Check the channel controls before you look at creative, audience or bidding. Google has been folding standalone Display into Demand Gen, so a campaign can start buying open-web inventory without anyone having chosen that.

On this page

• What should I check first when Demand Gen performance changes?
• What did I get wrong the first time?
• What do exploding views and collapsing sales mean?
• Does target CPC fix Demand Gen quality?
• What order should I diagnose a Demand Gen campaign in?

What should I check first when Demand Gen performance changes?

channel_controls, at ad group level — not campaign level. That is where the switches live.

Two fields matter:

• Channel strategy. One value means all channels. Another means owned-and-operated only, which is what keeps a Demand Gen campaign off third-party Display.
• Selected channels — YouTube in-stream, in-feed, Shorts, Discover, Gmail, Display.

Unset booleans come back blank, not false. On a telehealth account the two ad groups returned only display=true, gmail=true, with Discover and YouTube off — while the ad groups were named "Discovery."

What did I get wrong the first time?

Blank does not automatically mean off. I control-tested against a second account before concluding: its Demand Gen groups returned owned-and-operated strategy with every channel blank, and one group named "Open | Gmail" returned gmail=true alone. The flags matched the names exactly, which is what confirmed the reading.

Control-test on a known-positive before trusting any absence-based finding. A detector that has never been validated against a case it should catch is not evidence.

What do exploding views and collapsing sales mean?

Delivery moved to cheap inventory. More impressions and clicks for the same money, none of it converting.

The pattern on that telehealth account: $612.33 spent, 715 clicks, 0 purchases, 1 registration — 0.14% of clicks over 12 days, with 97.7% of spend landing on 844 URLs across 685 hostnames of made-for-advertising inventory.

Does target CPC fix Demand Gen quality?

Target CPC is available on Demand Gen at campaign and ad-group level, and it is the only steering lever that needs zero conversion data.

But it is a price control, not a quality control. It kills the $4–8 clicks and pushes spend toward $0.20–0.40 accidental-click game farms. Cheaper junk is still junk. On that account, at a 0.14% registration rate, cheaper clicks would not have registered either.

Also worth knowing: Maximize Clicks on Demand Gen does not support cost-per-click bid limits.

What order should I diagnose a Demand Gen campaign in?

1. Channel controls, at ad group level.
2. The biddable conversion goal — is the campaign bidding on something that actually fires?
3. The audience layer — is there one at all?
4. Placement quality, and set expectations honestly about what a blacklist can do.
5. Only then, creative.

Related questions

My average CPC doubled and I didn't change anything — what actually causes that?

The universal published answer is "a competitor entered the auction," which is unfalsifiable and useless. Before you accept it, separate a genuinely more expensive auction from a shifted traffic mix. Read the full answer.

My Google Ads conversions dropped and I didn't change anything. Is it broken, or is the data still filling in?

Before you diagnose anything, check whether the window you are reading is old enough to be complete. If any day in it is less than 4 days old, the number is wrong and there is nothing to diagnose yet. Read the full answer.

My Local Services Ads leads suddenly stopped and I didn't change anything — what do I check first?

Check your Google Verified status and your budget pacing before anything else. A profile suspension and a paced-out budget both produce a total stop, and everything else produces a decline rather than a stop. Read the full answer.

Written by Usama Khan, who runs Google Ads for e-commerce brands and home-services companies at AdventurAds. Figures are from accounts under active management and are anonymised. Last reviewed 2026-08-26.