How to fix: Demand Gen campaign without audience segmentation

TL;DR

A Demand Gen campaign running with a single broad audience pool wastes budget. Demand Gen is a visual-first, AI-driven format that serves across the YouTube feed, YouTube Shorts, Discover, and Gmail, and it needs distinct audience signals to learn which creative-audience pairings convert. Pool everyone into one ad group and the algorithm averages bids and creatives against noise, inflating CPA. The fix is structural: split the campaign into separate ad groups by audience layer — Customer Match plus website visitors as the warm core, a Lookalike segment seeded from that Customer Match list as the warm prospecting layer, a custom intent segment for cold high-intent traffic, and one in-market or affinity segment as the cold broad layer. Give each ad group its own creative tuned to funnel stage.

Why it matters

Demand Gen, rebranded from Discovery Ads in late 2023, is Google's only campaign type that exposes Lookalike segments [1]. Lookalike segments expand from a first-party seed list (Customer Match, website visitors, YouTube channel viewers) to find new users with similar behaviour. That expansion lever is the campaign's main strategic advantage over generic YouTube or Display targeting — and it only works if you actually feed it a clean seed.

When all audiences live in one ad group, three things break. First, Smart Bidding sees aggregated conversion data from very different user populations and calibrates bids to the average, which under-bids on high-intent Customer Match users and over-bids on cold in-market users. Second, you cannot see per-audience CPA in reports, so there is no signal to act on. Third, creatives optimised for cold prospects run against warm returning visitors, and warm-message creatives leak to cold traffic that will not convert on a retention pitch.

Google's audience targeting documentation lists the segment types available in Demand Gen and explicitly calls out Lookalike segments as the format-exclusive lever [2]. Customer Match requires at least 100 active members updated within the last 540 days to stay eligible [3] — below that threshold the list is paused and Lookalike seeding fails silently. For background on the first-party data substrate this rule depends on, see Customer Match readiness.

How to fix

1. Audit the current structure. Open the Demand Gen campaign and review its ad groups (Campaigns → click the Demand Gen campaign → Ad groups). If you see a single ad group containing all targeting, or two ad groups split only by creative type rather than by audience, you have the problem this rule flags.

2. Refresh or build the Customer Match seed list. Export your highest-value customers from CRM — recent purchasers, repeat buyers, or qualified leads for B2B. Hash and upload via Tools → Shared library → Audience manager → Segments → New segment → Customer list. The list needs at least 100 active members updated in the last 540 days [3]; in practice aim for 1,000 plus for stable Lookalike seeding.

3. Create the Lookalike segment. Audience manager → Segments → New segment → Lookalike segment. Choose your Customer Match list as the seed. Demand Gen and YouTube are the only formats where Lookalike segments serve [1][2], so this segment is wasted everywhere else — keep it scoped here.

4. Build a custom segment from intent signals. Audience manager → Custom segments → New custom segment. Enter 5-15 high-intent search terms (the same queries that perform on Search), competitor domains, or app names that match your ideal customer profile. Custom segments work across Demand Gen, Display, Video, and Search [2].

5. Split the campaign into one ad group per audience layer. Inside the Demand Gen campaign, create:

   • Ad group A — warm core. Targets the Customer Match list plus website visitors (your data segments). Creative messaging: retention, upsell, loyalty.
   • Ad group B — warm prospecting. Targets the Lookalike segment. Creative messaging: value proposition, social proof.
   • Ad group C — cold high-intent. Targets the custom segment. Creative messaging: problem-solution, specific feature or offer.
   • Ad group D (optional) — cold broad. Targets one or two relevant in-market or affinity segments. Creative messaging: brand introduction, top-of-funnel hook.

6. Allocate budget proportionally and let it run. Skew initial spend toward ad groups A and B (warm tends to convert fastest), leave C and D smaller. After 14 days, pull the per-ad-group CPA report and rebalance — kill underperforming segments, scale the winners.

7. Verify. The campaign should now show four distinct ad groups, each with its own audience and creative set, and per-ad-group conversion data should be visible in the reporting view.

Common mistakes

• One ad group with five audiences stacked. Adding multiple audiences to a single ad group still pools their signal — Google does not report or bid per audience inside an ad group, only per ad group.
• Skipping Customer Match because the list is small. A 200-member list still works as a Lookalike seed and as a warm targeting layer; do not wait for "enough" data. Upload what you have and grow it.
• Using the same creative across all ad groups. Defeats the point of the split. Warm retention messaging on cold traffic converts at a fraction of the rate; cold prospecting messaging on warm returning users ignores their context.
• Forgetting the 540-day refresh window. Customer Match lists silently expire if not updated within 540 days [3] — schedule a quarterly CRM export to refresh them.
• Adding in-market or affinity as the primary layer. These are broad cold segments and should be the smallest budget allocation, not the largest. Demand Gen earns its keep through Lookalike and first-party data, not generic interest targeting.

FAQ

Why is Demand Gen called Demand Gen and not Discovery Ads? Google rebranded Discovery Ads to Demand Gen in late 2023 and added YouTube Shorts and YouTube in-stream placements to the existing Discover feed and Gmail surfaces [1]. Same campaign type, broader reach, new name.

Can I use Performance Max audience signals instead? No — Performance Max uses signals as a starting hint and expands aggressively beyond them. Demand Gen ad groups actually target the audiences you select, which is why segmentation matters here in a way it does not in PMax.

How small can a Customer Match list be? The minimum to stay eligible is 100 active members updated within 540 days [3]. For Lookalike seeding aim for 1,000 plus to get a stable expansion model. Below 100 the list is paused.

Should I exclude my Customer Match list from the cold ad groups? Yes — add the Customer Match segment as an exclusion on ad groups C and D so warm users do not get cold messaging. Audiences are additive within an ad group but exclusions are honoured.

How long until I can compare ad groups? Each ad group needs roughly 30 conversions over 30 days before Smart Bidding calibrates reliably (the standard Smart Bidding learning gate). For lower-volume accounts give the split four to six weeks before pruning.

Sources

1. Google Ads Help — About Demand Gen campaigns. Demand Gen rebrand from Discovery, YouTube and Discover and Gmail surfaces, Lookalike segments as format-exclusive lever.
2. Google Ads Help — About audience targeting. Segment types matrix (in-market, affinity, custom segments, your data segments, Lookalike segments) and which campaign types each supports.
3. Google Ads Help — About Customer Match. Minimum 100 active members updated within the last 540 days to stay eligible.