Every time a client asks "should we be doing YouTube too?", the real question is a different one: does their market contain demand that never passes through a Google search? If the answer is yes — and in most categories it is — Demand Gen is how you buy that demand today.
It is also the campaign most often opened badly, measured worse, and switched off after three weeks with the verdict "YouTube doesn’t work for us".
What Demand Gen is, and where it came from
Demand Gen is not a new format: it is a consolidation. It replaced Discovery campaigns first, then absorbed Video Action Campaigns. What is left is a single campaign type serving YouTube (in-stream, in-feed and Shorts), Discover and Gmail, with a set of controls Performance Max does not expose.
That is the difference that matters from an account manager’s seat: in Demand Gen you decide which creatives compete, against which audiences and — within limits — on which surfaces. It is the opposite of the "hand everything to the system" logic that governs Performance Max.
| Search | Performance Max | Demand Gen | |
|---|---|---|---|
| Demand | Existing and expressed | Existing, across networks | Latent, to be created |
| Search network | Yes | Yes | No |
| Creative control | Copy and assets | Asset groups, little visibility | High: assets per format, previews, native A/B tests |
| Audience control | Indirect | Signals, not targets | Real targets: lists, lookalikes, interests |
| Guiding metric | CPA / ROAS | CPA / ROAS | CPA on a long window + incrementality |
Demand Gen does not compete with Search. It is the one modern campaign type that never touches the search network. If overlap is your worry — the legitimate concern that holds many accounts back on Performance Max, which we covered in PMax controls and reporting — here the problem simply does not exist.
In Demand Gen, creative is the targeting
On Search the message arrives after the intent: the user has already typed what they want. On YouTube and Discover the message comes first, and it decides who stops. Which is why a Demand Gen campaign with perfect audiences and weak creative has never worked in any account we have seen.
Three operating rules, in order of impact:
- Images and video together, not one of the two. Google’s creative guidance reports roughly +6% conversions at the same spend for advertisers using both versus advertisers uploading images only. It is the cheapest improvement available in this campaign type.
- Every aspect ratio. 9:16 vertical for Shorts, 1:1 square and 16:9 landscape for the rest. A missing asset is not a cosmetic detail — it is a surface you simply do not appear on.
- The first three seconds are the campaign. On a scrolling surface there is no "getting to the point after the intro". Hook, problem and brand all belong in the opening.
The test worth more than any optimisation
Demand Gen has native A/B experiments. Use them to compare concepts, not variants: "product demonstration" against "customer problem", not the same video with two soundtracks. A test between two executions of one idea almost always ends in a statistical tie and burns three weeks.
Audiences: start from your data, not Google’s
The hierarchy that works is always the same, warmest first:
- Remarketing and customer lists. People who already know you convert at a cost no cold audience matches. It is also the most honest test of whether the creative holds up at all.
- Lookalikes off your best segments. Not off "all visitors": off purchasers, or off people past a funnel stage that means something. The quality of the seed list is the only variable governing the quality of the lookalike.
- Interest and in-market audiences. Useful for volume, never the starting point of a test.
The prerequisite is the same as everywhere else in the account: lists fed by correct tracking and properly handled consent. When consent is fragile the lists drain quietly — the mechanism we described in Consent Mode v2 and tracking.
Measuring a campaign that creates demand
This is where most tests fall over. A campaign reaching latent demand produces part of its effect elsewhere: in a brand search two days later, in a direct visit, in a conversion attributed to another channel.
Three corrections of method, before you even look at the numbers:
- Do not judge it on last click. Under last click Demand Gen loses by definition. It needs data-driven attribution, for the reasons we set out in data-driven attribution across Google Ads and GA4.
- Watch brand demand. If brand query impressions and clicks rise during the flight at constant spend, the campaign is doing its job even when direct CPA is unimpressive.
- Use a longer observation window. The cycle between first exposure and conversion on these surfaces is longer than on Search. Closing the test at 14 days reads only the fast half of the result.
The honest test is geographic. If budget allows, the cleanest read is not a CPA comparison between campaigns but a geo test: exposed regions against held-out regions, same seasonality. It is the only way to measure incrementality instead of attribution.
When it makes sense, and when it does not
| Situation | Demand Gen | Why |
|---|---|---|
| Search saturated, impression share above 80% | Yes | Growth can no longer come from expressed demand |
| New product, or a category nobody searches by name | Yes | Demand has to be created before it can be captured |
| No video assets and no budget to produce any | Not yet | Images alone leave out the surfaces that carry the volume |
| Fewer than 30 conversions a month account-wide | No | Not enough signal to exit the learning phase |
| No first-party lists, or no consent to use them | Lists first | A lookalike has nothing to learn from |
| Ecommerce with a live feed and stable PMax | Yes, alongside | It works the upper funnel PMax harvests later |
The first-30-days checklist
- A sensible minimum budget. Below roughly ten conversions a week the campaign never leaves learning: one properly funded market beats five starved ones.
- At least three creative concepts, each in vertical, square and landscape, images and video.
- Start on remarketing to validate the creative, then widen to lookalikes.
- Data-driven attribution live before launch, not after.
- A dedicated landing page. Scroll traffic tolerates far less friction than search traffic — everything we write about landing page experience applies double here.
- An A/B experiment on concepts from week two, not week one.
- A review at 45 days, with brand demand and the long window included.
Demand Gen replaces nothing: it adds. But it is the last place in the account where a good idea can still beat a bigger budget, and that alone is reason enough to treat it properly.
If you want to know whether your account has the room — and the data — to open it, the free audit starts from how saturated your existing demand already is. You can also see how we work and which services cover this ground.