Getting Demand Gen to work comes down to one thing: finding the right audiences.
That sounds obvious. But unlike Search campaigns where keywords do the targeting for you, Demand Gen puts the entire burden on audience selection. Pick wrong, and you’re burning budget on people who will never buy. Pick right, and you’ve got a scalable acquisition channel.
This guide covers every audience type available in Demand Gen. But more importantly, it covers how to actually test and combine them. Because that’s where most advertisers get stuck.
Why Audiences Matter More in Demand Gen
Demand Gen is a completely different animal than Search campaigns.
In Search, the keyword does most of the targeting work. Someone searches “blue running shoes,” and Google shows your ad. Intent is already there.
Demand Gen doesn’t work that way. There’s no keyword to lean on. Google is showing your ads to people browsing YouTube, Gmail, Google Maps, and other properties hoping they’ll be interested. That means audiences are doing all the heavy lifting.
Get your audience targeting wrong, and you’re burning budget reaching people who’ll never buy from you.
Based on my experience, the difference between a profitable Demand Gen campaign and a money-losing one almost always comes down to audience selection. You can have beautiful creative and solid offers. But if you’re showing them to the wrong people, none of it matters.
For a complete overview, check out my guide on the Demand Gen complete guide.
Types of Audiences Available
Google gives you a lot of audience options in Demand Gen. Let me walk you through each one so you can figure out which ones work for your store.
Customer Match Lists
This is where you start. Customer Match lets you upload your own email lists, phone numbers, or customer data directly to Google.
It’s your most valuable audience because these are people who’ve already bought from you or engaged with your business.
For ecommerce, I always start by uploading your customer email list. Make sure your data is clean and you have consent to use it. Google will match your list against their user database and create an audience of logged-in users.
The match rate varies. For ecommerce stores, you’re typically looking at 20-50% of your list matching in Google’s system.
Don’t just upload your list once and forget it. Refresh it monthly with your latest customers. The more recent the data, the better the matches. And more importantly, the better your lookalikes will be (more on that below).
For more details, check out my guide on Google Customer Match.
Lookalike Segments (The Demand Gen Advantage)
This is where Demand Gen pulls ahead of other campaign types. Lookalike audiences are unique to Demand Gen, and they’re incredibly powerful for ecommerce scaling.
Here’s how they work: You provide a source audience (usually your Customer Match list or another audience), and Google analyzes the characteristics of those people. Then it finds similar users who aren’t yet in your audience and creates a “lookalike” segment.
Google gives you three flavors of lookalikes: narrow, balanced, and broad.
Narrow lookalikes are closest to your original audience. If your source is 10,000 customers, a narrow lookalike might find another 50,000 to 100,000 similar people. The match quality is highest, but the reach is limited.
I use narrow lookalikes when I’m being budget conscious or when I have a high quality source audience.
Balanced lookalikes cast a wider net. You might find 200,000 to 500,000 similar users. This is usually where I start for most ecommerce campaigns because you get good reach without sacrificing relevance.
Broad lookalikes expand even further. These can reach millions of users, but match quality drops. There’s still a connection to your source audience, but it’s weaker.
Start narrow. Always.
Narrow gives you the highest match with your seed list. That’s where you want to begin because it tells you whether Demand Gen can find more people like your customers at all. If narrow works, expand to balanced. If balanced works, test broad.
This is the scaling progression. Don’t jump to broad because you want more reach. Earn your way there by proving each level works first.
One thing that matters more than the lookalike type: the quality of your source list. If you upload your entire email list including people who bought once five years ago, your lookalikes will be diluted. Use customers from the last 6 to 12 months. If you can segment by repeat purchasers or high AOV, even better. That’s your strongest seed list.
Minimum list size: Google requires 500+ users. But practically, if you’re under 1,000, the lookalikes won’t be reliable enough to spend real budget on.
Custom Segments
Custom segments let you target people based on their actual behavior. What they’ve searched for, websites they’ve visited, or apps they’ve used.
This is different from Customer Match because you’re not uploading your own data. You’re defining behavior patterns.
For example, you could create a custom segment for “people who searched for running shoes in the last 30 days” or “people who visited competitor websites.” It’s useful for finding warm audiences beyond your customer list.
I find custom segments work best when combined with other audiences rather than as a standalone approach. They’re good for reaching people showing intent signals without the precision of Customer Match.
For a deeper dive, see my article on Google Ads Custom Audience Segments.
In-Market Audiences
These are Google’s pre-built audiences. People who’ve shown intent to buy in specific categories.
For ecommerce, in-market audiences are useful. But here’s my take: they’re not as precise as your own customer data or lookalikes. Google’s definitions are broad, and you’re competing with every other advertiser targeting the same audience.
They’re a good scaling lever, but don’t rely on them as your primary targeting.
Read more about In-Market Audiences here.
Affinity Audiences
These are interest-based audiences. Google groups people by what they care about.
For ecommerce, affinity audiences are mostly useful for awareness and testing new market segments. They’re not my first choice for performance focused campaigns, but they’re worth testing if you’re trying to reach a completely new audience.
Check out Affinity Audiences for more context.
Remarketing Audiences
These are people who’ve already interacted with your business. Visited your site, viewed products, abandoned carts.
Remarketing is essential in Demand Gen. You should always have a remarketing audience active because these people know who you are and are more likely to convert.
For ecommerce, I typically create separate remarketing audiences: all website visitors, product viewers, and cart abandoners. You can target them with slightly different messaging depending on where they dropped off.
See Remarketing Audiences for setup details.
Demographics and Detailed Demographics
You can narrow targeting by age, gender, parental status, and household income. I use these mostly to exclude unprofitable segments.
For example, if your product skews toward men 35-54, you might exclude younger age groups to improve efficiency.
Be careful though. Demographic targeting is blunt. It’s better to rely on your audience data to do the targeting than to manually layer demographics on top.
Here’s why: if your lookalike audience is already finding the right people, adding demographic restrictions might just reduce your reach without improving results.
Read up on Demographics and Detailed Demographics.
How to Actually Test Audiences
This is the part most guides skip. They list the audience types (which we just did) and then say “test them.” But they don’t tell you how.
Here’s the actual testing framework.
The Structure: One Audience Per Ad Group
Put one audience in each ad group. Use the same ads across all ad groups.
This is the only way to get clean data on which audience is actually working. If you stack three audiences in one ad group, you’ll never know which one drove the results.
Same ads, same bids, same budget allocation. The only variable is the audience. That’s how you isolate what’s working.
When to Make Decisions
You need both time and conversions before you can call a winner.
Don’t pause an audience after 3 days because the CPA looks bad. Google’s algorithm needs time to optimize delivery within each audience. And small sample sizes lie.
Give each audience at least 2-3 weeks and a meaningful number of conversions before deciding. The exact number depends on your volume, but the principle is the same: don’t make decisions on thin data.
The Testing Order
Start with lookalikes, not remarketing.
This surprises people. Conventional wisdom says start with warm audiences. But here’s the thing: your remarketing pool in Demand Gen is usually small. Demand Gen is a top of funnel channel. Most of the value comes from reaching new people.
So start with a narrow lookalike based on your customer list. This is your first real test of whether Demand Gen can find new customers for you.
Once that’s running, add remarketing as a separate ad group. Then test a balanced lookalike. Then in-market or custom segments.
The progression looks like this:
- 1. Narrow lookalike (from best customers)
- 2. Remarketing (site visitors, cart abandoners)
- 3. Balanced lookalike (expand reach)
- 4. In-market or custom segments (test new pools)
- 5. Broad lookalike (only if the above are working)
The Biggest Mistake: Random Audience Selection
The number one mistake I see is advertisers randomly selecting an audience and running with it. Without testing alternatives.
They pick a balanced lookalike, it does OK, and they never test whether narrow or broad would have been better. Or they use in-market audiences because the category name sounds right, without testing their own customer data as a seed list.
Getting Demand Gen audiences right is not about picking the “correct” audience. It’s about systematically testing combinations until you find what works for your specific store and products.
This takes time. Budget a few weeks of dedicated testing before you expect Demand Gen to be a reliable channel.
Optimized Targeting
Google enables Optimized Targeting by default in Demand Gen. It lets Google expand beyond your selected audiences to reach additional users it thinks will convert.
Leave it on for your prospecting audiences (lookalikes, in-market, custom segments). It generally helps Google find more people similar to your audience, and at the prospecting stage, more reach is usually good.
Turn it off for remarketing.
Why? Because the whole point of remarketing is targeting people who already know you. If you let Google expand beyond that, you’re basically turning your remarketing ad group into another prospecting ad group. It defeats the purpose.
This is one of those settings that’s easy to overlook but can quietly waste budget if you’re not paying attention.
Audience Exclusions (Don’t Skip This)
Here’s something most people don’t do well: audience exclusions. You should always be excluding people who’ve already converted.
Why. Because showing ads to someone who just bought from you is wasting money. They’re already a customer. That budget would be better spent on a prospect.
Here’s what I suggest: create an exclusion audience of customers who’ve purchased in the last 30 days. You can exclude this from your acquisition audiences (lookalikes, in-market, etc.). Save your budget for people who haven’t bought yet.
Also exclude across ad groups. If someone is in your remarketing audience, exclude them from your lookalike ad groups. Otherwise you’re paying prospecting CPMs to reach people you could be reaching through cheaper remarketing.
Exclusions are where a lot of ecommerce stores leave money on the table. Be intentional about who you’re not targeting.
How Demand Gen Audiences Differ from Performance Max
This is an important distinction that confuses a lot of people. In Performance Max, audiences are “signals,” not hard restrictions. Google treats them as suggestions and will still show your ads to people outside your audiences if it thinks they’ll convert.
In Demand Gen, audiences actually constrain your targeting. If you select specific audiences, Google will primarily show your ads to those people. Optimized targeting can expand beyond them, but the audiences are doing real work.
This means audience selection matters more in Demand Gen than in Performance Max. In Performance Max, you can be looser with audiences because the algorithm will find additional people anyway. In Demand Gen, your audience choices directly impact who sees your ads.
If you’re coming from Performance Max, don’t apply the same audience strategy to Demand Gen. Be more intentional and more restrictive with your audiences in Demand Gen.
For a full comparison, check out my guide on Demand Gen vs Performance Max.
If you’re coming from Performance Max, don’t apply the same audience strategy. In PMax you can be loose because the algorithm finds people anyway. In Demand Gen, your audience choices directly determine who sees your ads. That’s why the testing structure above matters so much.
Common Audience Mistakes
Too many audiences in one ad group. People think more audiences means better reach. It doesn’t. It means you can’t tell what’s working. One audience per ad group. Always.
Not testing alternatives. This is the big one. Picking an audience and sticking with it without ever testing whether something else would work better. Audience testing is not optional. It’s the whole game in Demand Gen.
Not excluding converters. If you’re not excluding customers who’ve already bought, you’re wasting budget retargeting people who are already in your funnel.
Using old customer data. A customer from three years ago is less valuable for lookalikes than one from three months ago. Refresh your lists regularly.
Relying only on Google’s stock audiences. In-market and affinity are useful, but generic. Your own customer data will outperform them. Upload your lists.
Giving up too early. Audiences need time and conversions before you can judge them. Killing an audience after a few days of bad data means you’ll never find what works.
Wrapping Up
Demand Gen audience targeting isn’t about finding the one perfect audience. It’s about building a systematic testing process that narrows down what works for your store.
Start with a narrow lookalike from your best customers. Add remarketing. Expand to balanced lookalikes and in-market as you scale. One audience per ad group. Same ads. Give each test enough time and data before deciding.
The advertisers who get the best results from Demand Gen are the ones who treat audience testing like a discipline, not a one-time setup task.
