Generative AI Display Advertising Is Now the Default

Most people searching for "generative AI display advertising" are asking whether they should use AI to make display ads. In 2026 that framing is already out of date. Google has answered the question on your behalf, twice: once by retiring the campaign type, and once by turning the generation on by default.

Here is what actually changed, what is switched on in your account right now, and which parts of it you should turn off.

Display campaigns are being retired into Demand Gen

Google announced in May 2026 that standalone Display campaigns are being consolidated into Demand Gen. The timeline it has published:

The important clarification, because the headlines got this wrong: the Google Display Network is not going away. GDN inventory continues to serve. It just serves from inside a Demand Gen campaign, alongside YouTube, Discover, Gmail and Google Maps.

So this is not an inventory change. It is a control-surface change — the campaign type through which you buy that inventory is being replaced by one that is designed around asset supply and audience signals rather than around placements and bids.

The part nobody mentions: five defaults are already on

Demand Gen ships eight creative enhancement and generative AI tools. Five of them are enabled by default on new campaigns. Verified against Google Ads Help on 2026-08-12:

Read that table as an operational fact rather than a feature list. A Demand Gen campaign built quickly — the way campaigns actually get built — is generating video from your stills, cropping your images, overlaying text you did not write, and screenshotting your landing page into the ad unit, all without an explicit decision.

That is not necessarily wrong. Some of it is genuinely useful; producing vertical cuts of a horizontal video is work nobody enjoys. But it is a materially different situation from "should we try generative AI for display", and it deserves an explicit review rather than an implicit yes.

There is no master switch

Each default is its own checkbox in a different part of the interface:

• Video enhancements → ad settings → uncheck "Get vertical and square versions of your videos" and "Get shorter versions of your videos"
• Auto-generated video → asset optimization → uncheck the Videos checkbox
• Image enhancements → asset optimization → uncheck Adaptive layouts and Animated images
• Pre-generated videos → deselect the suggested videos before publishing, or decline them on the campaign review page
• Landing page preview → Final URL card → uncheck "Show a screenshot of your landing page in your ads"

Five toggles in four places. This is worth a written checklist for whoever builds your campaigns, because "we turned the AI stuff off" is not a thing anyone can do in one action.

What the generation is actually bounded by

The published asset requirements tell you where the quality ceiling sits:

• Pre-generated videos: minimum 1 logo, 2 images, 2 text items (≥1 headline, ≥1 description), 1 business name.
• Auto-generated video ads: your text and images, plus optional brand guidelines for colour and font.
• Image enhancements: high-quality, sharp images, at least one headline, and compliance with editorial advertising policies.
• Generated images: Google Ads interface must be set to English; up to 20 image selections across aspect ratios.

Two practical consequences.

First, the output is a function of the seed assets. Feeding two mediocre product photos and expecting a good auto-generated video is not a fair test of the feature; it is a test of your asset library. Teams that report the generation being poor very often uploaded the minimum.

Second, the English-only requirement on generated images is a real constraint for non-English teams, and it is the kind of limitation that does not appear in any summary of the feature.

How to run the migration deliberately

The choice is not whether to migrate — the dates decide that. The choice is whether you migrate on your terms or inherit whatever the automatic migration produces. A sequence that produces a defensible outcome:

1. Export your Display baseline before you touch anything. Placement-level performance especially, because your exclusion lists are built on it and the post-migration reporting surface is organised around Demand Gen's structure, not Display's.
2. Audit your asset library against the minimums above. If you cannot supply a logo, two strong images, a headline and a description per campaign, fix that first — everything downstream is bounded by it.
3. Migrate one campaign, not the account. Preferably a mid-volume one where a wobble is survivable.
4. Decide each of the five defaults explicitly on that campaign and write down what you chose. This is your control setup.
5. Hold the campaign type constant when you test creative. This is the single most important methodological point below.
6. Re-baseline before you judge anything. The channel mix changed. Your CPA is not comparable to the Display-era number in any meaningful way.

The measurement trap

Here is the mistake that will be made across the industry over the next year, and it is worth naming precisely.

You migrate a Display campaign to Demand Gen. Results change. You attribute the change to generative creative — because that is what you were thinking about.

But migrating changed at least four things simultaneously: the inventory mix (GDN plus YouTube, Discover, Gmail, Maps), the audience expansion behaviour, the bidding surface, and the creative. A before-and-after comparison cannot separate them. Whatever conclusion you draw will be about the variable that was on your mind, not the variable that moved the number.

If you actually want to know whether the generated creative helps:

• Hold the campaign type constant. Migrate first, re-baseline, and only then toggle asset-optimization settings.
• Test one enhancement at a time. Auto-generated video and image enhancements are different mechanisms with different failure modes.
• Give it two full conversion cycles. A 14-day test on a 7-day lag is one cycle and tells you nothing.
• Look at the asset-level report, not just campaign totals. The question is whether generated assets outperform your uploaded ones, and that is visible at asset level or not at all.

Google has published no controlled figure for the creative-generation features specifically. Treat any vendor number that does not separate the creative effect from the campaign-type effect as unusable — including numbers you generate yourself.

What this means for creative production

Two structural implications worth planning around, beyond the mechanics.

The unit of creative work is shifting from "the ad" to "the asset set." When five features are generating variations from what you upload, your job stops being "make the display ad" and becomes "supply a well-formed asset set and constrain what gets made from it." That is a different brief, a different review process, and — for most teams — a different relationship with the design function.

Brand governance moves from approval to constraint. You cannot approve every generated permutation; there are too many and they are assembled at serve time. What you can do is control the inputs (asset quality, brand guidelines for colour and font) and the toggles. If your brand review process assumes a human sees every asset before it serves, that assumption no longer holds, and pretending otherwise is how off-brand crops reach production.

Where to go next

• AI Max vs PMax — the same "which controls does this remove" question, on the Search side
• Best AI media buying tools, ranked — where third-party tooling sits once the platforms generate creative themselves
• Best AI tools for Facebook & Instagram ad creatives — the Meta-side equivalent of this asset-supply problem
• Google Ads MCP: the complete guide — reading the account programmatically, which is how you audit five toggles across many campaigns

The short version

"Generative AI display advertising" is not a strategy you are choosing in 2026. It is the default configuration of the campaign type that Display is being folded into, on a timeline that ends with automatic migration in 2027.

The useful work is not deciding whether to adopt it. It is auditing the five defaults, supplying asset sets good enough to be worth generating from, and building a test that holds the campaign type constant so you can find out which of the generation actually earns its place.