.png)
Here's the mistake I see most often: advertisers treat Demand Gen and Performance Max as interchangeable because both are AI-driven, both hide most of the levers, and both are what Google's rep pushes when your Search account plateaus. They are not interchangeable. They serve different jobs, show ads in different places, and reward completely different setups. Pick the wrong one and you'll spend three weeks and a few thousand dollars learning that the hard way, watching a dashboard tell you conversions are up while your actual pipeline stays flat.
So before I get into structure, the 30-second answer. Performance Max is a conversion-harvesting machine. You point it at people already close to buying — across Search, Shopping, YouTube, Display, Gmail, Maps, Discover, all of it — and let Google chase the cheapest conversion wherever it can find one. Demand Gen is the opposite motion. It lives on YouTube, Discover, and Gmail, it's visual and social-feeling, and its job is to create interest in people who weren't searching for you yet. One captures existing demand. One manufactures it. If your goal is squeezing more sales out of an audience that already wants what you sell, PMax. If your goal is filling the top of the funnel with people who've never heard of you, Demand Gen. Most accounts I've worked on need both, running deliberately, not one bleeding into the other.
That last part is where the money leaks. Run them without a plan and PMax will quietly eat the conversions Demand Gen worked to create, then take credit for them in the report. I'll get to how to stop that. But first, what each of these things actually is in 2026, because both have changed enough that advice written even a year ago is half wrong now.
Performance Max started life in 2021 as Google's replacement for Smart Shopping and Local campaigns, and for the first couple of years it was a black box you fed a budget and a feed. It's grown up since. You now get campaign-level negative keywords, brand exclusions, asset-group-level reporting, channel performance data, and search-term insights that are still incomplete but no longer useless. What hasn't changed is the core deal: you hand Google a feed, a set of assets, and a conversion goal, and it decides where and to whom your ads show across every Google surface at once. You're not buying placements. You're buying outcomes and trusting the machine on everything in between.
Demand Gen replaced Discovery campaigns and took the good parts of what used to be YouTube's separate video action campaigns. Think of it as Google's answer to paid social. It runs on YouTube (in-feed and Shorts), Discover, and Gmail, it's built for scroll-stopping image and video creative, and it leans on lookalike audiences and detailed interest targeting the way Meta does. Crucially, it will not put your ad in the Search results. Nobody typing your product into the search bar sees a Demand Gen ad. That single fact should shape how you think about it: Demand Gen reaches people during the parts of their day when they aren't looking for you.
The inventory difference is the cleanest way to keep these two straight. Performance Max's whole pitch is total coverage: Search, Shopping, Display, YouTube, Gmail, Discover, and Maps, all from one campaign. Demand Gen deliberately fences itself to three visual surfaces — YouTube, Discover, and Gmail. There's overlap on those three. Both campaign types can serve a YouTube in-feed ad or a Gmail promotion. But PMax also owns the two surfaces that actually convert bottom-funnel intent — Search text ads and Shopping listings — and Demand Gen touches neither. If a meaningful chunk of your conversions come from someone typing a query with buying intent, only one of these campaigns can capture that click. That's not a preference. It's a hard line, and it settles most of the 'which one' arguments before you get to targeting.
Here's where the two diverge in a way that matters more than most people expect. Demand Gen gives you real targeting. You choose audiences — custom segments built from search behavior and URLs, your own first-party lists, lookalikes seeded off your converters, plus age, gender, and location. You define who Google is allowed to show your ad to, and Google optimizes within that box. If you know your customer, you can draw the box tightly. That's the paid-social muscle memory: you bring the audience, the algorithm finds the cheapest impressions inside it.
Performance Max does not work that way, and this trips up people coming from Search. In PMax, your audience signals are exactly that — signals, not targeting. You hand Google a list of audiences you think will convert, and Google treats it as a hint about where to start looking. It is free to ignore it the moment it finds conversions elsewhere. You cannot force PMax to only serve your custom segment. What you can do is fence it: campaign-level negative keywords, brand exclusions to stop it poaching brand-search conversions it didn't earn, account-level placement exclusions, and a data feed that constrains what it can show. So the control question comes down to this. Demand Gen lets you steer. Performance Max lets you set guardrails and then gets to drive. If you're the type who needs to know your budget is only reaching a defined audience, that difference alone will tell you which one you'll sleep better running.
One practical note on the audience signal, because it's the single most misused feature in PMax: the quality of your signal determines how fast the campaign exits its rough early phase, even though it doesn't cap who gets reached. Seed it with your actual converter list and high-intent custom segments and PMax finds its footing faster. Seed it with a vague 'people interested in our category' list and you'll pay for a longer, messier learning period. Garbage signal in, slow ramp out.
Strip away the feature lists and the choice comes down to where your demand is. If people are already searching for what you sell, Performance Max captures that intent and Demand Gen literally cannot. If demand for your product is thin — a new category, a problem people don't know has a solution, a brand nobody's heard of — then no amount of PMax will conjure searches that don't exist, and Demand Gen is the tool built to create them. I've watched founders pour budget into PMax for a genuinely novel product and get almost nothing, then blame the campaign type. The campaign type was fine. There was just no existing demand to harvest. That's a Demand Gen problem wearing a PMax costume.
Here's how I'd actually route the common cases:
The cannibalization risk is real and it runs in one direction: Performance Max eating conversions that belong to Search, brand, or the demand Demand Gen created. Left alone, PMax will happily bid on your brand terms, scoop up someone who was going to convert anyway, and report it as net-new. Then it looks like your hero campaign while your actual acquisition engine looks weak. Three fixes handle almost all of it. First, turn on brand exclusions in PMax so it stops poaching brand search — that alone cleans up the most common false-attribution problem. Second, keep a dedicated Search brand campaign running so those clicks land where you can see them honestly. Third, don't point both PMax and Demand Gen at the same retargeting audience with the same goal; give them different jobs — Demand Gen for cold reach and interest-building, PMax for capturing intent — so they're not bidding against each other for the same warm user.
The reporting trap is worse than the auction overlap, honestly. PMax's channel-level reporting improved but it still muddies which surface drove what. If you're running both, don't judge them on last-click conversions inside the Google UI alone. Look at whether total new-customer volume and blended CPA moved when you added the second campaign. If Demand Gen is doing its job creating demand, you'll often see it show up as lift in branded search and in PMax's numbers, not in Demand Gen's own conversion column. Judge the system, not the silo. This is exactly the kind of cross-campaign attribution that manual weekly check-ins tend to miss and that a system watching blended performance around the clock catches faster — but I'll come back to that.
If you've decided Demand Gen fits, here's the architecture I'd build. Resist the urge to over-segment. Demand Gen needs conversion volume to learn, and the fastest way to starve it is to split one modest budget across six tightly-themed campaigns that each get a trickle of data.
Structure Demand Gen by funnel stage and audience type, not by product line. A clean starting build for most accounts is two campaigns: one cold prospecting campaign and one retargeting campaign. Keep them separate because they need different creative, different expectations, and different budgets — mixing a cold audience with your warm site visitors in one campaign lets the algorithm spend most of the money on the easy warm conversions and call the whole thing a success while your prospecting goes nowhere. Inside each campaign, ad groups map to audiences. In the prospecting campaign, that's one ad group per distinct audience type: custom segments built from competitor and category search terms in one, lookalikes seeded off your converters in another. In the retargeting campaign, ad groups split by recency or behavior — all site visitors, cart abandoners, past purchasers for repeat-buy products.
One thing I got wrong for a while: I used to run separate ad groups for every audience idea I had, figuring more granularity meant more control. What it actually meant was every ad group stuck in learning, none of them getting enough conversions to optimize. If a prospecting audience isn't going to see at least a few dozen conversions a week at your budget, fold it into a broader one. Two or three well-fed ad groups beat eight starving ones. Let the algorithm find the pockets inside a bigger audience — that's the one thing it's genuinely good at.
Creative is where Demand Gen is actually won, because the format is visual and the auction rewards assets that stop the scroll. Give each ad group a real range: multiple portrait and landscape images, at least one short video (even a simple 6-to-15-second cut), several headlines and descriptions, and your logo. Demand Gen leans hard on video, so if you only bring static images you're competing with one hand behind your back. Match the creative to the stage — cold prospecting creative should introduce the problem and the brand, not scream a 20%-off code at someone who's never heard of you. Save the discount and the urgency for the retargeting ad groups, where the viewer already knows who you are. For audiences, lead with your first-party data: upload your customer list and build lookalikes off your best converters, not off all traffic. A lookalike of everyone who bounced is a lookalike of people who bounce.
PMax structure is a different discipline because you have fewer levers, so the ones you have matter more. The single biggest decision is how many asset groups and how many campaigns you run, and the default advice — 'one campaign, let Google sort it out' — is only right if your products are homogeneous. If you sell one type of thing at one margin, one campaign is fine. If you sell across categories with wildly different margins or conversion values, separating them lets you set different targets and budgets, which is the only real control you have over where PMax spends.
Asset groups in PMax are theme buckets, roughly analogous to ad groups but blunter. Split them by product category or by audience theme, and give each one its own tailored assets and a matching audience signal — don't let one generic asset group cover your whole catalog, because then every product gets the same creative and the same signal. Your feed is doing more work than your ads here: a clean, complete Merchant Center feed with accurate titles, product types, and custom labels is what actually lets PMax target and segment intelligently. Custom labels are underused. Tag products by margin or by bestseller status and you can build asset groups or campaigns around the ones worth pushing.
Then the two guardrails everyone forgets. First, final URL expansion: leave it on and PMax can send traffic to any page on your site it thinks will convert, including pages you never meant to advertise. For most accounts I'd turn it off or, at minimum, add URL exclusion rules so it can't route paid traffic to your careers page or a blog post. Second, brand exclusions and negative keywords — I covered why above, but structurally, set them at campaign level from day one rather than waiting to notice PMax has been living off your brand terms for a month. These aren't optimizations you get to later. They're setup steps. Skip them and you're paying to find out what PMax does when nobody's fencing it in.
There's no universal split, but there is a sane default. If you're capturing existing demand and just adding reach, put the majority of budget where the intent is — for most ecommerce accounts that's PMax and Search taking 70-80%, with Demand Gen funded as a deliberate top-of-funnel investment at the remainder. If you're in demand-creation mode for a new product, invert it: Demand Gen and awareness lead, and you scale PMax as the demand you're building starts showing up as searches. The mistake is funding Demand Gen out of guilt — a token $500 a month that's too thin to ever exit learning, producing nothing, confirming your suspicion that it 'doesn't work.' Either commit enough budget for it to generate real conversion volume, or don't run it. A starved Demand Gen campaign isn't a small test. It's a slow way to waste money and draw the wrong conclusion.
On measurement: give each campaign a fair window before you judge it. Demand Gen especially takes time, because you're measuring demand you created, which shows up on a delay and often through other channels. Watch blended CPA and total new-customer volume as you add or scale each campaign, not the in-platform conversion column in isolation. If adding Demand Gen lifts your branded search volume and your overall new customers while your blended CPA holds, it's working — even if its own conversion number looks modest. If PMax's conversions climb but your blended new-customer count is flat, it's probably harvesting demand you'd have won anyway, and it's time to check those brand exclusions again. Judge the account, not the campaign.
This is the part that's genuinely hard to do by hand, and I say that as someone who did it by hand for years. Watching cross-campaign cannibalization, catching PMax drifting onto brand terms, spotting a Demand Gen ad group quietly stuck in learning, rebalancing budget between two campaigns that each report their own siloed success — that's not a weekly-check-in job, it's a continuous one. It's exactly the kind of always-on optimization work that a platform like groas is built to run around the clock, with a strategist watching the blended numbers instead of the vanity ones. I'm biased about the tooling because I work on it, but the underlying point stands whoever's doing the watching: these two campaign types only play well together when someone's actually refereeing them.
Can I run Demand Gen and Performance Max at the same time? Yes, and for most established accounts you should — they do different jobs. The catch is the refereeing above: brand exclusions on PMax, separate retargeting pools, and judging them on blended numbers. Run them without that discipline and they'll fight over the same warm users and hand you a report that flatters the wrong one.
Which is cheaper? Wrong question. Demand Gen usually shows a lower cost-per-click because it's top-of-funnel and less competitive, and PMax often shows a lower cost-per-conversion because it's harvesting people already ready to buy. Neither number means one is 'more efficient' — they're pricing different stages of the same funnel. Compare them on what they cost to acquire a genuinely new customer, blended, and the gap usually closes a lot.
Should a brand-new advertiser start with either of these? Rarely. If you have no conversion history and no existing search demand, start with a tight Search campaign to prove the offer converts and to generate the conversion data both AI campaign types need to learn from. Feeding PMax or Demand Gen a cold account with zero signal is the fastest way to a long, expensive learning phase and a conclusion that Google Ads 'doesn't work for you.' Build the signal first. Then let the machines chase it. That order is the whole game.
