Performance Max: Setup, Asset Groups and Audience Signals

by Francis Rozange | Apr 4, 2026 | Google Ads

Lede: one campaign, the entire Google network

Performance Max is the campaign type that ate Google Ads. In 2026, on most retail and e-commerce accounts we audit, it now absorbs 70% or more of paid search budget. The reason is structural: a single Performance Max campaign serves Search, Shopping, Display, YouTube, Discover, Gmail, and Maps from one asset library, one audience signal stack, and one budget. You stop thinking in channels. Google’s machine learning decides where the next impression goes.

That consolidation is also where most accounts fail. Performance Max rewards advertisers who feed it correctly: clean conversion tracking, a Merchant Center linked properly, asset groups built around themes rather than dumped into one bucket, audience signals treated as hints not as targeting, and a learning period the advertiser does not interrupt. It punishes everyone else with quiet budget waste, the kind that does not show up as a single broken metric but as a slow erosion of return on ad spend over six weeks.

This guide walks through the 2026 setup, end to end. What Performance Max actually is, how to pick a campaign goal, why Merchant Center matters for retail, how to structure asset groups, the full asset library Google expects, the difference between audience signals and audience targeting, URL expansion and brand controls, the negative keyword controls that arrived in 2023 and 2024, the 6-week learning period, the 3x target CPA budget rule, how to read insight cards, and what the campaign now looks like inside a 2026 budget mix.

What Performance Max actually is

Performance Max is a goal-based campaign type that runs across all Google inventory in a single unit. According to Google Ads Help, one Performance Max campaign can serve Search text ads, Shopping listings, responsive display ads, YouTube in-stream and Shorts ads, Discover feed ads, Gmail promotions, and Maps placements. There is one daily budget, one bidding strategy, and one set of conversion goals.

Mechanically, the campaign does not work like a Search campaign with extra surfaces bolted on. There are no keywords as targeting in the classic sense. There is no manual placement list. You upload a creative asset library and (for retail) a product feed, you set a goal and a bid strategy, you provide audience signals as hints, and Google’s ML model decides which user gets which creative on which surface. The advertiser controls the inputs (what to feed it), the constraints (what not to bid on), and the goal (what to optimize for). Everything between the input and the conversion sits inside a black box that Google has slowly opened over the past two years.

That opening matters for 2026. As Search Engine Land documented, Google rolled out channel-level reporting through 2024 and 2025, so you can now see how spend and conversions break down across Search, Shopping, YouTube, Display, Discover, and Gmail. Combined with search terms insights, asset performance ratings, and the audience insight reports, the campaign is far more inspectable than the original 2021 release. It is not transparent in the Search Network sense, but you are no longer optimizing in the dark.

Campaign goal selection: pick one objective, mean it

The first decision in the campaign creation flow is the business objective. Performance Max offers Sales, Leads, Local Store Visits and Promotions, and Awareness and Consideration. The choice steers the bid strategy options and the conversion actions that the campaign considers eligible.

Sales is the default for retail and e-commerce. It pairs with Maximize Conversion Value or Target ROAS bidding and uses your Merchant Center feed as the spine of the campaign. Most ecommerce Performance Max campaigns we run live in this mode.

Leads is the right pick for service businesses, B2B, and any account where the conversion is a form fill, a call, or a qualified lead rather than a transaction. It pairs with Maximize Conversions or Target CPA. The trap here is letting low-quality micro-conversions (newsletter signups, PDF downloads) pollute the signal: if your primary conversion is “Demo Request” and you also count “Newsletter Signup” as primary, the algorithm chases the cheapest of the two, which is almost never the demo.

Local Store Visits and Promotions targets brick-and-mortar businesses with verified Google Business Profile locations. It optimizes for store visits, calls from Maps, direction requests, and local actions. Used correctly it is a strong replacement for the old Local campaigns Google deprecated.

Awareness and Consideration is the newest objective, rolled out for advertisers who use Performance Max more like a Demand Gen extension. It optimizes for impressions and reach rather than conversion value, and most performance accounts should ignore it.

Search Engine Journal coverage stresses one rule that we see violated in nearly every account audit: pick one primary conversion goal per campaign and set the secondary goals as observation only. If you mix Purchase, Add To Cart, Lead Form, and Newsletter Signup as primary, the algorithm balances among them and your blended cost per real purchase quietly drifts upward.

Merchant Center linking for retail Performance Max

For Sales-objective Performance Max with a product feed, Google Merchant Center is the foundation. The product feed is no longer a side input, it is the core of the campaign. Google Merchant Center documentation requires a verified and claimed website, valid product data including GTINs where applicable, accurate availability and pricing matched to the destination page, and structured data that does not contradict the feed.

Linking Merchant Center to Google Ads is done from the Tools menu under Linked Accounts. Once linked, you select the Merchant Center account and the country of sale during the Performance Max creation flow. The campaign can then use the entire feed or a subset filtered by listing groups (by category, brand, item ID, custom label, product type, or condition).

The custom label fields are the lever here. Tinuiti’s published case studies on Performance Max optimization repeatedly call out custom labels as the single most underused tool in retail PMax. Use them to tag products by margin tier, by stock level, by best-seller status, by seasonality, and by price band. Then build asset groups or separate Performance Max campaigns that filter on those custom labels. A high-margin best-seller running at 30% gross margin should not share an asset group, and arguably should not share a campaign, with a clearance item running at 5% margin.

Feed quality drives 70% to 90% of retail Performance Max performance. Feed errors that look small (missing GTIN, incomplete title, wrong availability, mismatched price) suppress impressions before the algorithm ever gets a chance to optimize. Audit the Merchant Center diagnostics tab weekly during launch and monthly thereafter.

Asset group structure: one per theme, not one per campaign

Asset groups are the structural unit inside a Performance Max campaign that organize creative around a theme or audience. A campaign holds up to 100 asset groups. Each asset group holds its own set of headlines, descriptions, images, videos, logos, business name, audience signals, and final URLs.

The most common mistake is using one asset group per campaign. The reasoning sounds logical (consolidate signal, let the algorithm learn faster) but it is wrong for any account selling more than one product line or serving more than one audience. With one asset group, all creative competes against itself across all queries and all audiences. A women’s running shoe headline ends up paired with a men’s hiking boot search. The algorithm learns slower because it cannot match creative to context.

The right structure is one asset group per theme. A theme is a coherent slice of your business: a product category (Men’s Athletic Shoes), an audience segment (Premium Customers), a use case (Office Furniture for Remote Workers), or a seasonal angle (Holiday Gifts). Each theme gets its own creative library, its own URL set, and its own audience signals. The algorithm then has clean buckets to optimize within.

For most accounts, three to seven asset groups per campaign is the right range. Below three, you are leaving organization on the table. Above seven, the asset groups start fighting for impressions inside the same campaign budget and per-group conversion data thins out.

A workable e-commerce structure: Asset Group 1 covers Men’s Athletic Wear with male-targeted lifestyle imagery, headlines emphasizing performance, and a customer match list of male buyers. Asset Group 2 covers Women’s Athletic Wear with parallel female-targeted creative. Asset Group 3 covers Accessories at lower price points with bundle-focused messaging. Each group filters on the corresponding listing groups in the product feed via custom labels.

A workable B2B structure: Asset Group 1 targets Enterprise prospects with case-study videos and ROI headlines pointed at a high-value demo landing page. Asset Group 2 targets SMB prospects with self-serve trial messaging pointed at the pricing page. Asset Group 3 targets existing customer expansion with feature-launch creative pointed at upgrade flows.

Asset requirements: the 30+ piece library Google expects

Each asset group has minimum and maximum asset counts that determine ad strength. Filling the maximums is not vanity, it is what gives the algorithm enough combinations to find the winning ones.

The asset spec for one Performance Max asset group, per Google Ads Help:

Headlines: up to 5 short headlines (30 characters max each), with at least 3 required
Long headlines: up to 5 long headlines (90 characters max each), with at least 1 required
Descriptions: up to 5 descriptions (90 characters max), with at least 1 short description (60 characters max) and 1 longer description required
Business name: 1 required (25 characters max)
Images: 1 to 20 images, with multiple aspect ratios required (1.91:1 landscape minimum 1200×628, 1:1 square minimum 1200×1200, 4:5 portrait recommended)
Logos: 1 to 5 logos, with 1:1 square required and 4:1 landscape recommended
Videos: 0 to 5 videos. If you do not provide a video, Google auto-generates one from your other assets, which is rarely as effective as a real video
Call to action: 1 selected from a fixed list (Shop Now, Learn More, Sign Up, Get Quote, etc.)
Final URL: 1 required, plus optional final URL paths

Practically, a single well-built asset group has at least 5 headlines, 5 long headlines, 4 to 5 descriptions, the business name, 10 to 15 images split across the three aspect ratios, 3 to 5 logos, and 1 to 3 videos of varying lengths. That is a 30-piece minimum library per group. Multiply by your asset group count and you understand why Performance Max is creative-hungry in a way that legacy Search campaigns were not.

Headlines should not be variations of the same sentence. Each one should attack a different angle: price, quality, speed, brand, range, social proof, guarantee. The algorithm rotates and combines them with descriptions across surfaces, so redundant headlines waste slots.

Images should mix product-on-white shots (essential for Shopping placements), lifestyle shots (essential for Discover and YouTube), and contextual shots showing the product in use. WordStream’s testing data shows lifestyle imagery typically outperforms product-on-white for click-through rate but underperforms it for conversion rate, with a 60-40 lifestyle-to-product mix usually delivering the best blended ROAS.

Videos do not need to be expensive. Phone-shot 15-second product demos and unscripted customer testimonials regularly outperform polished brand films in Performance Max. Provide a 15-second short, a 30-second mid-length, and where possible a 60-second longer-form video. If you skip videos entirely, the auto-generated stitched-together video runs on YouTube placements and typically performs at half the engagement of a real one.

Audience signals versus audience targeting

The audience signal concept confuses people because the name lies. An audience signal is not an audience target. Google’s documentation is explicit: audience signals are hints to the algorithm about who is likely to convert. They influence the model’s exploration, but they do not constrain delivery. The campaign will serve to users outside the signal if the algorithm predicts they will convert.

This is the single most misunderstood feature of Performance Max. Advertisers who treat audience signals as targeting (as they would in a Display or YouTube campaign) become frustrated when impressions land on users who do not match. Advertisers who treat audience signals as exploration hints unlock the campaign’s actual capability: finding incremental converters who look nothing like your stated audience but who behave like converters.

The audience signals to use, ranked by impact:

Customer Match is the highest-impact signal. Upload a hashed list of your existing customers (emails or phone numbers). The algorithm uses this list as a similarity seed. Tinuiti has documented case studies where Customer Match upload alone drove 25% to 35% ROAS improvement within three weeks for retail accounts with sufficient list size. The list should be at least 1,000 hashed records to activate, and ideally segmented to your highest-LTV customers (top 25% by lifetime value) rather than everyone who ever bought.

Custom segments are audiences you build from search keywords, URLs, or app interests. A custom segment of “users who searched for [competitor brand] in the last 30 days” gives the algorithm a strong commercial-intent signal. Custom segments are the closest thing Performance Max has to keyword targeting and are particularly powerful for B2B and high-consideration purchases.

Website visitor segments built from your remarketing tag: all site visitors, product page visitors, cart abandoners, past converters. These are easy wins because the data already exists in your account.

In-market segments are Google’s first-party data on users actively researching a category (running shoes, business software, life insurance). They provide modest improvement on their own (5% to 8% ROAS lift in our experience) but compound when stacked with Customer Match.

Demographics (age, gender, household income, parental status) are weak signals individually but useful as exclusions for products that genuinely should not serve outside a demographic.

Allow two weeks for new audience signals to integrate. The algorithm rebalances exploration when a signal is added, and short-term performance often dips before recovering and improving. Pulling a signal at day three because performance dropped is the most common self-inflicted Performance Max wound.

URL expansion controls

Final URL expansion is the feature that lets Google substitute your designated final URL for a more relevant page on the same domain when the algorithm predicts a better match. It is on by default for Sales-objective campaigns. For most retail accounts with deep, well-structured product catalogs, leaving expansion on is correct: the algorithm sends a user searching for “men’s trail running shoes” to your trail-running category page rather than your homepage, and conversion rate goes up.

For sites with weak internal navigation, thin category pages, or content quality issues, URL expansion can hurt. The classic case is a furniture e-commerce site that saw conversion rate drop 12% after launch because URL expansion sent users to broad category pages that lacked the specific product they had searched for. Disabling expansion (Campaign Settings > Final URL expansion > Off) recovered the loss.

The middle ground is leaving expansion on but adding URL exclusion rules. You can exclude specific URLs or URL patterns (anything containing /blog/, /support/, /careers/) from expansion eligibility. This is the right setup for sites that have good commercial pages but also a content marketing footprint that should not be served as a destination.

Brand inclusion and exclusion: the 2024 update

Brand controls in Performance Max evolved sharply in 2024. Search Engine Land tracked the rollout of campaign-level brand inclusion and exclusion lists, which moved Performance Max from a campaign type with almost no brand control to one where brand traffic can be precisely steered.

Brand exclusions let you prevent your ads from serving on searches for specified brands. The use case is competitor protection: a luxury watch retailer can exclude searches for direct competitor brands so budget is not spent on users with strong intent toward another brand. Brand exclusions cover the brand name and its known misspellings and variants.

Brand inclusions let you tell the campaign to focus only on searches for specified brands. The primary use case is reseller scenarios: a retailer that carries Brand A, Brand B, and Brand C can run a Performance Max campaign that includes only those three brands and skips generic category searches. This is also useful for brand-defense campaigns where you want to bid on your own brand searches without accidentally chasing generic queries.

Brand lists are managed at the account level under Tools and Settings > Shared Library > Brand Lists, then attached to the campaign in the Performance Max settings. The lists update as Google maintains the underlying brand database, so you do not have to manually track misspellings.

Negative keywords: 2023 account-level, 2024 expansion

Performance Max negative keyword controls have a complicated timeline that determines what is possible in 2026.

Originally Performance Max had no negative keyword support. You could ask Google support to add account-level negatives manually, which was operationally absurd at scale.

In 2023 Google rolled out account-level negative keywords for Performance Max, applying to all PMax campaigns in the account. As Search Engine Land reported, this finally let advertisers exclude obvious waste terms (free, cheap, used, jobs, careers, lyrics) without going through support.

In January 2025 Google expanded the controls again with campaign-level negative keywords available to all advertisers, with up to 10,000 negatives per campaign (matching the Search campaign cap). This is the meaningful unlock for 2026: you can now run separate Performance Max campaigns with separate negative lists, the same way you would manage Search campaigns. A clearance-product PMax campaign can permit “cheap” and “discount” while a premium-product PMax campaign blocks them.

Practical negative keyword stack for a typical retail Performance Max in 2026:

Account-level negatives for hard-blocks: jobs, careers, lyrics, wikipedia, reddit, free, DIY, instructions, manual
Account-level brand-protection negatives: any internal terms you do not want anywhere
Campaign-level negatives for the specific PMax: irrelevant categories your campaign does not serve, low-margin product types you want to suppress, geographic terms that fall outside your shipping zone
Brand exclusion lists for competitor brands

Even with all this, treat negatives as directional hints. Performance Max will still occasionally serve on terms in the list if it predicts very high conversion probability. The hint is strong, but it is not absolute.

The 6-week learning period

Performance Max enters a learning phase the moment you launch and exits it (or stabilizes) somewhere between 4 and 6 weeks later, depending on conversion volume. Google’s guidance says the algorithm needs roughly 50 conversions of historical data inside the campaign to fully calibrate.

What this means in practice:

Days 1 to 7 are pure exploration. Performance is noisy. Cost per conversion can be 2x to 3x your eventual stable level. Do not panic.

Days 7 to 21 are partial calibration. The algorithm starts identifying winning combinations. ROAS begins to trend toward target. Resist the urge to cut budget, swap creative, or change bid strategy.

Days 21 to 42 are the stabilization window. By the end of week 6, on a properly funded campaign, performance should be at or above your target metric. If it is not, the diagnosis is structural (feed quality, asset quality, audience signals, conversion tracking) rather than an algorithm problem.

The single largest cause of Performance Max failure is interrupting the learning period. Pausing a campaign at day 14 because ROAS is below target, or cutting budget by 50%, resets the learning. The next launch starts at day zero again. Many accounts cycle through three or four interrupted learning periods before realizing the algorithm never had a chance to stabilize.

Substantive changes during learning trigger a partial reset. Substantive includes: changing bid strategy, swapping the primary conversion goal, cutting budget by more than 20%, removing all assets in an asset group. Adding new creative, adding new audience signals, and adding new negatives are not full resets.

Starting budget recommendations: 3x target CPA

Google’s recommended starting daily budget is 3x your target cost per acquisition. The math is simple: at 1x target CPA daily budget, you generate roughly one conversion per day, which gives the algorithm 30 conversions across the 30-day learning period. That is the absolute minimum for the model to function. At 3x, you generate three conversions per day, 90 across the learning period, and the algorithm has enough signal to actually optimize rather than just survive.

Concretely: if your target CPA is 50 USD, the starting daily budget is at least 150 USD. If your target CPA is 20 EUR, the starting daily budget is at least 60 EUR. If your target ROAS is 4:1 with an average order value of 100 USD, your effective target CPA is 25 USD, and the daily budget should start at 75 USD or above.

For ROAS-bidding campaigns, the equivalent rule is to fund the campaign so that target conversion volume reaches at least 30 conversions per 30-day window, ideally 50 to 100. If your target ROAS implies a CPA so high that 3x daily budget is unaffordable, the diagnosis is that Performance Max is the wrong campaign type for your account at its current scale, not that the rule is wrong.

Underfunded Performance Max campaigns are the second-largest source of poor results we see in audits. A 25 USD daily budget on a 50 USD target CPA campaign delivers what looks like inconsistent performance but is actually just statistical noise from too few daily conversions for the algorithm to read.

Insight cards interpretation

The Insights tab in Google Ads has become the primary diagnostic tool for Performance Max in 2026. The cards rotate based on what the algorithm has detected, and reading them correctly distinguishes operators from button-clickers.

Search terms insights show the actual queries triggering your ads, grouped into themes. Treat this as a continuous source of negative-keyword candidates and as a sanity check that audience signals are functioning. If a “premium customer” asset group is serving on “free” and “cheap” themes, your audience signal is being overridden by feed and asset content.

Asset group performance rates each asset (headline, description, image, video) as Best, Good, or Low. Replace Low-rated assets every 4 to 6 weeks. Best-rated assets become templates for new asset variations.

Audience insights reveal which audience segments and signals contributed to conversions. This is also where you discover audience overlap (the same users converting from multiple signals) and audience expansion (the algorithm finding converters outside your stated signals).

Channel performance shows the spend and conversion split across Search, Shopping, YouTube, Display, Discover, Gmail, and Maps. Optmyzr’s analysis of cross-account channel reporting shows wide variance: some retail accounts run 60% of conversion value through Shopping and 25% through Search, while others run 40% through YouTube. The channel mix is account-specific, and the only way to influence it is through asset quality (more videos shifts spend toward YouTube, better feed shifts spend toward Shopping).

Bid strategy and budget insights flag whether your campaign is budget-constrained, whether your target ROAS is too aggressive for the available demand, and whether the bid strategy is in learning. Read these weekly.

Performance Max in the 2026 budget mix

The structural reality of Google Ads in 2026 is that Performance Max accounts for the majority of paid search spend on most e-commerce and retail accounts. Tinuiti’s quarterly benchmark reports consistently show 65% to 75% of total Google Ads spend on retail accounts flowing through Performance Max, with the remainder split between branded Search, generic Search, and Demand Gen.

The campaign mix that works in 2026:

Branded Search stays separate from Performance Max. Bidding on your own brand terms inside PMax dilutes the campaign’s signal and makes incrementality measurement impossible. Run a dedicated Search campaign on brand keywords with manual or Maximize Clicks bidding.

Performance Max handles the bulk of acquisition: non-brand commercial intent, Shopping inventory, YouTube and Discover demand. One to three campaigns depending on margin tier, brand, or geography segmentation.

Generic Search remains useful for high-intent non-brand keywords where you want manual control over copy and landing page. Especially relevant for B2B and for retail categories where the search query already implies a strong product match.

Demand Gen covers upper-funnel YouTube, Discover, and Gmail when you want to drive demand creation rather than conversion harvesting. Pairs well with Performance Max because demand created by Demand Gen flows into Performance Max as remarketing audiences.

Shopping standard is now mostly redundant for accounts running Performance Max with the same feed. Some advertisers maintain a Standard Shopping campaign at low priority as a fallback for products excluded from Performance Max, but the use case is shrinking.

Common mistakes that drain Performance Max budget

One asset group per campaign. Single-bucket structures prevent the algorithm from matching creative to context. Build three to seven themed asset groups instead.

Multiple primary conversion goals. Pick one primary goal. Demote everything else to observation. The algorithm cannot optimize for “Purchase OR Lead Form OR Newsletter Signup” simultaneously.

Underfunded budget. Below 3x target CPA, the campaign never accumulates enough conversion data to optimize. Fund it correctly or do not run it.

Interrupting the learning period. Pausing or substantially modifying the campaign before week 6 resets learning. Be patient.

Treating audience signals as targeting. They are hints. The campaign will serve outside the signal. That is the feature, not the bug.

Ignoring the product feed. Feed quality drives 70% to 90% of retail PMax performance. A clean Merchant Center is worth more than 30 hours of asset optimization.

No videos. The auto-generated video Google stitches together is meaningfully worse than even a phone-shot real video. Provide at least one per asset group.

Stale creative. Refresh 20% to 30% of assets every 4 to 6 weeks during the active phase. Stale creative drives ad fatigue, even on a black-box campaign type.

No negative keywords. The 2023 and 2024 expansions made negatives one of the highest-leverage Performance Max controls available. Use them.

No URL expansion control. Either confirm expansion is sending traffic to good destinations, or disable it, or add URL exclusion rules. Do not leave it on default and never check.

Conclusion: feed it correctly, then leave it alone

Performance Max is the campaign type that demands the most preparation and rewards the least intervention. The work is front-loaded: clean tracking, a properly linked Merchant Center, asset groups built around real themes, a 30-piece asset library per group, audience signals that hint without trying to constrain, URL and brand controls that match how your business actually works, negative keywords that block obvious waste, a budget that respects the 3x target CPA rule, and a learning period the advertiser does not interrupt.

Do that, and the campaign compounds. Do not, and Performance Max becomes the most expensive way to discover that your product feed is broken.

In 2026, with Performance Max absorbing the majority of paid search budget on most retail accounts, the operators who treat the campaign as an input optimization problem outperform the ones who treat it as a magic box. Feed it correctly. Set the constraints. Let it run for six weeks. Then read the insights, refresh 25% of the creative, and let it run for six more.