What you'll learn in this article
- The dividing line I use between what I optimize by hand and what I deliberately leave to the AI, and the reasoning behind it
- Why conversion data quality is the optimization that does the most work, even though no checklist lists it first
- How I handle structure, budget, creative and bid targets and the exact thresholds I move within
- How google ads demand gen vs performance max optimization actually differs in day-to-day practice
- The levers I refuse to touch because every time I override the optimizer it costs me efficiency
Most lists of the top demand gen campaign optimization strategies google ads read like a settings tour: adjust bids, add audiences, refresh creative, tweak targeting. The trouble is they never tell you which of those levers you should actually be pulling and which ones you should keep your hands off entirely. After a couple of years running these campaigns on live accounts, the most useful thing I've learned isn't a new lever to pull it's a clear line between what I optimize and what I deliberately hand to the automation, because Demand Gen is an AI-driven, auction-time system and half the "optimizations" people make are just fighting it.
So this article isn't a feature checklist. It's the actual division of labor I work to: what I tune by hand because it changes what the system learns from, what I leave to the algorithm because it reads the auction better than I ever could, and the reasoning behind each call. If you want the platform basics first, the primer on what Demand Gen is covers the ground; here I'm assuming you know what the campaign type does and want to know what a practitioner changes versus what they leave alone.
One framing sets up everything below. In an automated campaign, my job stops being "set the bid for this auction" and becomes "manage the inputs the optimizer learns from." Get the inputs right clean data, enough signal, good creative, sane budget and the AI does the optimization. Reach past those inputs to micro-manage placements or daily bids and you mostly just reset learning. That input-versus-override distinction is the lens for every strategy here.
The dividing line: inputs I own, outputs I leave
If I compress the top demand gen campaign optimization strategies google ads into one rule, it's this: I own the inputs, the algorithm owns the outputs. Inputs are everything that shapes what the system learns from conversion tracking, goal choice, budget level, ad group structure, creative quality and refresh, and the bid target. Outputs are everything downstream which placement serves, which user sees the ad, the auction-time bid, and how spend splits across YouTube, Shorts, Discover and Gmail. I optimize the first list aggressively and I leave the second list almost entirely alone.
The reason is mechanical, not philosophical. Demand Gen uses auction-time Smart Bidding, which sets a bid for every individual auction using far more signal than I can see in any report. When I reach in to hand-pick placements or nudge bids by the hour, I'm not adding intelligence I'm subtracting data and forcing a relearn. So the test I apply to any proposed change is simple: does this change the inputs the model trains on, or does it override the model's own decision? If it's the former, it's my job. If it's the latter, it's almost always a mistake dressed up as optimization.
This is also why "do more" is the wrong instinct on Demand Gen. The campaigns that scale in my accounts aren't the ones I touch most they're the ones where I got the inputs right and then resisted the urge to fiddle. The sections below are just that list of inputs, in the order they matter, with the reasoning for how far I move each one.
The optimization that does the most work: conversion data
The highest-leverage thing I optimize isn't visible in the campaign at all it's the conversion signal feeding it. Most Demand Gen underperformance I'm asked to diagnose traces back to a broken, double-counted or too-sparse conversion setup, not to creative or bidding. The auction-time bidder is only ever as good as the data it learns from, so before I touch anything else I confirm tracking is firing cleanly and counting what I think it counts. Scaling on a broken signal just buys bad outcomes faster.
Beyond "is it firing," I optimize the structure of the signal. I point the campaign at a deep action I actually care about a purchase or a qualified lead but I also tag shallower events across the funnel, like add-to-cart or site visit, and set them as non-biddable at the account level. That gives the system more events to learn from during the ramp without letting cheap actions hijack the bid. It's the closest thing to a free optimization I know: it speeds learning and stabilizes the bidder, and it costs nothing but careful setup. This is where Demand Gen and the rest of the account meet, because the same clean signals that power Smart Bidding across campaign types are what let Demand Gen ramp at all.
If I had to rank a single optimization above every creative or bidding tweak, this is it. It's also the one no "top strategies" listicle puts first, because it isn't a button in the campaign it's plumbing. But on the accounts I manage, fixing the data has moved CPA more than any asset refresh, and it's always the first thing I check before I conclude a campaign is "just not working."
Structure and budget: the two inputs I get right before launch
The next two inputs I optimize are structure and budget, and both are about giving the algorithm enough signal density to learn. On structure, I consolidate. Google AI learns at the campaign level, so the Search habit of slicing into tight ad groups is actively harmful here each split divides the data the optimizer needs. I combine similar audience themes into broader ad groups and only carve out a separate one when I genuinely need different creative. My practical rule is a signal floor: if an ad group is generating very sparse conversions over a week or more, I merge or broaden it rather than let it limp along, because thin ad groups don't just underperform, they can cause spend spikes when the bidder loses confidence.
On budget, the lever that quietly decides everything is funding the bid strategy properly. Google's own guidance is to set the budget to at least 10 times your target CPA, and in my experience an underfunded budget is the single most common reason a new campaign stalls if the budget caps you at a handful of conversions a day, the system never gathers enough signal to optimize. So I'd rather run one well-funded campaign than three starved ones. The same consolidation logic that applies to ad groups applies to budget: concentrate it so the AI has something to read. This is the foundation every Demand Gen bidding strategy sits on, because no bid strategy can optimize a budget too thin to produce signal.
Creative refresh and bid targets: the levers I move slowly
Creative is the input I optimize most actively, because on a discovery campaign it's the part the audience actually responds to and the part I can steer most directly. But I optimize it for coverage and freshness, not for chasing a perfect ad-strength gauge. I keep full ratio and format coverage so the ad can serve everywhere, then I watch for fatigue and refresh before performance decays rather than after. Crucially, I add before I remove and change only a portion of the assets at a time, so the system never loses its whole reference set at once. The reasoning is the same as everywhere else: preserve what the algorithm has learned while feeding it something new to test.
Bid targets are the input I move most carefully of all. Google's optimization guidance is explicit that large changes can amount to a learning reset, so I cap any single adjustment at 15% and usually flex 5 to 10 percent at a time, then let it settle for a few days before moving again. Google's Demand Gen performance guide spells out both the 15% ceiling and the rule to make no changes at all during the initial learning period, and every aggressive jump I've made taught me why: the CPA spikes while the system re-stabilizes, and the spike usually eats whatever extra volume the bigger target bought. With targets, slow genuinely is faster.
Demand Gen vs Performance Max: optimizing two different machines
People ask me about google ads demand gen vs performance max optimization as if they're the same job with a different label, and they aren't. The biggest practical difference is how much creative control you have and what intent you're working with. Demand Gen is creative-led and upper-funnel: I'm optimizing image and video assets, ad formats and audience signals across discovery surfaces, and I can see and steer the creative directly. That makes creative refresh and audience signals my main levers, and it's why I treat Demand Gen as my demand-creation engine rather than a demand-capture one.
Performance Max is a different animal. It spans Search, Shopping, Display, YouTube and more from a single campaign, so I optimize through asset groups, audience signals and for retail the product feed, with far less visibility into where spend actually goes. The optimization mindset shifts from "steer the creative" to "feed the machine good asset groups and trust the cross-channel distribution." In day-to-day terms, I tune Demand Gen by improving what people see, and I tune Performance Max by improving what the system has to work with and then largely standing back. Forcing either to behave like the other is how the comparison goes wrong: they're optimized for different jobs, and the right strategy is to let each do the one it's built for.
That difference also changes how I split budget between them. I lean on Demand Gen when I want efficient reach and creative-driven consideration, and on Performance Max when I want full-funnel coverage and conversion volume across every surface at once. They're complementary engines, not rivals, and optimizing them well means respecting what each one is actually for.
What I deliberately leave to automation
Part of having a real optimization strategy is knowing what not to touch. I don't hand-pick placements or hand-restrict channels unless there's a hard brand-safety reason Google's guidance is to run across all channels unless there's a specific reason not to, and in my experience narrowing the surface reliably raises CPA more than it protects anything. The system distributes spend to whatever channel performs best for my goal and concentrates there over time on its own; overriding that just removes options it would have used to find cheaper conversions.
I also leave the auction-time bid, the audience expansion through optimized targeting, and the day-to-day placement decisions entirely to the algorithm. These are the outputs in my inputs-versus-outputs rule, and every time I've reached in to override them I've paid for it in efficiency. The last thing I leave alone is the campaign itself during learning: I don't judge performance or make changes until it's cleared the learning period and gathered enough conversions, because acting on noise is worse than waiting. Restraint isn't passivity here it's a deliberate strategy, and on Demand Gen it's one of the most reliable optimizations I have.
The practical takeaway
If you take one thing from these top demand gen campaign optimization strategies google ads: stop optimizing the outputs and start optimizing the inputs. Get the conversion data clean, consolidate structure, fund the budget to roughly 10x your target, keep creative fresh, and move bid targets in small steps then hand placements, channel mix and auction-time bidding to the AI and leave them there. The whole game is feeding the optimizer well, not out-guessing it.
The version of optimization that holds up on live accounts is more about discipline than activity. Own the inputs, respect the learning period, change less and more slowly, and treat Demand Gen and Performance Max as the different machines they are. Do that and the automation does the heavy lifting it's genuinely good at. Override it out of a need to feel in control, and you'll have a busy account that quietly underperforms a patient one.