Demand Gen on YouTube: where targeting control ends and automation begins, compared to PMax

Demand Gen on YouTube: Targeting and Differences from PMax

Written by Nicolò Caiti I have been managing real Google Ads accounts since 2018, from manual Search campaigns to Performance Max and Demand Gen automation.

Published: Jun 17 2026 Updated: Jun 17 2026 Google Ads · Demand Gen · YouTube

In brief

What you will find in this article

How Demand Gen targeting actually works on YouTube: audience segments, custom segments, Lookalike and optimized targeting
The difference between a "binding" targeting signal and a "suggestion" signal, the point that changes everything compared to PMax
The complete matrix: what you control in Demand Gen and what you control in Performance Max, line by line
What Google's official documentation says and what instead is inference from real campaigns, clearly separated
Behaviours observed on YouTube, Shorts and in-feed, explicitly flagged as inferences
The operational workflow for deciding when to use Demand Gen and when to use PMax on YouTube inventory

You load the same videos, the same Lookalike and the same budget into two campaigns. One is Demand Gen, the other is Performance Max. After two weeks on YouTube you get different distributions, different CPAs and a completely different level of control over where impressions are served. The cause is not the creative: it is how the two campaign types treat targeting.

Demand Gen targeting is not "PMax with videos". It is an audience-first model where the segments you select carry real weight, while in PMax audience signals are only a starting point for automation. This article clearly separates what is officially documented from what emerges from directly managing Demand Gen on YouTube campaigns.

The goal is not to decide which campaign "is better", but to understand where your control ends and where Google's begins, because that is where the difference between the two is decided.

How Demand Gen targeting works on YouTube

Demand Gen targeting is often misunderstood. The most common confusion is treating it as deterministic targeting like classic Video, or as full automation like PMax. It is neither: it sits in between, offering targeting options that range from tightly controlled segments to broad machine learning expansion, depending on how you configure it.

✅ Confirmed by Google Ads: Demand Gen has an audience-first approach and reaches users as they scroll YouTube, YouTube Shorts and the Google feeds, including Gmail and Discover. Audiences are made up of segments, groups of people with interests, intents and demographics estimated by Google, and include custom segments, your own data, interest and in-market segments, plus the Lookalike segments exclusive to Demand Gen. Source: Google Ads, About Demand Gen campaigns.

The main control lever sits at ad group level: you choose the segments, decide whether to enable or disable optimized targeting, and separately build Lookalike segments with a reach chosen between narrow, balanced and broad. This is the real decision centre of a Demand Gen campaign on YouTube.

✅ Confirmed by Google Ads: During creation, audiences are set at ad group level by choosing custom segments, your data, Lookalike segments, interests or detailed demographics. Optimized targeting can be kept or deselected; when it is on, demographic expansion also takes effect, so the system can go beyond the demographic signals you selected such as age, gender, household income and parental status. Source: Google Ads, Create a Demand Gen campaign.

The practical question for anyone managing YouTube inventory is: is the segment I select in Demand Gen a constraint or a suggestion? The documented answer is that it depends on optimized targeting. With optimized targeting off, segments behave much more like a constraint; with optimized targeting on, segments become signals the system can expand from.

⚠️ Inference on product direction, not documented as current policy: it is reasonable to hypothesise that Google will push Demand Gen increasingly towards automation, over time narrowing the control gap relative to PMax. The shift of primary bidding towards tROAS points in this direction. This is a plausible trajectory for the next 2–3 years, not a current fact: today the control difference between the two campaigns is real and measurable.

The three types of audience signal in Demand Gen: constraint, suggestion and expansion

Before comparing Demand Gen and PMax it is necessary to distinguish three completely different types of audience signal, because they have a different impact on how much you actually control distribution on YouTube. Demand Gen offers more control here than PMax, but that control only holds if your creative assets and segments are configured deliberately.

✅ Segments as a constraint, the optimized targeting OFF case

Custom, in-market and affinity segments selected at ad group level
With optimized targeting disabled the system stays close to the chosen segments
It is the configuration with the most control over Demand Gen targeting
Suited to high-intent first-party audiences and remarketing
More predictable distribution on YouTube, smaller reach

⚠️ Optimized targeting, the segment becomes a suggestion

The selected segment becomes a signal, not a boundary
Demographic expansion also kicks in beyond age, gender, income
Google looks for conversions even outside the manual segments
Increases reach and volume, lowers predictability of who sees the ad
This is where Demand Gen starts to approach PMax behaviour

✅ Lookalike segments, exclusive to Demand Gen

Built from a seed list with narrow, balanced or broad reach
Not reusable outside Demand Gen: on other campaigns they show as "Eligible - Limited"
Give control over the balance between similarity and reach
Work well on YouTube for prospecting new users
They are the main control lever PMax does not offer in the same way

⚠️ PMax audience signals, always just a suggestion

In PMax the segment is never a constraint, it is an input for automation
The system expands beyond the audience signal by product definition
No granular control over the specific YouTube placement
Lookalike segments cannot be used as in Demand Gen
Maximum cross-channel reach, minimum control over who and where

Demand Gen vs PMax matrix: what you control on YouTube

This table summarises the current state (June 2026) line by line. Official sources are indicated where the information is documented; where it comes from directly managing campaigns this is stated explicitly.

Targeting lever	Demand Gen	Performance Max	Control on YouTube	Source
Manual audience segments (custom, in-market, affinity)	Yes, at ad group level	Signal only	High in Demand Gen	Official Google Ads
Lookalike segments	Yes, exclusive	No	Demand Gen only	Official Google Ads
Optimized targeting OFF	Can be disabled	Cannot be disabled	High in Demand Gen	Official Google Ads
Audience exclusions	Yes	Limited	Medium	Official Google Ads
YouTube placement control	Partial	Almost none	Medium-low	Direct campaign analysis ⚠️
Controlled creative format	Yes (video/image)	Mixed assets	High in Demand Gen	Direct campaign analysis ⚠️
Cross-channel expansion	YouTube/Discover/Gmail	Entire Google network	Broader in PMax	Official + direct analysis ⚠️
Transparency on where the ad appears	Partial	Low	Medium-low	Direct campaign analysis ⚠️

⚠️ Methodological note on the table: rows marked "Direct campaign analysis" come from observing the behaviour of Demand Gen and PMax campaigns run on the same audiences and the same YouTube inventory. These are not official Google sources; product behaviour can change without notice. The table reflects the state observed in June 2026 and should not be used as a permanent guarantee.

What it really means on YouTube: reach, Shorts and in-feed

✅ Confirmed by Google Ads: Demand Gen lets you reach up to 3 billion monthly active users as they scroll YouTube and their Google feeds, with YouTube Shorts averaging over 50 billion daily views. The campaigns offer robust audience controls, including Lookalike segments to find new customers similar to existing ones, and differentiate between Lookalike and optimized targeting as distinct levers. Source: Google Ads, Demand Gen FAQ.

On YouTube this translates into a practical point: Demand Gen distributes across three main surfaces, YouTube in-stream/in-feed, Shorts and Discover/Gmail. You do not have granular control over the individual placement like in a classic Video campaign, but you do have audience control that PMax does not give you, because in PMax the YouTube inventory is just one of many surfaces inside a single black box.

For anyone aiming at prospecting on YouTube with some governance over who sees the ad, Demand Gen is the more sensible choice today: Lookalike and optimized targeting OFF provide a level of predictability that PMax, by product definition, does not offer. It also lets you keep a conversion focused setup without losing the brand awareness value of the discovery surfaces, as long as you feed it high quality creatives. A low rating or thin volume in Demand Gen is almost always a creative or seed-list issue, not targeting "switched off" by the system.

Inferences from directly managing campaigns

⚠️ This section describes direct observations from managing real campaigns, not official documentation from Google.

Optimized targeting OFF: the real control switch. On the Demand Gen campaigns on YouTube where I disabled optimized targeting, distribution stayed much closer to the selected segments, with a more stable CPA but lower reach. Turning it back on, volume grew quickly but the audience reached widened beyond the demographic signals I had set, exactly as demographic expansion would predict. It is the lever that most moves Demand Gen towards or away from PMax behaviour.

Lookalike narrow vs broad: reach matters more than the segment. On YouTube, narrow Lookalikes performed better in retention and traffic quality, while broad ones gave volume but with a drop in conversion rate. Verified across multiple seed lists and sessions. It is not an official figure, but the pattern is consistent enough to treat Lookalike reach as the first lever to test, before the creative.

PMax vs Demand Gen on the same audiences: real cannibalisation. Running PMax and Demand Gen in parallel in the same account, I observed audience overlap and moments of cannibalisation on YouTube. PMax tends to absorb the easier, lower-CPA conversions, leaving Demand Gen the more expensive prospecting. This does not show up in standard reports but in cross-analysis of placements and audiences, and it should be kept in mind before comparing the two CPAs as if they were independent.

💡 The observation that surprised me most: I compared the same Lookalike used as an audience signal in PMax and as real targeting in Demand Gen, on the same budget. In Demand Gen the audience actually reached was much closer to the seed list; in PMax the signal was effectively diluted in cross-channel expansion within a few days. A rigorous approach would require a dedicated incrementality test, but the pattern is clear enough to treat Demand Gen as the right choice when control over the YouTube audience really matters.

tROAS bidding does not change the placement. The shift of Demand Gen's primary bidding towards tROAS pushes you to read the campaign as conversion-driven, but on YouTube the surfaces stay discovery-type: tROAS optimises towards the most valuable users inside a top-of-funnel channel, it does not turn Demand Gen into a search campaign. The distinction between bidding objective and channel nature is real and should be kept separate when reading the results.

Operational workflow: when Demand Gen and when PMax on YouTube

1. Choose the campaign based on how much audience control you need. If you need to govern who sees the ad on YouTube, prospecting on Lookalike, exclusions, controlled remarketing, start from Demand Gen. If the goal is maximum cross-channel reach with conversions driven by automation on CPA/ROAS and the specific placement does not matter to you, PMax is a better fit. In a full funnel approach the two are complementary, not interchangeable: Performance Max and Demand Gen answer two different needs. When you compare Performance Max vs Demand at the targeting level, the split is clear: Demand Gen focuses on governed audience prospecting, while Performance Max (PMax) focuses on automated, network-wide reach.

2. In Demand Gen, decide optimized targeting deliberately. Start with optimized targeting OFF when you want control and predictability, turn it on when you need to scale volume and accept expansion beyond the segments. Remember that turning it on also brings demographic expansion into play, so your demographic signals become indicative.

3. Build Lookalikes with the right reach. For quality prospecting on YouTube start from narrow or balanced; move to broad only when you need reach and have a creative that holds up. Create separate ad groups for Lookalikes, as the creation flow provides, so you isolate performance and read it cleanly.

4. Do not compare PMax and Demand Gen CPAs as if they were independent. If they run in parallel in the same account, check audience and placement overlap before drawing conclusions. Cannibalisation on YouTube is real and skews the direct comparison between the two CPAs.

⚠️ Long-term caveat: the current state is not guaranteed. Google is pushing Demand Gen towards automation and the migration of Display into Demand Gen is underway. The control gap relative to PMax may narrow over time. The structural solution is not to depend on a single campaign type, but to build solid seed lists and first-party audiences, which keep their value whatever targeting lever the product offers.

FAQ on Demand Gen, YouTube targeting and differences from PMax

How does Demand Gen targeting work on YouTube?

Demand Gen uses an audience-first approach: at ad group level you select custom, in-market, affinity, first-party and Lookalike segments, and decide whether to enable optimized targeting. With optimized targeting disabled the system stays close to the chosen segments; turning it on, segments become signals Google expands from, including demographic expansion. Distribution happens on YouTube, Shorts, Discover and Gmail.

How does Demand Gen targeting differ from PMax?

In Demand Gen audience segments carry real weight and optimized targeting can be disabled, so you keep control over who sees the ad. In PMax audience signals are only an input for automation, which always expands beyond the signal and distributes across the entire Google network with little placement transparency. Lookalike segments, moreover, are exclusive to Demand Gen.

Do Lookalike segments work in Performance Max too?

No. Lookalike segments are only available for Demand Gen. If you put a Lookalike into an audience used elsewhere, that audience shows as "Eligible - Limited" on other campaign types, because Lookalikes do not work outside Demand Gen. It is one of the sharpest differences between the two campaigns on YouTube targeting.

Should I disable optimized targeting in Demand Gen?

It depends on the goal. Disabling it gives more control and predictability over the selected segments, useful for remarketing and high-intent first-party audiences. Enabling it increases reach and volume but extends the audience beyond the signals, including demographic expansion. In my experience it is the first lever to manage deliberately when comparing Demand Gen with PMax behaviour.

Do Demand Gen and PMax cannibalise each other on YouTube?

From direct experience yes, when they run in parallel in the same account on the same audiences. PMax tends to absorb the easier, lower-CPA conversions, leaving Demand Gen the more expensive prospecting. It does not show up in standard reports but in cross-analysis of audiences and placements. Before comparing the two campaigns' CPAs it is worth checking the overlap.

Does tROAS bidding turn Demand Gen into a pure conversion campaign?

No. tROAS optimises towards higher-value users, but Demand Gen's surfaces on YouTube, Shorts, Discover and Gmail stay discovery-type, top-of-funnel. Changing the bidding strategy does not change where the ads appear: it changes how the system looks for value inside a discovery channel, not the nature of the channel itself.

All articles

See all →