A Performance Max asset group accepts up to 15 videos per orientation, requires at least one of each of the three, and none may be shorter than 10 seconds. And if you upload none, Google may generate one itself from the rest of your assets. That last sentence decides the quality of most Spanish campaigns, because almost nobody reads it in time.

The video Google builds in your place

When the asset group has no video, the system assembles one automatically by combining the images, the text and, where it exists, the feed information. Technically it works: the campaign runs and serves impressions.

The problem is what that video is. A montage of product photos with captions laid over it and a generic transition shows nobody using anything, and it competes on surfaces where the other ads do.

Why it is spotted late

Because nothing fails. There is no error, no warning, the campaign spends normally, and the report shows mediocre performance that gets blamed on budget or bidding. The automatic video is the suspect nobody interrogates.

The check takes two minutes: open the asset group and look at whether the videos in it are the ones you uploaded.

What an asset group asks for

OrientationRatioRecommended resolutionRequired minimum
Landscape16:91920 x 10801 video
Square1:11080 x 10801 video
Vertical9:161080 x 19201 video

Three orientations, not three crops of the same shot. A landscape video cropped to vertical leaves the head out of frame or the product on the edge, and that detail shows more in the small format than in the large one.

The documentation also recommends that at least one vertical run between 10 and 60 seconds, because that is the length that fits the Shorts inventory.

The consequence for the shoot

You shoot one session designed from the start for all three. In practice that means framing with room around the subject, not pushing the product to the edge, and avoiding captions in the corners, which disappear in at least one of the three versions.

A creator told this in advance solves it at no extra cost. A creator told afterwards has to be paid for a second session.

Where it actually runs

Here is why so many briefs are written wrong. Performance Max is not a YouTube campaign: it spreads the same asset group across several Google surfaces, and the video can appear in contexts very unlike each other.

That forces a piece that works with no prior context. It cannot assume the viewer arrived looking for the category, nor that they saw an earlier ad from the brand, nor that the sound is on.

What that changes in the first seconds

The product and the problem have to be clear without audio, with on-screen text, in the first seconds. This is not an aesthetic preference: it is the only way the piece works on every surface the system may choose to put it.

What the video has to show

The content changes compared with an organic piece, and not in tone but in what the viewer does not have.

The product in use from the first shot. No introduction, no presenting the speaker, no opening logo. The viewer did not ask to see this and does not grant the margin they grant in a feed they chose.

The reason to buy in plain words. What problem it solves, in a sentence understandable without knowing the category.

The price or the offer if they are an argument. Here yes, by contrast with an organic piece meant to last months, because a campaign is switched off and rebuilt when the offer changes.

The brand at the end and also small at the start. Many viewers never reach the end, and the one who does already knows who is speaking.

What does not work

The video opening on a rhetorical question, the one that takes eight seconds to show the product, and the one depending on a voice-over nobody will hear. All three work on surfaces the viewer chose and fail on surfaces where the ad interrupts.

The feed changes the equation

If the campaign is connected to a product feed, the system already holds images, prices and availability for each reference. The video then stops having to describe the catalogue and takes on another job: giving the proof a listing cannot give.

That means showing the product in a real context, in somebody's hands, solving something. The listing says what it is and what it costs; the video makes somebody believe it.

The consequence for how many videos you need

With a feed, a handful of generic category pieces performs better than one video per reference, which is unworkable on large catalogues. You produce per product family, not per SKU.

How the session is prepared

Three takes per message, not three edits. Each idea is filmed in all three orientations during the session, rather than shot landscape and reframed later.

One message per video. The system tests combinations; giving it five videos that say the same thing wastes the only lever the brand controls.

Different lengths. A short piece and a fuller one of the same message give the system material for different surfaces.

The clean file. No watermarks from other networks and no burnt-in subtitles if the piece is to be reused elsewhere.

What can be measured and what stays opaque

This is worth saying before somebody asks for the report that does not exist. Performance Max does not break performance down by placement with the granularity of a traditional campaign, so the question "how did the video do in each place?" has no complete answer.

What can be read is asset-level performance inside the group, which shows which pieces the system uses most and which it ignores. That is signal enough to decide what to film next time, even if it is not the breakdown one would like.

The honest reading

If the goal is understanding which creative works and why, this campaign is not the right instrument, and part of the budget should be reserved for a format that allows it. If the goal is selling with minimal management, the opacity is the price, and it is one many brands accept knowingly.

Sources

Checked on 8 September 2026. Where this guide and the official documentation disagree, the documentation prevails.