Performance Max Assets: Specs, Quality and What to Feed It
Every Performance Max asset spec in one place - headlines, images, video - plus what Ad Strength actually rewards and when to ignore its advice.
Assets are the last creative control you hold in Performance Max. Bidding is automated, placement is automated, assembly is automated - the one thing the machine cannot do without you is the raw material, and everything it shows on every surface is built from what you put in the asset group. That makes the asset upload the highest-leverage half hour in the whole campaign, and also the place where I see the most careless work: one squeezed image in three crops, fifteen headlines that say the same thing, and a video slot left for Google to fill.
This post is the reference I wish the Google Ads interface showed on one screen: the full spec sheet with minimums that actually matter, what Ad Strength rewards and how much to trust its promises, and the video decision that quietly decides how your brand looks on YouTube. It assumes the campaign structure is already right - that part is the setup walkthrough - and it feeds directly into the diagnostic post when a live campaign underperforms.
The spec sheet
Everything below is from Google’s asset requirements page, current as I write this. Ranges are minimum required to maximum allowed.
Text assets:
| Asset | Count | Limit |
|---|---|---|
| Headlines | 3-15 | 30 characters, at least one of 15 or fewer |
| Long headlines | 1-5 | 90 characters |
| Descriptions | 2-5 | 90 characters |
| Business name | 1 | 25 characters |
Image assets (.jpg or .png, 5,120 KB max each):
| Asset | Count | Size |
|---|---|---|
| Horizontal 1.91:1 | 4-20 | 1200 x 628 recommended, 600 x 314 minimum |
| Square 1:1 | 4-20 | 1200 x 1200 recommended, 300 x 300 minimum |
| Vertical 4:5 | 2-20 | 960 x 1200 recommended, 480 x 600 minimum |
| Square logo 1:1 | 1-5 | 1200 x 1200 recommended, 128 x 128 minimum |
| Horizontal logo 4:1 | 0-5 | 1200 x 300 recommended, 512 x 128 minimum |
Video: up to 15 per orientation (horizontal 16:9, square 1:1, vertical 9:16), each 10 seconds or longer, 1080p recommended. Video is technically optional. Treating it as optional is the mistake the whole next section is about.
Two spec details do real work. The one headline under 15 characters is not bureaucracy - short headlines are what fit the smallest placements, and without one the machine crops or skips your best lines. And the minimums on images (4 horizontal, 4 square, 2 vertical) are floors, not targets: the floor gets you served, and nothing more.
What quality means to the machine, and how much to trust it
Google scores your asset group with Ad Strength, and its Performance Max Ad Strength page is unusually explicit about what an “Excellent” rating takes: all 15 headline and 5 description slots used with real variety, a full spread of asset types, at least 6 sitelinks, and your own uploaded videos in all three orientations rather than the auto-generated kind.
The incentive Google attaches: advertisers who improve Performance Max Ad Strength to Excellent see, on average, 6% more conversions. I read that number the same way I read the 15% claim on responsive search ads: the footnote first. In Google’s own 2024 features recap the claim is footnoted to internal global data comparing asset groups before and after an Ad Strength change, Q1 to Q3 2023. Before-and-after comparisons on advertisers who chose to improve their assets measure motivated advertisers as much as they measure assets. A 6% average from that design is a reason to take assets seriously, and not a promise your account collects by filling slots.
The counterweight comes from outside Google. Optmyzr’s study of 24,702 Performance Max campaigns found that 57% of advertisers ran full creative sets - and that text-only asset groups often outperformed the “complete” ones. My reading, which matches what I see in accounts: slot-filling is not the lever. A complete asset group of mediocre, repetitive creative loses to a lean one with genuinely distinct messages. Ad Strength measures completeness because completeness is what a scoring system can count. Distinctness is what actually feeds the combination machine, and only you can count that.
So my rule: build toward Excellent, but never by padding. Fifteen headlines that earn their slot beat fifteen that fill it, and if I only have eight genuinely different angles, I ship eight and add more when I have something new to say. The ad copy prompts exist for exactly this job - a full asset group is a real volume of copywriting, and forcing one distinct angle per headline is the constraint that makes the volume worth having.
Video: the slot Google fills if you don’t
Here is the behavior every advertiser should know before launch, in Google’s own words: if you don’t add a video to your Performance Max asset group, one or more videos may be auto-generated from the assets in your asset group. Your images and headlines get assembled into a video, with templates, colors, fonts and music the system picks to match your brand - and lately, generative AI that animates static images into 10-second clips. That output then represents your brand on YouTube, chosen by a system whose goal is to have something to serve.
Uploading your own videos switches auto-generation off - Google states the disable behavior plainly on the same page. That single fact is why my advice in the setup post is to bring a video, even a modest one, before launch. A plain honest 15-second product video you control beats an automated slideshow you have never seen. And note the adjacent behavior: Google may also flip your uploaded horizontal video into square and vertical versions so it can serve on YouTube Shorts. If your horizontal video has text near the edges, check what a 9:16 crop does to it, because the machine will make that crop whether you checked or not.
If you genuinely cannot produce video, my order of preference: a simple self-made video from screen recordings or product shots, then Google’s own creation tools where you at least pick the inputs and review the output, and only then the silent default of letting auto-generation run. Whatever you choose, watch what actually serves. The video report shows you what YouTube viewers are seeing with your logo on it, and I have rarely met an advertiser who had looked.
Images: native crops, not one file squeezed three ways
The image spec table above hides the real instruction: the three aspect ratios are three different compositions, not one image exported at three sizes. A horizontal 1.91:1 lifestyle shot center-cropped to 1:1 loses its subject, and stretched to 4:5 it loses its credibility. I supply each ratio composed for that ratio - product centered for square, headroom for vertical - and I keep text off the images entirely, because the placements overlay your headlines anyway and image text at 300 x 300 renders as noise.
Variety matters here for the same reason it matters in headlines. Four near-identical pack shots satisfy the minimum and give the machine nothing to test. My working set is product-in-use, product-alone, outcome, and something with a human in it, per theme - which also keeps the asset group honest with the one-theme-per-group rule, because images from different product stories jammed into one group produce combinations you would never approve by hand.
Prune with the asset report, on a schedule
Asset work does not end at launch. Google now reports per-asset impressions, clicks, cost and conversions in Performance Max, which turns pruning from guesswork into maintenance. My cadence is monthly: sort assets by impressions, and for anything carrying real volume with performance labeled “Low” or visibly worse than its peers, replace it with a new angle rather than leaving it to drag the mix. Swap assets in batches, not in a weekly trickle - the same learning-period logic that governs every other change to this campaign type applies to creative, and a constant drip of asset edits keeps the machine perpetually re-learning.
One caution from the same Optmyzr data: a single strong headline dominating every combination usually means the other fourteen were filler. The fix is not to delete the winner. It is to write challengers that are actually different, which is the one part of this format no report automates.
FAQ
How many assets should I upload at launch? Enough that every one earns its place - for me that is usually 10-12 headlines, all 5 descriptions, 8-12 images across the ratios, and at least one real video per orientation. The maximums are ceilings for mature asset groups, not launch targets.
Does Ad Strength affect delivery directly? Google presents it as a diagnostic, and its own guidance is that improving it correlates with more conversions - the footnoted 6% above. In practice I treat a Poor rating as a real warning (the machine has too little to assemble) and the gap between Good and Excellent as advisory. I will not pad an asset group with filler to change a label.
Can I just let Google’s AI generate the images and video? You can, and the automatically created assets toggles in settings govern the text side of that - the setup post covers why I launch with them off. Generated creative is a reasonable way to explore volume once a campaign is stable. It is a bad way to introduce your brand to placements you have never reviewed.
Are the auto-generated videos really that bad? They have improved - the image-to-video generation is better than the old slideshow templates. The problem is not always quality. It is accountability: an asset representing your brand that nobody on your team has watched. Review what serves, then decide.
This post is part of the paid media guides. The campaign around these assets is built in the Performance Max setup walkthrough, protected by the guardrails checklist, and debugged in the troubleshooting guide. The copywriting method for filling fifteen slots with real variety is in the ad copy prompts, and the wider practice starts at the complete guide to AI marketing.