How to Build a Google Ads Demand Gen Campaign with AI Video Ads and Portfolio Bidding

Google keeps nudging advertisers to hand more of the wheel to its AI. This session was a live build that worked with that trend instead of fighting it: we set up a Google Ads Demand Gen campaign from scratch, generated ad creative with the AI tools now baked into the platform, and talked through the settings that quietly decide whether your budget gets spent well or burned. The bigger theme was leverage, knowing when to let Google’s automation run, where to put up guardrails, and how a one-person operation can build its own tools with AI instead of paying for someone else’s.

Why Demand Gen Instead of Performance Max

The whole direction of Google Ads is “give our AI more control.” Performance Max takes that to the extreme: hand over your site and locations, and Google handles the rest. That can work well for ecommerce, but it’s a black box that often disappoints home-services advertisers who need more say.

Demand Gen sits in the sweet spot. It still leans on Google’s AI, but it hands you back real control over the audiences, demographics, and placements you target. And it runs across Google’s visual surfaces, YouTube and Shorts, Discover, Gmail, the Display Network, and now Maps, all from a single campaign.^{[1]About Demand Gen campaignssupport.google.com}

• Optimize for conversions, not clicks: When AI builds a campaign for you, it tends to default to clicks, which buys cheap, low-intent traffic. Choosing a conversion goal lets Google find the people who actually pick up the phone.
• Lock your locations: Forget to restrict targeting and you’ll pull penny clicks from all over the world with zero value. Review the whole campaign before you scale.
• Lean mobile-first: Most of this traffic is on phones, often seconds away from a call, so mobile is where the early budget belongs.

Generating Ad Creative Inside Google Ads

The headline feature this week: Google now lets you generate ad creative right in the ad builder. You can create and edit images from a text prompt, then turn them into short videos without leaving the platform.^{[8]Creative enhancements and generative AI in Demand Gensupport.google.com} The video step is powered by Google’s Veo model and produces clips up to about ten seconds long from your images, at no extra cost.^{[7]Veo in Google Adsbusiness.google.com}

• Free, and getting better: It taps Google’s image and video models inside the ad flow, and clip lengths should climb as the models improve.
• Strict policies: Generation is picky, people in AI-made creatives get flagged, and even tame images can bounce. Expect to work around the rules.
• Image ads still run: When video generation stalls, a single strong image ad still serves on YouTube, so it’s a fine fallback.

The New Maps Channel

One genuinely new setting: Demand Gen channel controls now let your ads appear on Maps, alongside YouTube, Gmail, Discover, and the Display Network.^{[2]Channel controls in Demand Gen campaignssupport.google.com} As someone scrolls a listing, your ad can surface further down, a placement that didn’t exist a few weeks ago.

• Promising for home services: A local-intent surface is a natural fit for service businesses.
• Early and uneven: It’s the kind of new inventory that can be cheap before everyone piles in, but it’s still settling, so watch your delivery closely.

Match Types Quietly Got Renamed

A member asked about broad match, and the answer reframes all of search. The match types still carry their old names, but in practice each one now behaves a notch looser than it used to.^{[4]About keyword matching optionssupport.google.com}

• Broad is really “intent match”: It reaches the widest set of loosely related searches, which means a lot of off-target traffic.
• Phrase behaves like the old broad: It at least keeps the meaning of your term in the query.
• Exact behaves like the old phrase: It’s the closest you can get to genuine relevance today.
• For search, lean phrase or exact: Broad burns budget on customer-service searches and accidental clicks. In Demand Gen audiences, keywords are broad no matter what you type, it’s all AI in there.

Portfolio Bidding Saves Your Budget

This was the most actionable tip of the call. With a plain “maximize conversions” target, Google can decide a single click is sure to convert and pour a big chunk of your target CPA into that one click, then the visitor doesn’t convert and the money’s gone.

The fix is a portfolio bid strategy. Unlike a standard strategy, a portfolio strategy lets you set a target CPA and a hard maximum-CPC bid limit at the same time.^{[5]About Target CPA biddingsupport.google.com} In plain terms: “I’ll pay up to this much for a conversion, but never more than this for a single click.” That cap is what stops the budget bonfire.

Call Extensions vs Landing-Page Calls

A member running phone-lead campaigns asked whether to drive calls from the landing page or from call extensions. First, a platform change worth knowing: the pure call-only ad format is being retired, with no new ones allowed and existing ones losing impressions, so calls now flow through responsive search ads with call assets instead.^{[6]Transition from call ads to call assetssupport.google.com}

• Call assets pull lower-quality calls: Accidental taps, butt-dials, and people who think they’re reaching customer service skew the volume.
• Landing-page calls convert better: When someone clicks through and calls from the page, the intent is higher.
• Run both and let Google optimize: Use a conversion goal and let the system sort window-shoppers from real buyers, just make sure your landing-page calls are actually being tracked back into Google.

Hand Google the Keys: Server-Side Conversion Tracking

The most powerful conversion isn’t the form fill, it’s the server-side import. Your back end captures the signup, then passes the conversion (with Google’s click ID) back to the Ads account. That gives Google clean, confirmed signals about who actually converts, so its AI can hunt down more of those people.

Build Your Own Tools: A Video Editor Made with AI

To tie it together, the session closed on a bigger idea: treat your computer as a blank canvas, not a hammer. Instead of fighting an off-the-shelf video editor’s bugs and limits, you can have an AI coding assistant build an interface that works exactly the way you want.

• A custom “Premiere Pro”: A full video editor, built over time on a flat-rate AI subscription rather than at per-token API rates.
• The AI edits, too: Feed it a long clip and ask for a short narrated version, and it can write the script, generate the voice, and cut it together.
• Two models beat one: Running two AI assistants as a tag team puts two sets of eyes on the same project, so less slips through.