What you'll learn in this article
- The real timeline of how Discovery campaigns were upgraded to Demand Gen, and why "no action required" was misleading on live accounts
- What actually changed in performance after the migration, observed firsthand, and why the CTR drop was the wrong number to panic about
- What broke, not literally, but in the mental model and reporting I relied on to read a Discovery campaign
- What genuinely was not backward-compatible once the inventory and audience model widened under the same campaign name
- The before-and-after playbook I now run so a forced migration never catches me without a clean baseline again
When discovery campaigns replaced by demand gen stopped being an announcement and became a thing happening inside accounts I actually managed, the lesson arrived fast: the campaigns kept their names, kept their budgets, kept their history, and yet did not behave like the same campaigns afterwards. This is the transition seen from the chair of someone who lived through it on live spend, not from a launch blog. The migration was sold as automatic and painless, and in the narrow sense of "you don't have to click anything" that was true. In the sense that mattered to me, what the campaign reached and how I was supposed to read it, plenty changed, and a few of my own assumptions broke quietly. If you want the platform-level context first, the primer on what Demand Gen campaigns are is worth reading before this account of what the switch felt like.
The timeline, and why "no action required" was misleading
The factual spine is simple. Google announced Demand Gen as the successor to Discovery, the upgrade rolled out through late 2023, and the remaining Discovery campaigns were automatically moved to Demand Gen between January and March 2024. On the accounts where I had no dedicated Google rep, the switch simply happened, and Google kept the historical data and learnings attached to the upgraded campaigns. On paper, the cleanest migration I have seen Google run.
The misleading part is the phrase "no action required." It is accurate about the mechanics and wrong about the consequences. Nobody had to rebuild a campaign, but the campaign that came out the other side could serve in places the Discovery version never did, and that single fact reshaped the numbers. So I stopped treating the upgraded campaign as "my old Discovery campaign, now renamed." I started treating it as a new campaign that inherited an old name and an old budget, because that framing is the only one that survived contact with the reporting. When something is google ads demand gen formerly discovery, the "formerly" is doing a lot more work than the word suggests.
What actually changed in performance
The first thing I noticed was a CTR drop paired with an impression jump, and on the first account I read it exactly wrong. My instinct was that the upgrade had degraded the campaign. It hadn't. The inventory had widened, that's all. Discovery lived mostly across Gmail, the Discover feed, and YouTube's feed surfaces; Demand Gen pushed the same campaign into a much bigger slice of YouTube, including skippable in-stream and Shorts, where click behaviour is naturally lower and volume is naturally higher. A lower CTR over a far larger, cheaper impression base is not failure, it's a different mix.
Google's own guidance flags this directly: after migration, metrics can look volatile, especially once YouTube placements become available, and a stark CTR decline does not by itself mean the campaign stopped working. You can read the official framing in the official Google guidance on Demand Gen campaigns. Once I re-baselined against conversions and cost per action instead of the inherited Discovery CTR, the picture calmed down. On most accounts the conversion volume held or improved at a comparable CPA once I gave the algorithm a couple of weeks; the panic had been entirely about reading the wrong metric.
What broke (and it wasn't a setting)
Nothing was deleted out from under me. No budget vanished, no targeting was wiped. What broke was subtler and more annoying: the way I read a campaign. With Discovery I carried a rough surface-by-surface intuition in my head, Gmail does this, Discover does that, the feed behaves like so. After the upgrade, the report in front of me described a campaign serving across a wider, more video-heavy surface set, and my old intuition mapped onto it badly. The dashboard hadn't lied; my mental model had simply gone stale overnight.
The second thing that broke was creative fit. A Discovery setup that leaned on static images was suddenly eligible to serve on video-first surfaces where an image asset is, at best, a weak citizen. The campaign still ran, but it was running image creative through inventory that rewards video, and that mismatch quietly capped performance until I addressed it. This is where I treat the upgraded campaign like any other Demand Gen build and apply the same discipline I'd use when following Demand Gen best practices on a fresh campaign, because the inherited assets were chosen for a narrower world than the one the campaign now lived in.
What genuinely wasn't backward-compatible
Some things did not carry forward cleanly, and these are the ones worth naming precisely. Reporting expectations were the big one: the surface-level view I used to reason about a Discovery campaign no longer matched how Demand Gen reports formats and audiences, so any comparison I tried to draw straight across the migration boundary was muddy. The audience model widened too, with lookalike segments and multi-format logic that a pure Discovery campaign never had, which means the targeting that produced the old numbers is not the targeting producing the new ones, even inside the same campaign shell.
Bidding and creative scope expanded in ways that make a one-to-one before/after dishonest. Demand Gen handed me more formats and a different bidding surface, so the campaign could chase outcomes the Discovery version structurally couldn't. The practical consequence: you cannot benchmark post-migration Demand Gen against pre-migration Discovery as if they were the same product with a new badge. The frequency and exposure behaviour shifts too once YouTube enters the mix, which is why I think about it the same way I do the wider trade-offs between Demand Gen and a Video campaign, because the same video inventory drives both stories. Backward compatibility, in the honest sense, did not survive the surface expansion.
What survived the migration cleanly
It wasn't all breakage, and it would be unfair to paint it that way. The single best-behaved part of the whole transition was the historical data and the bidding learnings, which carried over rather than resetting to zero. The algorithm did not start cold, the conversion history stayed attached, and that continuity is the main reason most campaigns settled within a couple of weeks instead of relearning from scratch. Credit where it's due: Google got the hard part, not losing the learning, right.
What also survived was the core intent of the campaign, mid-funnel demand across Google's visual feeds, and the budgets transferred without drama. So the honest summary is split: the data layer survived intact, and the interpretation layer did not. The numbers were trustworthy; the assumptions I had built on top of the old numbers were the casualties. Knowing which side of that line a given thing falls on is most of what made the difference between a smooth account and a chaotic one, and it informs how I now structure the wider account-level Google Ads strategy around campaigns I expect Google to keep changing under me.
The before-and-after playbook I run now
The migration taught me one durable habit: capture the baseline before the platform changes it for you. Before any forced upgrade, I now export the last 30 to 90 days of CTR, CPA, conversions, and conversion value, and I screenshot the Discovery-era report so I have a clean before-picture that the upgrade can't muddy. Once the campaign becomes Demand Gen, that old comparison degrades fast, so the export is the only honest reference point I'll have left.
The second habit is creative readiness. Instead of letting an image-led setup serve as-is across video surfaces, I prepare video and multi-format assets ahead of the switch so the campaign isn't running weak creative through inventory that demands strong creative. Then, after the upgrade, I deliberately hold judgement for roughly two weeks and re-baseline against conversions, not the inherited CTR. The migration will happen whether I prepare or not; the only thing I actually control is whether I walk in with a baseline and a creative plan, or walk in blind and spend a month rationalising a CTR drop I should have expected.
The practical takeaway
If you remember one thing from someone who sat through this on live spend: the Discovery to Demand Gen move was automatic in the mechanics and anything but automatic in the consequences. Nothing got deleted, the data and learnings carried over cleanly, and the budgets transferred fine. What changed was the inventory the campaign could reach, and with it the metrics, the creative requirements, and the way the report had to be read. The CTR drop was a distraction; the surface expansion was the real story.
So I don't treat a forced platform migration as a non-event anymore. I treat it as a new campaign wearing an old name, I capture the baseline before it lands, I plan creative for the wider surfaces, and I re-baseline against conversions after. Do that, and "discovery campaigns replaced by demand gen" becomes a managed transition instead of a month of staring at numbers that look broken but aren't.