Google on October 1, 2026 published an episode of its Ads Decoded series in which Alejandro Osio, a product manager for Demand Gen and YouTube performance ads, described a $100 daily budget as the minimum best-practice level for a Demand Gen campaign, declined to set any interval for replacing fatigued assets, and said Gemini Omni had reached Google Ads for the first time through Asset Studio.

In Short

Google posted a 20-minute video in which one of its product managers explained how advertisers can make, test and replace the pictures and videos in their Demand Gen ads on YouTube. It matters to anyone buying those ads because he described $100 a day as the bare minimum for such a campaign, which is 20 times the $5 that Google's systems actually require. There is no set schedule for swapping out tired ads, so the guidance is to watch how each one performs over time and keep the ones that are still pulling their weight.

An episode without performance figures

The episode pairs Ginny Marvin, Google's Ads Product Liaison, with Osio, a product manager for Demand Gen and YouTube performance ads. Titled "Demand Gen campaign creative, explained: asset variety, testing, and fatigue," it runs 19 minutes and 47 seconds on the Google Ads YouTube channel, which lists 919,000 subscribers; when captured for the material reviewed here, the page showed 3,913 views and 35 comments. A companion post on Google's Ads & Commerce Blog, published the same day, summarised three takeaways. Its image file is labelled S2E5, suggesting a second season after Google closed the first with a backstage episode recorded at Google Marketing Live in May.

The source material contains one discrepancy. The YouTube description and Google's blog post spell the guest's surname Osio, while the machine-generated transcript renders it "Ocio." This article follows Google's published spelling. Because the same transcript contains other plain errors, such as "non-feeded" where the context indicates non-feed assets, direct quotation below is limited to passages where the wording is unambiguous.

Across nearly 20 minutes, Osio cited no conversion uplift, no return on ad spend figure and no internal experiment result - the numbers Google's Demand Gen communications usually lead with. The one hard figure here is a budget.

Coverage before anything else

Osio opened by calling creative "one of, if not the most critical lever to drive strong results for your business." His first rule was coverage: images and videos across aspect ratios, and an ad strength rating of at least Good, preferably Excellent, on every Demand Gen ad. Marvin stressed that ad strength in Demand Gen works as it does in Search, as a diagnostic of asset variety; Osio agreed, calling it an indicator of whether an ad is "covering the breadth of inventory" the campaign type can reach.

That matches Google's May 6 Ads Decoded episode, which explained that ad strength uses cosine similarity to measure asset diversity rather than performance. Independent data points the same way. A 2024 Optmyzr analysis of more than 1 million ads across 22,000 accounts found no strong correlation between Ad Strength and performance, with ads rated Average often beating those rated Excellent on cost per acquisition. The ratios themselves are fixed by the format's specifications: 1.91:1 landscape, 1:1 square and 4:5 portrait for images, with landscape, square and vertical video.

On starting material, Osio was blunt. Social assets can be brought over directly - "lift and shift," in his words - without waiting for a bespoke YouTube set. Google's blog post adds a condition the episode skipped: existing creative has to meet Google's asset specifications and guidelines.

The $100 figure and the numbers around it

The budget reference arrived in passing, in an argument that volume is not testing. Osio asked listeners to picture a Demand Gen campaign with "a hundred dollars daily budget, which is bare minimum best practice," and then to picture tens of thousands of non-feed images and videos added to it. In that case, he said, "the system's not feasibly gonna be able to explore all of those assets," because "its primary objective is to deliver performance against your set goals." Feed-driven retail campaigns were excluded from the example; according to Osio, Google's systems are built for much larger volumes when a product feed supplies the creative.

The more assets a campaign holds, he said, "the more time and budget you likely need to reach a clear conclusion." Marvin observed that this departs from Google's Search recommendations.

The figure sits awkwardly beside Google's other published numbers. Since April 1, 2026, the Google Ads API has enforced a minimum daily budget of $5 for every Demand Gen campaign, a rule set out on February 27. Osio's level is 20 times that. A Fospha programme run with Google across 25 retail ecommerce brands in late 2025 referenced a recommended minimum of approximately $130 per day per campaign, and advised starting with at least nine assets - three horizontal videos, three vertical and three square. When Google moved YouTube video action buyers in Display & Video 360 onto Demand Gen in 2025, it recommended a daily budget greater than 20 times the expected CPA for conversion-based bidding. At the Display & Video 360 integration in 2024, the multiple was 15 times the target CPA.

The numbers measure different things - an enforcement threshold, a podcast characterisation, rules of thumb tied to bid strategy - but they do not converge. Run backwards, the 20-times rule implies that $100 a day supports a target CPA of no more than $5, and over a 30-day month that budget comes to roughly $3,000 before asset count enters the calculation. What does a "bare minimum" mean when Google's own API accepts a twentieth of it?

Fatigue without a number

Asked how real asset fatigue is on YouTube, Osio said Google fields many questions about refresh frequency but has no definitive interval to give. "When we look at the data, we find that the answer is really that it depends," he said, naming the type of asset, the inventory it serves on and audience behaviour as variables. "I would actually recommend against setting a blanket rule for your business on how frequently to replace assets."

His alternative runs through asset reporting. An asset in marked decline against its performance when fresh is a stronger candidate for replacement; one still serving strongly is better left running, with new material layered in where there is room. Active removal makes sense, according to Osio, only when an ad is already at capacity and a specific asset has declined significantly from its starting point. Capacity is not small: a Demand Gen ad accepts up to 20 images.

That cuts against common practice. A survey of 111 US marketers by Perion and EMARKETER found 59.5% relying on planned rotation schedules to trigger creative refreshes, with only 11.7% using AI-driven performance signals. The same research found nearly three-quarters of marketers refresh creative only after performance has declined - roughly the point at which Osio's guidance makes replacement defensible.

One measurement complication went unmentioned. Judging decline against the moment an asset was added depends on conversions arriving in time to be counted, and at Google Marketing Live in May, Google said only 40 percent of Demand Gen conversions occur in the first 30 days. An asset that looks weak in its first month may simply be early.

Two meanings of asset testing

Osio separated two uses of the phrase, noting that "different advertisers use the term to refer to different things." For refreshing an existing campaign, he favoured gradual change: top-performing, highest-traffic assets stay live while new ones are added, which "helps keep performance stable while the system ramps up on any new inputs you've added." For bulk testing, he put "new asset quality over extreme quantities."

Neither answer mentioned Google's formal testing tools for the campaign type. The November 2025 Demand Gen release added asset uplift A/B experiments for text, images and video, alongside custom experiments supporting up to 10 arms. In May 2026, Google added 1-Click Creative Testing, which evaluates asset variants against each other on performance goals. The episode's guidance concerns rotation inside a live campaign, not controlled experiments.

What asset optimizations now cover

Osio grouped the enhancement features by asset type. For video, Google AI generates new videos, shortens existing ones and resizes them. Images get three treatments: adaptive layout, which takes a square image and "extrudes it out to fit natively on vertical"; landing page previews, which give viewers a look at the destination page; and animated images, still in beta, which add motion effects to images on display inventory. Text gets generated variants.

Video enhancements were part of the November 2025 release. The animated images beta is the episode's most specific status detail, and display inventory now belongs to Demand Gen: Google is retiring standalone Display campaigns and folding the Google Display Network into the campaign type, with the transition expected to complete in 2027.

The default state of these features is not settled by the record. PPC Land's coverage of the November 2025 release described creative enhancements as opt-in; its coverage of the August 2026 Demand Gen Drop described video and image enhancements as enabled by default for new campaigns. Osio's checklist for existing advertisers included an instruction to "toggle asset optimizations on," consistent with at least older campaigns not running them.

Asked whether advertisers with large creative budgets have reason to use video enhancements, Osio said they are "absolutely worth testing regardless of your creative input." Marvin stressed that outputs remain "still grounded in your original assets," and Osio said the enhancements use an advertiser's own creative as their base. Google's May 6 episode had cited an internal experiment in which Demand Gen ads with video enhancements saw more than a 16% increase in conversion value at the targeted ROI on YouTube - a figure from Google's own data.

Review before anything runs

On quality assurance, Osio pointed to two controls. The first is shareable previews, which he said Google recently released: a link showing how ads will render with the assets and inputs supplied, for review by creative production teams and other approvers before spend begins. The second is asset reporting, where variants produced by Google AI can be reviewed and paused.

Performance Max gained a preview link usable without account access in early 2024, and version 17.1 of the Google Ads API added shareable preview URLs with expiration dates later that year. Asset Studio's September 2025 rollout included shareable links for creative review. Microsoft Advertising's equivalent for its Performance Max campaigns produces a public, view-only link that expires after 30 days. Osio gave no expiry or arrival date for the Demand Gen links.

Gemini Omni's first appearance in Google Ads

The most newsworthy section was also the least fresh. Marvin said multimodal video creation in Asset Studio is generally available, taking advertisers from storyboarding to vertical and horizontal assets in a single workflow. That status was first published on August 27, when Google's twelfth monthly Demand Gen Drop moved the feature to general availability - in a post that also carried a 30% average conversion uplift drawn entirely from Google's internal experiments. The capability had been set out on May 20, when Asset Studio gained Gemini Omni with a summer rollout in English.

Osio described the feature as "actually the first time Gemini Omni has made it into" Google Ads, with an advertiser's existing creative and URL as inputs. The model had already reached consumer products: YouTube brought Gemini Omni remixing to Shorts at Google I/O on May 19, initially excluding the EU and UK, and Google Vids added Gemini Omni editing for paid subscribers on July 16.

His own test case was domestic: he had used the tool the day before recording to help his cousin set up her first Demand Gen campaigns. The flow is prompt-based, output quality tracks input quality, and the storyboard can be edited "scene by scene." He compared the prompting to giving feedback to a creative production team.

The intended users include advertisers with good photographs but no video production, and B2B businesses weak on visual assets. Google built these tools, Osio said, "to reduce the costs and burden" of producing the variety of creative Demand Gen can serve. Labelling did not come up, although Google's process for marking existing assets as created or edited with AI runs through Asset Studio's asset library.

Before a first campaign: data, seeds, variety

Asked what a new advertiser needs before a first Demand Gen campaign, Osio listed three items. The first was data strength, which he broke into three parts: Google tag gateway set up, a CRM linked through Data Manager, and enhanced conversions switched on. Together, he said, these "pass high quality signals directly to Google AI."

Google has steadily pushed advertisers toward each component. Customer Match uploads through the Google Ads API stopped on April 1, 2026, leaving the Data Manager API as the route. Enhanced conversions for web and for leads merged into a single on/off setting in June 2026. On September 10, Google added a Data Strength Uplift Metric and stated that tag gateway adopters see an average 14% conversion uplift, and over 20% for Demand Gen; PPC Land has noted that the footnote behind the 14% figure scopes it to Google's internal finance-vertical data.

The second item was targeting: first-party customer data uploaded as an audience seed list, with optimized targeting switched on "to capture high intent, net new users." Marvin closed the episode by explaining that optimized targeting uses campaign conversion data and provided audience, custom or customer data segments to find additional people predicted to convert. Exclusions can be set, traffic is reported in the "total expansion and optimized targeting" row, and the feature is on by default in Demand Gen, managed at the ad group level.

That default has history. In 2024, practitioners flagged the default setting for reaching users outside remarketing lists. In February 2026, Google converted Demand Gen Lookalike segments from hard targeting into suggestions, ending the 2.5%, 5% and 10% similarity thresholds that once bounded them. In Osio's sequence, the seed list is a starting signal, not a boundary.

The third item repeated the opening: horizontal, vertical and square versions of images and video, and Excellent ad strength throughout.

Why this matters for the marketing community

Demand Gen has become the container for most of Google's visual buying outside Search and Performance Max: video action campaigns were absorbed by April 2025, standalone Display is following, and Maps arrived in May. Nor is the feed exception in Osio's budget example marginal: Smarter Ecommerce found 35% of Demand Gen campaigns using a product feed in 2026, up from 16% in 2024.

The episode exposes a tension in Google's pitch to smaller advertisers. Asset Studio and asset optimizations exist to lower the cost of producing variety, and Osio's cousin is exactly the first-time advertiser they are built for. Yet his own testing guidance says more assets need more time and more budget per asset before the system reaches a conclusion. Cheaper production does not make evaluation cheaper. On Osio's own reasoning, an advertiser running near the $5 enforced minimum and filling an ad with generated variants faces arithmetic that does not work in its favour.

Timing adds another layer. Six weeks before the episode, Google began forcing overperforming targets up for budget-limited campaigns running target-based bidding across five types, Demand Gen included. Low-budget campaigns of the kind Osio used as his example are among the likeliest to hit that limit.

Then there is evidence. The figures Google has attached to these tools elsewhere - the 30% Demand Gen uplift, the 16% video enhancement gain, the 14% tag gateway lift - all come from its internal data, one narrowed by footnote to a single vertical. Buyers get a clear account of how the product team believes the system works, and little that can be checked independently.

Timeline

Summary

Who: Google, through its Ads Decoded series, with Ginny Marvin, Ads Product Liaison, interviewing Alejandro Osio, Product Manager for Demand Gen and YouTube performance ads. The guidance is aimed at advertisers running or considering Demand Gen campaigns on YouTube and Google's other visual surfaces.

What: A 19-minute 47-second episode on Demand Gen creative. Osio described a $100 daily budget as bare minimum best practice, declined to set any refresh interval for fatigued assets, distinguished gradual refresh from bulk testing, listed the current asset optimizations (including animated images in beta), pointed to recently released shareable previews for review, and said multimodal video creation in Asset Studio marks Gemini Omni's first appearance in Google Ads. No performance figures were cited.

When: Published October 1, 2026, on the Google Ads YouTube channel and in a companion post on Google's Ads & Commerce Blog. Multimodal video creation had already reached general availability on August 27, 2026.

Where: Google Ads globally, covering Demand Gen inventory across YouTube, Discover, Gmail, Maps and the Google Display Network, with Asset Studio located inside the Google Ads interface.

Why: Demand Gen is absorbing most of Google's visual inventory, so its creative guidance increasingly governs non-search spend. The $100 figure is 20 times the $5 minimum Google's API enforces and differs from other Google budget rules of thumb, while the argument that more assets need more budget per asset sits uneasily with tools designed to let advertisers produce more assets cheaply.