Performance Max AI Video Reframing: Opt Out by September 4

On September 4, 2026, Google starts generating video creative you never made, for placements you never picked, and running it under your brand name. Unless you opt out, that is the default.

The change is narrow on paper: Google Ads will use generative AI to extend existing Performance Max video ads into missing aspect ratios. If you uploaded a 16:9 horizontal cut and never produced a 9:16 vertical or a 1:1 square, Google's model will now build those versions for you and serve them across inventory your campaign previously couldn't reach.

Google's own framing is that the goal is to "maximize video ad reach while improving the viewing experience." Advertisers who don't want it can contact their account team or submit an opt-out form before September 4. Video settings remain editable in the account afterward.

It lands three days after a second default-on deadline, which is what makes this week worth a calendar block rather than a shrug.

What Actually Changes in Performance Max Video

Google has reframed video before. The existing enhancement crops and pans a horizontal video into vertical and square formats — mechanical, predictable, and easy to eyeball. What ships now is different in kind, not degree.

Per PPC News Feed's write-up, the system scans active Performance Max campaigns for missing ratios and uses generative AI to extend the frame. That means synthesizing pixels that were never shot: outpainting the top and bottom of a 16:9 clip to fill a 9:16 canvas, or inventing the edges of a square crop.

The practical differences:

• It applies to existing assets, retroactively. This isn't a setting on new uploads. It sweeps campaigns already running.
• The default is on. Taking no action is a choice to participate.
• The opt-out is an account-level request, not a campaign checkbox. Reporting describes contacting your account management team or filing Google's form — a heavier lift than a toggle, and one that doesn't segment by campaign or brand.
• Generated frames are model output, not your footage. Whatever the model invents in the extended region is what a customer sees next to your logo.

That last point is where most of the industry commentary has landed. Dataconomy noted the obvious risk: automatically generated versions may not suit every brand, and can introduce visual inconsistency or inaccurate product representation.

The Other September Deadline

September 1 is when Google begins automatically upgrading Search campaigns to AI Max — specifically those using Dynamic Search Ads, automatically created assets, or campaign-level broad match. New DSA creation ends across Google Ads platforms in the same window, with the full DSA sunset extended to February 2027 after advertiser feedback. Google says upgrades will mirror existing legacy settings, and advertisers keep brand and location controls, text guidelines, legacy URL controls, and one-click experiments.

We covered the mechanics when the migration was announced in DSA is becoming AI Max. The point here is the pattern: in one week, Google flips the default on both what your search campaigns target and what your video creative looks like. Two of the three legs of a campaign — targeting and creative — move to model output by default, with measurement the only leg you still fully own.

The Case for Opting Out Immediately

The strongest version of the argument against generative reframing:

You are liable for output you didn't produce. If the model outpaints a background that misrepresents a product, invents packaging detail, or renders a person's proportions oddly, the ad still carries your brand. Regulated advertisers — finance, health, alcohol, anything with mandatory disclosure placement — have a concrete compliance problem, because extending a frame can push a legal line out of the safe area or leave it floating in synthesized space.

You can't audit what you can't see. Performance Max asset reporting has improved, but there's no obvious workflow for reviewing every AI-generated ratio across every campaign before it serves. In practice, most teams will discover a bad generation from a screenshot someone sends them, not from a report.

The opt-out is deliberately high-friction. A form and an account team, on a two-and-a-half-week clock, in late August, when half the industry is out. That is not how you design a control you expect people to use.

Reach is not the metric. Filling vertical inventory with a degraded asset buys impressions, not incrementality. If the generated cut converts worse than the horizontal original, blended campaign metrics may still look fine while efficiency quietly erodes.

Why Blanket Opting Out Is the Wrong Move

All of that is fair. Here's where it breaks down.

The counterfactual isn't "my beautiful horizontal video runs everywhere." It's "I don't show up." Vertical placements are where the volume is — Shorts, in-feed, mobile interstitial. If your campaign has no 9:16 asset, Google isn't holding that inventory open for you. Something else fills it. Opting out doesn't protect your creative; it removes you from the auction where attention actually is.

Most advertisers were never going to produce the missing ratios. That's the uncomfortable truth behind the feature. Agencies with a real production budget cut native vertical. Everyone else uploads the one asset they have and lets Google's crop do the rest. For that majority, generative extension is competing against mechanical cropping — which already chops heads, buries text, and centers on nothing. A model that extends the frame is plausibly better than a crop that destroys it.

Blanket opt-out is a decision made without data. Nobody has performance evidence on this feature yet, because it hasn't run. Opting out of an untested change before you've seen a single generated frame is the same reflex that kept teams off responsive search ads and Advantage+ for a year longer than was rational. The disciplined move is to look at the output, not to preempt it.

And the risk isn't uniform. A SaaS demo video with a flat background outpaints cleanly. A hand-held product shot with a busy environment does not. Treating "generative reframing" as one binary decision across an entire account discards that distinction — which is exactly the distinction that determines whether this helps or hurts you.

What to Do Before September 4

1. Inventory your video assets by missing ratio

Pull every active Performance Max campaign and list which video assets have a 16:9 only, versus which already have native 9:16 and 1:1. Campaigns with complete ratio coverage are unaffected by this change — the system fills gaps. Knowing your exposure takes an afternoon and scopes every other decision below.

2. Sort the gap list by brand risk, not by spend

For each asset missing a ratio, ask a single question: what is in the region the model would have to invent? Solid or blurred backgrounds, studio product on seamless, motion graphics — low risk. Human subjects near the frame edge, packaging with legible copy, regulated disclosures, anything where the edges carry meaning — high risk. That list, not your top-spend list, is what justifies the opt-out request.

3. If you opt out, start the request this week

The path runs through your account team or Google's form, and September 4 is a hard date. Requests filed on September 3 through a channel that takes days to process are functionally not filed. Note also that the opt-out is account-level — if part of your account benefits and part doesn't, you're choosing for all of it.

4. Preempt the feature by shipping native vertical

This is the answer that survives whichever way the feature goes. An asset that already has a native 9:16 cut is not a candidate for generative extension, gets full vertical reach, and outperforms both crops and outpaints because it was framed for the canvas. If you're already running AI creative testing at scale, adding ratio variants to the test matrix costs almost nothing. The same asset set feeds Reels and TikTok, so the production spend amortizes across platforms rather than sitting in one channel.

5. Set an asset-level review cadence for September

If you let it run, check asset performance weekly through the month and screenshot generated ratios where you can surface them. You're looking for two things: generations that are visually wrong, and generated ratios that underperform the original badly enough to drag campaign efficiency. Judge that on incrementality rather than last-click ROAS — added vertical reach will inflate impression and click volume regardless of whether it's working.

6. Confirm your AI Max upgrade settings mirrored correctly

Separately, after September 1, verify that upgraded Search campaigns carried over their brand controls, location settings, and URL controls as Google says they will. "Mirrors existing legacy settings" is a claim worth checking against your own account rather than trusting. Google's AI Brief natural-language controls are the lever for steering the upgraded campaigns once they're live.

The Pattern Worth Naming

Neither of these announcements is scandalous on its own. Reframing video for missing placements is a sensible thing for an ad platform to automate, and AI Max is a better product than the DSA it replaces. What's changed is the mechanism: both arrive as defaults with an opt-out window, not as features you evaluate and enable.

That is now the standard shape of platform change across the industry — the same shape as advanced matching switching on by default in ChatGPT Ads, and the same shape as Advantage+ absorbing manual campaign setup. The operational skill that matters isn't having an opinion about AI creative. It's having an inventory of what's about to change by default, a segment of where the risk actually concentrates, and a calendar entry before the window closes.

Most teams will miss September 4 not because they decided to participate, but because nobody was watching. The advertisers who do well here won't be the ones who opted out or opted in. They'll be the ones who knew which of their assets were exposed, shipped native vertical for the ones that mattered, and let the model handle the rest.

Sources: Search Engine Land — generative AI resizing for Performance Max video, PPC News Feed — PMax AI-generated aspect ratios, Dataconomy — generative AI optimization in video ads, Google Ads & Commerce Blog — DSA upgrade to AI Max