Demand Gen and Performance Max are not competitors. Performance Max closes demand across all of Google’s inventory from one product feed. Demand Gen creates demand on YouTube, Shorts, Discover and Gmail with creative and audiences you control. Most ecommerce brands should build Performance Max first, then add Demand Gen once new customer cost is measured and holding.
What each campaign actually does
Google describes Performance Max as one campaign that serves across YouTube, Display, Search, Discover, Gmail and Maps. You supply conversion goals, an optional ROAS or CPA target, a set of assets, audience signals and a product feed, and Google’s system decides where each ad appears and who sees it. The controls that matter to an ecommerce brand are the ones Google lists as ways to keep the campaign off placements you never wanted: brand exclusions, negative keywords and search themes.
Demand Gen is tighter on inventory and looser on control. Google says it runs across YouTube including Shorts, Discover, Gmail, Maps and the Display Network, reaching up to 3 billion monthly users, with conversion, value based and maximum clicks bidding. In exchange for the narrower inventory you get to choose placements and formats, build lookalike segments from your purchasers and site visitors, test creative concepts against one another, and attach a product feed in a format that behaves more like a paid social ad than a Shopping tile.
The shortest way to hold the difference: Performance Max is a closing campaign that will happily spend on your own brand name unless told not to, and Demand Gen is a prospecting campaign that only works when the creative earns attention from people who were not looking for you.
| Question | Performance Max | Demand Gen |
|---|---|---|
| Inventory | YouTube, Display, Search, Shopping, Discover, Gmail, Maps | YouTube and Shorts, Discover, Gmail, Maps, Display Network |
| Main input | Product feed, assets, goals, audience signals | Creative, audiences, lookalike segments, optional product feed |
| Placement control | Low. Reporting shows channels; you cannot choose them | High. You pick the placements and the formats |
| Creative testing | Google assembles the combinations | You test concepts against each other |
| Audience control | Signals only | Lookalike segments, in-market, custom and your own lists |
| Biggest risk | Spending on branded demand you already own | Spending on reach that never turns into customers |
| Judge it on | New customer share and cost, with brand exclusions on | Blended new customer cost and attributed branded searches |
Where the two overlap, and how to stop them paying twice
Both campaigns can serve on YouTube, Discover, Gmail and Display, so an account running both with no rules will often pay for the same person twice. Performance Max reaches a shopper on Discover at lunchtime, Demand Gen reaches her on YouTube that evening, and both record the purchase inside Google’s attribution. Google’s reporting will not tell you which touch did the work, because from Google’s side both of them did.
Three settings remove most of the collision. The first is brand exclusions in Performance Max, kept on permanently, so neither campaign gets credit for people typing your name into the search bar. The second is giving Demand Gen the audiences Performance Max cannot build as well: lookalikes of your best purchasers, people who watched your YouTube content, lapsed customers you want back. The third is splitting the feed by job. Performance Max carries the full catalog, organized into asset groups by margin tier. Demand Gen carries a short list of hero products, each with creative made specifically for it.
Then read both on one blended number rather than on two Google reports. If the account’s blended cost per new customer holds or falls as Demand Gen spend rises, Demand Gen is reaching people Performance Max was missing. If the blended number does not move while Demand Gen’s own report looks strong, the campaign is mostly collecting credit for demand the rest of the account already created, and its budget should go back where it came from.
When Demand Gen earns its budget
Demand Gen earns a place when three conditions hold. The product has to work in short video, because YouTube Shorts and Discover reward creative that stops a thumb, not a product tile with a price under it. The account has to have clean purchase data already, because value based bidding and lookalike segments both learn from your conversions, and they learn the wrong lesson from duplicated or missing ones. And there has to be a plan for measuring a campaign that mostly never gets clicked, because much of what Demand Gen produces shows up later as a branded search rather than as a conversion in its own column.
Google has been adding ways to see that effect. Its January 2026 Demand Gen update introduced attributed branded searches, which report the volume of branded searches a campaign drove on Google and YouTube, and claimed that campaigns including TV screens drive an average of 7% additional conversions at the same ROI. The September 2026 update added one click image ads on Shorts and Gmail and cited Google internal data showing advertisers who added Gmail to Demand Gen saw a 40% increase in conversions at the same ROI. Both numbers come from Google, which sells the inventory, so treat them as the argument for a test rather than as a result you can put in a forecast.
The practical test for an ecommerce brand is short. Run Demand Gen with lookalike segments built from purchasers, three distinct creative concepts for the hero range, a product feed attached, and a defined budget. Then watch two numbers for four weeks: attributed branded searches, and blended cost per new customer across the whole account. If branded searches rise but the blended number does not improve, you bought reach. If both move together, you found a prospecting channel.
When Performance Max earns its budget
Performance Max earns its budget the moment the feed is clean and brand exclusions are on. It is the campaign type that takes a product catalog and finds buyers for it across Shopping, search, YouTube and Display at the same time, and for most ecommerce accounts it is the first campaign built and the last one anyone would switch off. The question is never whether to run it. The question is what it is allowed to do.
Rising costs make the guardrails more important, not less. Triple Whale’s 2026 Google Ads benchmark, built on more than 21,000 brands over the twelve months to July 2026, reports a median cost per purchase of $28.14, up 9.96% year over year, a median CPM of $15.35, up 13.34%, and a median ROAS of 3.27, down 3.54%. For apparel and accessories brands the medians are a 3.99 ROAS and a $25.40 cost per purchase. With conversions getting dearer at that rate, a Performance Max campaign left free to serve on branded searches will report a healthy ROAS assembled largely from customers who were already yours.
The remedy is structural rather than clever. Brand exclusions on. Asset groups split by margin tier, so a high margin product can run to a lower return target than discounted stock. Real creative supplied, rather than images Google assembles itself. And reporting that separates first time buyers from returning ones, because Performance Max will optimize toward whoever converts cheapest, and that is often somebody who bought last month.
Not sure how much of your Performance Max revenue is branded? The free 30 minute growth audit splits it on your own numbers and leaves you with three fixes, whatever you decide afterwards.
How to split the budget between them
There is no fixed ratio, because the right split depends on how much new demand Google can find for your products and on what proportion of your existing Google revenue is branded. The table below is the starting point we use in audits, and it gets adjusted after the first month of data rather than defended against it.
| Situation | Performance Max | Demand Gen | Why |
|---|---|---|---|
| Google spend under $10K a month | 100% | 0% | Too few conversions to feed two learning systems; fix the feed and exclusions first |
| Brand exclusions not yet set | 100% | 0% | Until branded demand is separated out, neither campaign can be read |
| Clean feed, exclusions on, new customer cost measured | 80% | 20% | A Demand Gen test with hero products and lookalike audiences |
| Demand Gen holding blended new customer cost for four weeks | 65% | 35% | Scale Demand Gen in steps while the blended number holds |
| Product sells on video and Meta prospecting is already strong | 60% | 40% | Demand Gen on Shorts and Discover behaves more like paid social than like search |
| Branded share of Google revenue above 50% | Rebuild first | Hold | Google is mostly capturing demand other channels created; fix measurement before adding reach |
The split inside Google follows the same logic as the split between channels. In a plan judged on blended results, Google as a whole competes with Meta, TikTok and every other channel on the cost of a new customer, and the two campaigns inside Google compete with each other on exactly the same terms. One womens fashion brand I work with turned Google into a core channel across January to August 2026: Google accounted for 27% of the combined Google and Facebook budget in that window, up from roughly 10% a year earlier, while blended cost per new customer fell rather than rose. The channel earned that share by showing new customers in the store data, not by posting a good number in its own report.
What the table cannot tell you is where your own account sits, because that depends on what proportion of your Google revenue is branded, and almost nobody knows that figure until someone measures it. It is the first thing worth finding out, and it usually changes the split more than anything else on this page.
How to judge both without trusting Google’s report
Google records every conversion its tag can see inside the attribution window. Branded searches, view through conversions from YouTube, and purchases that Meta or an email started all land in Google’s column. Performance Max blends several inventory types into one reported figure, and Demand Gen can claim view based conversions from videos nobody clicked. Both reports are useful for comparing assets and audiences against each other inside the campaign. Neither is a verdict on the channel, and treating either as one is the most common way an ecommerce account ends up paying Google for its own customers.
The scorecard that survives contact with the bank statement has four lines. New customer share of each campaign’s conversions, read in the store or in Triple Whale rather than in Google. Blended cost per new customer across the whole account, week over week. Attributed branded searches for Demand Gen, so lift that appears as brand search is counted once rather than claimed twice. And contribution margin by product group, because a campaign that mostly sells discounted stock can post a fine ROAS and still lose money on every order. The measurement that produces those four lines sits under tracking and attribution, and building it is the first week of any Google Ads engagement we run, before a single campaign is touched.
A 30 day test plan
Week one is measurement: confirm that the Google tag and enhanced conversions record each order once, that first time and returning buyers can be separated in reporting, and that brand exclusions are live in Performance Max. Week two is the feed and the asset groups: titles rewritten for the products carrying most of the revenue, custom labels for margin and stock, asset groups split by tier. Weeks three and four are the Demand Gen test: lookalike segments from purchasers, three creative concepts for the hero range, a budget agreed in writing, and a daily read on blended cost per new customer beside attributed branded searches.
By day 30 you should know three things. What share of Google revenue is branded. Whether Performance Max still finds new buyers once it cannot serve on your name. And whether Demand Gen moved the blended number or only its own. Sometimes the right call is to keep Performance Max alone for another quarter and put the Demand Gen budget into creative instead. That is a cheaper outcome than running two campaigns that both report brilliantly while acquiring the same customers twice.
Want the split worked out on your own account rather than from a table? Google Ads management at Plaid Testing begins with the audit, and you keep the account, the feed and the data whatever happens next.
Common questions
Yes, and most ecommerce accounts spending more than $10K a month on Google eventually should. The conditions are that Performance Max has brand exclusions on, the feed is clean, and both campaigns are judged on blended cost per new customer rather than on their own reported ROAS, so the overlap on YouTube, Discover and Gmail does not get paid for twice.
Yes, with guardrails. Performance Max is the fastest way to put a product catalog in front of buyers across Shopping, search, YouTube and Display. Without brand exclusions and new customer reporting, its results tend to be built largely from branded demand you already owned, so the guardrails are the difference between a campaign that grows the business and one that only grows the report.
No. Demand Gen gives you placement and creative control on YouTube, Shorts, Discover and Gmail, plus lookalike audiences, but it carries no Shopping or search inventory. Performance Max does. For an ecommerce brand they do different jobs: one creates demand and the other closes it.
Demand Gen usually buys cheaper impressions and Performance Max usually buys cheaper conversions, because Performance Max serves closer to the moment of purchase. Comparing them on cost per conversion inside Google misleads, since Performance Max can claim branded conversions and Demand Gen can claim view based ones. Compare them on blended cost per new customer and on attributed branded searches instead.