Demand Gen's advantage over PMax is that you still choose who. That advantage only exists if the audience structure is designed to learn: one group per audience, comparable with each other, with expansion under control. This lesson gives you that structure.
One ad group per audience
Each Demand Gen ad group carries its own audiences and its own creatives. Mixing audiences inside a group makes it impossible to know which one performs and forces the same message on everybody. The base structure:
| Group | Audience | Message |
|---|---|---|
| Hot remarketing | Cart / product viewed in 14 days | Closing: the product, a reason to come back |
| Warm remarketing | Engaged visitors, 30-60 days | Brand + value proposition |
| Customers | Customer list | New arrivals, cross-selling, loyalty |
| Lookalikes | Similar to buyers / to qualified leads | Problem → solution, without assuming they know you |
| Custom segment | People who searched for your core terms | Approximate intent: a more direct message |
| In-market | Purchase intent in your category | Comparison and social proof |
Five or six groups, the same campaign budget, and in three weeks you have the comparison.
Lookalike segments
Demand Gen allows lookalike segments: Google looks for people similar to a seed (one of your lists). Three rules:
- The seed rules: lookalikes of "buyers" or "qualified leads" work; lookalikes of "all visitors" bring in people similar to browsers. The recommended minimum: 1,000 users in the seed (the more, the better).
- Adjustable reach: narrow (more similar, less volume), balanced, broad. Start narrow; widen if the CPA holds up.
- Exclude the seed (and your customers) from the lookalike group: you want new people.
It is the best criterion-driven prospecting tool left in Google since Display lost its similar audiences.
Custom segments
Two types that are useful in Demand Gen:
- By searches: people who searched Google for certain terms (your Search keywords that convert). It brings intent into an inventory with no search: the custom segment group is usually the most profitable of the prospecting groups.
- By interests/URLs: people who visit websites in your sector or use certain apps. Broader, for consideration.
Expansion: Optimized targeting
As in Display (module 7 of the Intermediate level), Demand Gen offers to expand beyond your audiences when the algorithm believes it will convert. The rule is identical: off in remarketing and in groups you want to measure cleanly; on only in prospecting with solid measurement, and after seeing how the "expansion" row performs in the report.
Exclusions
- Recent customers out of prospecting and hot remarketing.
- The seed out of the lookalike group.
- Employees, closed leads.
- Demographics: Demand Gen allows age/gender; use it as a filter if the data justifies it, not out of prejudice.
Comparing audiences by the same yardstick
After 3-4 weeks, by group: cost, click-through conversions, CPA, conversion rate, and frequency. Rules for reading it:
- Hot remarketing always wins on CPA: that is not to the campaign's credit, it is the audience. Don't use it to justify the prospecting.
- Between prospecting groups (lookalikes, custom segment, in-market), the click-through CPA decides how budget is shared out (move it in steps).
- A group with a good CPA and little volume deserves a wider reach (lookalikes: from narrow to balanced) before it deserves more budget.
- High frequency with a worsening CPA is fatigue (lesson 4), not a bad audience.
💡 Ninja trick: comparing audiences is the second job of Demand Gen Optimizer: a table by group/audience with cost, click-through conversions, CPA and trend, sorted from best to worst, with a proposal to move budget (when the groups have their own budgets) or to pause the audiences that never convert. The "one group per audience" structure is what makes that reading possible.
What you should remember
- One group per audience, comparable; 5-6 base groups.
- Lookalikes with a quality seed (buyers, qualified leads), narrow reach first, seed excluded.
- Custom segments from searches = intent in visual inventory.
- Expansion off in remarketing; on only with data.
- Compare on clicks and CPA; hot remarketing does not justify the prospecting.