Performance Max Video Ads for Ecommerce: The 2026 Playbook

Performance Max is the default campaign type for most ecommerce accounts in 2026 — and that is exactly why most of them look identical. Google's system assembles the final ad from the assets you feed it, and when you don't feed it video, it generates one for you from your images and text. That auto-generated video is the single biggest differentiator between accounts that win the auction and accounts that donate to it: it is generic by construction, because it has to work for every advertiser who didn't bring their own creative.

The same story plays out on the demand side. Demand Gen, Google's feed-first campaign type for YouTube, Discover and Gmail, reaches up to 2.9 billion people — but only if you hand it creatives that stop a thumb mid-scroll. Video is the highest-impact asset in both campaign types, and it is also the most expensive one to produce at volume. That tension — "Google wants video, producing video is slow and expensive" — is the exact problem this playbook solves.

This guide covers the verified 2026 asset specs for Performance Max and Demand Gen, how AI-generated UGC video fits into asset groups and rotation, hook patterns per ecommerce niche, the PMax vs Demand Gen decision, and the real cost math of building a video asset library that keeps up with Google's appetite for fresh creative.

Why PMax and Demand Gen are different from every other ad surface

Both campaign types are machine-learning driven, but they punish creative laziness differently:

• Performance Max is an asset-assembly system. You supply the parts (images, videos, headlines, logos, product feeds), and Google AI assembles and tests combinations across Search, Shopping, Display, YouTube, Discover, Gmail and Maps. The campaign has no ad groups in the classic sense — it has asset groups, which are thematic buckets. Google's algorithm does the rotation, which means your job is not to pick the winning creative but to give the system enough good raw material to find it.
• Demand Gen is a discovery-feed system. It runs video and image ads across YouTube (including Shorts), Discover and Gmail, with more explicit targeting controls (custom segments, lookalikes, demographics) than PMax. Creative here competes against organic content, not against other ads in a search result. The bar for "native feel" is higher, and the payoff for video that looks like content rather than an ad is correspondingly larger.

The structural insight most ecommerce teams miss: both systems prefer you to provide video, and both will happily serve you with mediocre video you didn't provide. PMax auto-generates videos when an asset group has none. Google has been explicit about this since the format launched, yet most accounts still run on auto-generated clips assembled from product images — then blame the campaign type when CPA drifts.

The 2026 video spec sheets (verified against Google Ads Help)

Performance Max video assets

Verified against Google's official video asset documentation on the day this guide was written:

Parameter	Requirement	Recommended
Format	MPG (MPEG-2 or MPEG-4)	MP4, H.264
Resolution	HD	1080p: 1920×1080 (16:9), 1080×1920 (9:16), 1080×1080 (1:1)
Duration	10 seconds or more	15–30s action, under 20s for reach
Count per asset group	Up to 15 videos	1 of each orientation (min 1, max 15)
Shorts eligibility	—	At least one vertical 9:16 video between 10 and 60 seconds
File size	—	No practical ceiling per current Help docs (256 GB listed)
Auto-generation	If you add no video, Google creates videos from your other assets	Bring your own; auto-generated clips are generic

Google AI may also adapt your uploaded videos — cropping a horizontal spot to square or vertical, or trimming a long video into shorter versions — and serve those automatically if they pass quality review. That is useful reach, but it means your original creative needs clean safe zones: keep the hook and the product in the center of the frame in every orientation.

Demand Gen video assets

Verified against Google's Demand Gen asset specifications:

Parameter	Requirement	Recommended
Format	MPG (MPEG-2 or MPEG-4)	MP4, H.264
Duration	Minimum 5 seconds	15–30s; under 10s won't serve on YouTube In-stream
Orientations	1:1, 4:5, 9:16	9:16 1080×1920 for Shorts; 1:1 1080×1080; 4:5 1080×1350
File size	Max 256 GB	Keep under 100 MB for fast upload
Videos per ad group	Up to 5	3+ per orientation ("rule of three" from Google's Creative Excellence guide)
Companion assets	Headline up to 40 chars, description up to 90, business name up to 25, CTA up to 10	Fill all headline and description slots

Google's Creative Excellence Guide for Demand Gen recommends at least three images or videos in each aspect ratio per ad group, and emphasizes native-feeling creative — for Shorts, that means vertical video that is "relatable, authentic, and connects with your audience."

Anatomy of a PMax video asset that earns its place

A video asset in PMax is rarely watched in isolation — Google tests it in different combinations, crops and placements. So the anatomy is about robustness across contexts:

1. Hook — 0–2s. Because Google may crop your video to square or vertical, the hook has to survive a center-crop. Put the speaker, the product, and the promise in the middle third of the frame. Text overlays near the edges will get cut.
2. Product reveal — 2–10s. Show the product in motion against a clean background. For AI-generated assets, this is where product-page imagery becomes a hero visual — a demo, a texture close-up, a before/after. Avoid busy backgrounds; they compress badly in smaller placements.
3. Proof — 10–20s. One social-proof beat: a star rating, a review quote, a customer count. In feed placements, stacked proof reads as clutter; pick the single strongest signal.
4. CTA — last 5s. State the action verbally and visually. PMax overlays its own CTA buttons, so the verbal ask and the on-screen button should agree. Don't put a hard end-card with text in the bottom third — it collides with the overlay.

Because PMax serves the same asset as a skippable in-stream ad, a Shorts tile, and a Discover image-with-video unit, think of the asset as a system, not a spot: center-weighted composition, clear audio, on-screen captions for sound-off placements.

Per-niche hook patterns for PMax and Demand Gen

The same asset pipeline produces different winners per category. These are the patterns that consistently hold up across Google's inventory:

Skincare and beauty

Hook: problem-first ("Why your skin still looks tired by 3pm"). First frame should be the texture or the concern, not the bottle — the bottle comes at the reveal. This category over-indexes on sound-off placements, so burned-in captions are non-negotiable. Compliance is the hard ceiling: no medical claims, no "cures", no before/after that implies treatment. Google's health claims policy applies to every placement, and AI-generated creatives get the same review as human-shot ones.

Supplements and wellness

Hook: outcome-first ("30 days in, my afternoon crashes are gone"). PMax and Demand Gen shoppers in this category convert on identity and routine more than ingredient lists. Keep claims to "supports" / "may help with" / "many customers report" — treatment, prevention, and diagnosis claims get accounts flagged regardless of whether the video is AI-generated or human-shot.

Fashion and apparel

Hook: identity-first or transformation ("Here's what actually fits a 5'4 frame"). The reveal is the fit on a body, not the garment on a hanger. Movement sells in feed placements; a static product shot reads as an ad. AI pipelines with motion control — reference-to-video from the brand's lookbook — beat a static hero image every time, and the same asset can feed both PMax and Demand Gen.

Home and kitchen

Hook: transformation ("This 12-second install changed my kitchen"). First frame shows the result, not the packaging. Home content overperforms in Discover and Shorts because the placement skews toward lifestyle and DIY — lean into the before/after and the "renters can do this too" angle.

Asset-group strategy: where AI creative fits

The most common PMax setup mistake is one asset group per product with three images and no video. The structure that works:

Strategy	Why it works
One asset group per theme (product category, audience, or margin tier)	Lets Google match the right assets to the right queries
1 video per orientation per asset group (16:9 + 1:1 + 9:16)	Maximizes placement eligibility; Google recommends 1 of each
3+ images per orientation alongside video	The "rule of three" gives the algorithm enough combinations to test
Swap in fresh video every 2–3 weeks	Google's auction rewards new creative; stale assets plateau
Keep the offer and landing page stable while testing creative	You isolate the variable — creative, not the offer

For Demand Gen, the same asset discipline applies at the ad-group level, with one addition: because Demand Gen gives you more targeting control, you can split creative by audience (new customer creative vs repeat customer creative) without guessing — the format tells you which assets each segment responds to.

PMax vs Demand Gen: which campaign gets which video

Decision	Performance Max	Demand Gen
Primary goal	Conversions across the full funnel	Consideration, discovery, and conversion from feeds
Inventory	Search, Shopping, Display, YouTube, Discover, Gmail, Maps	YouTube (Shorts, In-stream, Home, Search, Watch Next), Discover, Gmail, GDN, Video Partners
Creative control	Low — Google assembles and tests combinations	Medium — you control the creative per ad group
Targeting	Automated, goal-based	Custom segments, lookalikes, demographics, exclusions
Video role	Strong differentiator; optional because Google auto-generates	Strong differentiator; feeds are video-first
Learning period	Expect a learning period after launch or significant edits (third-party sources commonly cite 2–6 weeks; Google's own guidance is that duration depends on conversion volume)	Similar — budget stability shortens it

A common winning structure: run a PMax campaign for capture and a Demand Gen campaign for discovery, and feed both from the same video asset library. The same 9:16 AI UGC creative that wins in Shorts placements can be uploaded to both campaign types — Google will crop and trim it to fit.

Cost math: AI pipeline vs UGC creator vs in-house

The budget question for a PMax video library (say, 12 assets — 3 hooks × 2 products × 2 orientations — refreshed monthly):

Channel	Cost per creative	Turnaround	12-asset library
UGC creator (mid-tier)	$400–$800	7–14 days	$4,800–$9,600
In-house studio shoot	$1,500–$5,000	2–4 weeks	$18k–$60k
Agency-managed	$1,000–$2,500	4–6 weeks	$12k–$30k
AI pipeline (Prizmad)	≈$2–8 per render	under 10 min	under $120, same day

The AI line is a conservative estimate based on current plan pricing (Launch $39/mo for 30 tokens, roughly 6–7 renders; Starter $99/mo for 80 tokens; Pro $249/mo for 350 tokens, per Prizmad pricing). The cost difference moves the bottleneck from budget to script-writing — which is the right bottleneck, because script variety is what Google's auction actually rewards. An account that rotates 12 fresh video assets a month consistently out-tests an account that runs 3 for a quarter.

Compliance — the notes that trip up AI creatives

PMax and Demand Gen follow the same Google Ads policies as every other placement, plus the synthetic-content rules:

1. AI disclosure. Google requires ads that contain synthetic or altered content — realistic people doing or saying things they didn't — to be labeled. Ad accounts need to self-identify in Google Ads, and YouTube's "altered or synthetic content" disclosure applies to uploads. Apply the label; the algorithm and the reviewer both check.
2. Avatar likeness. Don't generate avatars that look like real identifiable people, competitors' founders, or your customers. This is a platform-policy issue and a right-of-publicity issue in the US and EU.
3. Health and money claims. The policy does not change because the creative is AI-generated. Skincare and supplements are the two categories that get accounts flagged fastest.
4. Misrepresentation. PMax auto-assembles ads from your assets — make sure every asset's claims match the landing page, because Google's misrepresentation policy applies to the assembled ad, not just the individual assets.

Frequently asked questions

---

Ready to feed Google the video it wants? Paste your product URL into the Prizmad generator and get a 9:16, Square-ready mp4 in about five minutes — upload the same file to PMax and Demand Gen asset groups, and let the auction do the rest.