts hand Google four images, a handful of headlines, and no video, then wonder why Display eats the budget while Search barely shows up.
The algorithm assembles and places ads across Search, YouTube, Display, Discover, Gmail, and Maps. It can only work with what you give it. A thin asset group means fewer combinations, weaker placements, and reports that tell you almost nothing about what is working.
AI creative changes the economics. Filling a group to Google’s recommended depth used to cost weeks of design time. Now it costs an afternoon, if you know what to load and how to organize it.
The short version: Build each asset group around a clear product, use case, or audience theme. Include distinct messages and formats instead of several near-identical images. AI can produce that variety affordably; human review protects accuracy. Use channel and asset reports to guide replacements, and remember that unequal delivery limits direct comparisons.
What Does Performance Max Need From Your Creative?
Google publishes the requirements, yet many accounts stop at the minimum. According to the Google Ads API asset requirements, a non-retail group needs 3 headlines and 1 long headline. It also needs 2 descriptions and square and landscape images.
Our working target goes further: at least 11 headlines and 4 descriptions, with up to 20 images across three orientations. Add 6 sitelinks and custom video to give Google more combinations to assemble.
The gap between eligible and useful is large. Google will accept 3 headlines and 1 image, but the campaign has almost nothing to test. You get an ad that runs, not one that improves.
| Asset type | Bare minimum | What to load |
|---|---|---|
| Headlines | 3 | 11 or more, varied in length and angle |
| Long headlines | 1 | 2 to 5 |
| Descriptions | 2 | 4 to 5, different lengths |
| Images | 1 per orientation | Up to 20, all three orientations |
| Video | None required | 3 or more, multiple aspect ratios |
| Sitelinks | 0 | 6 or more |
Skipping video does not remove YouTube from the mix. Google may generate a stand-in from your other assets and serve it anyway. You end up running creative on YouTube that nobody at your company designed or approved.
Why Does AI Creative Change the Performance Max Math?
Filling an asset group to that depth used to require a production budget. Three video cuts, 20 images, and 15 headlines could take a design team two weeks. That cost left many groups half empty and dependent on ecommerce feed assets.
AI tools make that depth affordable. A small team can generate variants, resize images, and try different copy angles in a single session. The production bottleneck is gone.
The new risk is sameness. Generate 20 similar product images and Ad Strength may read as complete. But the algorithm still lacks useful comparisons, because diversity matters more than file count.
Before uploading, ask why each asset differs from the one beside it. The answer should be a different claim, format, audience, or moment. If it is the same idea in a different crop, it is padding.
How Should You Theme Asset Groups So the Algorithm Learns Something?
Google recommends separating asset groups by category, theme, language, or audience. Each asset group becomes a reporting unit. Mix three unrelated propositions into one group, and the data will not support a clear decision.
Start with 2 to 4 asset groups built around distinct propositions, not around minor creative variations. One per product line, one per core use case, or one per audience segment.
| Asset group theme | Why it is a valid split | What the data will tell you |
|---|---|---|
| By use case | Different buyer problem and language | Which problem converts cheapest |
| By product line | Different margin and price point | Where to push budget |
| By audience stage | Cold versus in-market messaging | Whether you need more demand creation |
| By crop or color | Not a valid split | Nothing, this belongs inside one group |
The last row shows a common mistake: separating groups by appearance alone. That splits conversion data across groups that each learn slowly. Keep cosmetic variations together so the algorithm can compare them within one group.
Running PMax on a half-empty asset group?
We pair AI performance creative with the paid media program that spends against it, so the algorithm has real options to choose from.
Book a Free Strategy CallWhat Can You Finally See Inside Performance Max?
Channel performance reports now show how Search, YouTube, Display, Discover, Gmail, and Maps contributed during a chosen period. Asset reports have improved too. Together, they give you a clearer picture of where PMax delivers and where it wastes impressions.
Google says asset reports now support device, time, conversion, and network segments. You can compare one asset on YouTube versus Search instead of seeing only an aggregate. The update also added cost charts, ROI columns, and account-level downloads.
That is a real change for anyone deciding what to replace. Before it, “this image is Low rated” was most of what you had.
How Do You Read Asset Reporting Without Fooling Yourself?
Use asset reporting as a clue, then confirm it with a structured test. Performance Max gives assets unequal delivery, concentrating early impressions on a few predicted winners. The others may never receive a fair sample, so the report alone cannot settle the comparison.
Three habits keep the reads honest:
- Wait before judging. Google advises 2 to 3 weeks before evaluating and replacing assets. Pulling an asset after 4 days can kill something that never got distribution.
- Compare within a network, not across. A headline that looks weak overall may be carrying Search while a video absorbs YouTube. Segmenting by network is what makes that visible.
- Replace, do not empty. Swapping a Low asset for a new one keeps the group complete. Deleting without a replacement shrinks the menu. It can also leave the group without enough coverage for every placement.
For a clean concept read, use a channel where you control delivery. Our creative testing framework for paid social explains the process. Add proven winners to PMax instead of guesses.
What Guardrails Keep AI Creative From Degrading Performance Max?
Volume without judgment can hurt an account. Half-empty groups give Google too few options. At the other extreme, 20 unreviewed AI assets can expose valuable audiences to false or off-brand claims.
Four rules hold the line:
- Human approval on every asset that ships. Generation is cheap, so every asset still needs a reviewer. Nothing enters an asset group unseen.
- Claims get checked, not generated. Pricing, guarantees, and outcome numbers come from the business, never from a model filling in a plausible sentence.
- One brand system across variants. Logo placement, palette, and type stay fixed while message and format vary.
- Refresh on a schedule. Rotate new assets into high-spend groups regularly rather than waiting for performance to visibly decay.
These rules are familiar to a good creative team. The same discipline now applies to a pipeline that produces far more assets each week than it used to.
What Should You Do Next?
The menu metaphor holds all the way through. You cannot tell Performance Max where to serve your ads, but you decide what it has to serve. An account running on four images and no video is handing Google a short menu and complaining about the meal.
Load each group properly and theme it so the reporting means something. Use channel and network data to decide what to swap. For the wider Google Ads picture, see why YouTube can beat PMax outright for some niches.
If you are unsure whether the gap is your assets or how they are grouped, we can help you work through it. Bring your asset groups and last 90 days of channel reporting. Book a Free Strategy Call to discuss what to add, what to replace, and what the data cannot yet tell you.
Like this? Get the next one.
Short emails. New posts as they ship.
