Performance Max 2026: The Campaign Structure That Actually Works

performance max is the campaign type everyone runs and nobody trusts. the distrust was earned in 2022 and 2023, when it was a black box that ate brand search and reported a roas you could not check. the controls that fix that have shipped, one at a time, since january 2025. most accounts have not turned them on. this guide is the structure for an account that has.

---

what performance max actually is

performance max is one campaign that serves across search, shopping, youtube, display, discover, gmail, and maps, with google's ai deciding channel, placement, creative combination, and bid against a conversion goal. for a retailer with a product feed it is mostly a shopping campaign: in smec's analysis of more than 4,000 retail performance max campaigns across 500 advertiser accounts, feed-based ads took between 74% and 97% of campaign cost, with a median of 90% (smec, state of performance max 2025).

here is what most guides get wrong: they describe performance max as one product and give you one opinion of it. it is three different campaigns depending on what you feed it. feed plus assets for retail, assets only for lead gen, and a store-goals variant that now includes waze pins. the controls, the failure modes, and the right structure differ for each. this guide is written for retail and lead gen on the web, with the differences called out.

---

the fundamental shift: from black box to steerable

2022 to 2024: no search terms report, no campaign-level negatives without a google rep, no brand exclusions, no channel breakdown, and performance max automatically took priority over standard shopping for the same products. the only levers were budget, target roas, and audience signals nobody could verify.

october 2024: google changed the shopping overlap rule so that "ad rank determines which campaign serves ads rather than automatic performance max prioritization," and shipped asset experiments and final url expansion experiments (google ads help, new performance max features, october 23, 2024).

january 2025: campaign-level negative keywords rolled out to all advertisers, a new customer acquisition goal with high value mode arrived, brand exclusions could be applied to search text ads while keeping shopping ads on branded traffic, and asset group reporting became segmentable (google ads and commerce blog, january 23, 2025).

august 2025: campaign-level negative keyword lists, search themes raised from 25 to 50 per asset group, device targeting, age and gender controls, search terms reporting for all performance max campaigns and through the api, and full asset statistics with conversions, cost, and clicks (google ads help, unlock more visibility and control in performance max, august 7, 2025).

2026: first-party audience exclusions ("you can now exclude specific customer lists from your performance max campaigns"), a budget report inside performance max that projects end-of-month spend, demographic performance breakdowns, and placement reports segmented by network (google ads, new performance max steering and reporting updates coming in 2026).

in plain english: performance max now has most of what a search campaign has. the remaining difference is that you steer it with exclusions and signals instead of keywords and bids.

---

what this broke in existing accounts

1. brand search inside performance max made the campaign look like the best thing in the account

haus ran 3-cell geo experiments on performance max with and without branded search terms. excluding brand drove new customers more efficiently in 100% of experiments, with "a 19% to 60% lower cac" and "a 40% reduction in acquisition cost on average." excluding brand also drove 24% more incremental revenue on average, and in the caraway case study google's reporting overstated performance max by 33% (haus, pmax experiments including vs excluding branded search terms). accounts that scaled performance max on its reported roas scaled a campaign that was partly reselling their own brand demand.

2. campaigns below the conversion floor that never stabilized

smec's dataset puts the floor at "at least 30 monthly conversions" per campaign for stability and "ideally 60 or more." accounts that split one product catalog into six performance max campaigns to "control budget" gave each one a fraction of that and then blamed the algorithm.

3. target roas set from the reported number

median target roas in the smec dataset rose from about 4.7 to 6.0 between 2022 and 2025, and campaigns typically hit 95% to 116% of target. a target set from a roas that included brand and double-counted conversions is a target the campaign will hit by serving exactly the traffic you did not want.

4. standard shopping running alongside with no plan

since october 2024 the two campaign types compete on ad rank for the same product. accounts that kept a standard shopping campaign "as a fallback" now have two campaigns bidding on the same queries with different targets, and the reporting does not say which won.

5. the 2023 asset group setup

one asset group with everything in it, or twenty asset groups with three products each. asset groups are where creative and audience signals attach; they are not budget containers. accounts with too many groups have no group with enough data to rate assets; accounts with one group cannot tell which products or messages are working.

6. no search terms review because there was no search terms report

there has been one since august 2025. most accounts still do not look at it, and the campaign-level negative list stays empty.

7. final url expansion sending traffic to pages that were never meant to convert

final url expansion picks landing pages on its own unless excluded by url rules. blog posts, careers pages, and out-of-stock products have all served as landing pages. the url contains rules and the final url expansion experiments exist to test and constrain this; most accounts have neither.

---

the data

• smec, 4,000+ retail performance max campaigns, 500+ accounts, 2021 to early 2025: performance max cost share peaked "just shy of 82%" of google ads spend in may 2024 and then fell about 6 points; feed-based ads take a median 90% of performance max cost; campaigns need at least 30 monthly conversions, ideally 60+; 95% of campaigns used maximize conversion value as of 2023; median target roas rose from about 4.7 to 6.0; campaigns achieve 95% to 116% of target. (smec, state of performance max 2025)
• haus, 3-cell geo experiments, published august 2024: excluding brand terms lowered cac by 19% to 60%, 40% on average, and was more efficient for new customers in every experiment; excluding brand drove 24% more incremental revenue on average with a 50% win rate on revenue; brands with average order value above $238 drove 27% more incremental revenue including brand; google's reporting overstated one brand's performance max results by 33%. (haus, pmax brand experiments)
• google, april 15, 2026, updated june 11, 2026: dynamic search ads, automatically created assets, and campaign-level broad match settings are moving to ai max; ai max auto-upgrades for automatically created assets and broad match begin september 2026, and the dynamic search ads sunset and auto-upgrade moved to february 2027; "ai max for search campaigns see an average of 7% more conversions or conversion value at a similar cpa/roas when using the full feature suite" (google ads and commerce blog, dynamic search ads are upgrading to ai max).
• google, channel performance reporting: launched september 2024, search partners added november 2024, available in manager accounts with bulk download; waze "promoted places in navigation" inventory added for store-goals campaigns in the us (google ads help, channel performance reporting).

---

what changed in 2026 specifically

customer list exclusions. you can exclude a customer match list from a performance max campaign. this is the missing half of new customer acquisition: the goal tells the campaign to prefer new customers, the exclusion stops it from serving existing ones at all.

budget report inside performance max. end-of-month spend projection per campaign, so pacing is visible without a script.

demographic and network placement breakdowns. performance by age and gender, and placement reports segmented by network. the brand-safety question "where did my ads run" now has a report.

ai max is the search-side sibling. dynamic search ads, automatically created assets, and broad match are being folded into ai max on a september 2026 to february 2027 timeline. performance max stays the cross-channel product; ai max is what your search campaigns become. the question "performance max or ai max" is usually "both, with brand and remarketing carved out of performance max."

unchanged: the conversion floor, the shopping share of spend, and the need to verify the reported roas against something outside google.

---

what actually works now

1. decide what performance max is for before you build it

retail: performance max is your shopping engine for non-brand, new-customer demand. lead gen: it is a prospecting campaign on youtube, display, and search that needs a real lead quality signal, not a form submit. store goals: it is maps and waze. write the job down; every setting below follows from it.

2. carve brand out and keep it out

add a brand exclusion list at the campaign level. for retail, use the format-level control so shopping ads can still serve on branded queries if you want them to while search text ads do not. run a separate brand search campaign with its own budget and a manual or target-impression-share strategy. if your average order value is well above $238, haus's data says including brand can win on revenue; test it with a geo holdout rather than assuming either way. see incrementality testing on a small budget.

3. hit the conversion floor with fewer campaigns

one performance max campaign per distinct goal and margin band, not per product category. if a campaign cannot reach 30 conversions a month, merge it. if it can reach 60, it is stable enough to hold a target roas.

4. set targets from reconciled numbers

use google ads "conversions (by conv. time)" and your store's revenue, after brand is excluded, for the first 30 days at maximize conversion value with no target. then set target roas at what the reconciled number supports. see ga4 and google ads conversion import for the reconciliation.

5. structure asset groups by message, not by budget

three to six asset groups per campaign. each group is one product line or one customer type with its own headlines, images, video, audience signals, and search themes. every group needs enough traffic for the asset ratings to mean something. feed-only for the first two to four weeks on a new retail campaign, then add assets and read the asset experiment.

6. use the controls that shipped

• campaign-level negative keyword list, reviewed against the search terms report every two weeks.
• search themes up to 50 per asset group, checked against the usefulness indicator and the source column so you know which themes drive incremental queries.
• url contains rules and a final url expansion experiment, so the campaign cannot land traffic on pages you did not choose.
• device and age exclusions where the data supports them.
• new customer acquisition goal with high value mode, fed by a refreshed customer match list of real buyers, plus a customer list exclusion of existing customers on the prospecting campaign.

7. reconcile standard shopping

if you still run standard shopping on the same products, give the two campaigns different jobs (standard shopping on brand or on a fixed margin set, performance max on the rest) and different products. two campaigns with the same products and different targets is a bidding war you pay for.

8. read the channel and placement reports monthly

channel performance shows where the spend went. if search and shopping are under 80% of spend on a retail campaign, the campaign is buying display and youtube reach to spend budget. placement by network shows what to exclude.

---

what to stop doing immediately

• stop reporting performance max roas with brand included.
• stop running performance max campaigns under 30 conversions a month.
• stop setting target roas from the campaign's own reported number.
• stop splitting one catalog into many campaigns to control budget.
• stop leaving the campaign-level negative list empty when the search terms report exists.
• stop running standard shopping and performance max on the same products with different targets.
• stop treating final url expansion as harmless without url rules.
• stop assuming the 2023 structure still applies. the controls changed; the structure has to.

---

the orbit agent instructions

{
  "guide": "performance-max-2026",
  "agent_tasks": {
    "audit_brand_carve_out": {
      "check": "for each performance max campaign, confirm a brand exclusion list is applied, read the search terms report for branded queries over 30 days, and confirm a separate brand search campaign exists",
      "flag": "no brand exclusion list, branded queries above 10 percent of performance max search spend, or no dedicated brand campaign",
      "recommend": "apply the brand exclusion list, move brand to its own search campaign, and re-baseline performance max roas after 30 days"
    },
    "audit_conversion_floor_and_targets": {
      "check": "for each performance max campaign, count conversions in the last 30 days, read the bid strategy and target, and compare reported roas with conversions by conversion time and store revenue",
      "flag": "fewer than 30 conversions in 30 days, a target roas on a campaign under 60 conversions, or a target set above what reconciled revenue supports",
      "recommend": "merge sub-floor campaigns, run maximize conversion value without a target until 60 conversions a month, then set the target from the reconciled number"
    },
    "audit_negatives_and_search_terms": {
      "check": "read the performance max search terms report for the last 30 days sorted by cost and the campaign-level negative keyword list",
      "flag": "more than 10 percent of search spend on terms with zero conversions and no negative coverage, or an empty campaign-level negative list",
      "recommend": "add negatives for the zero-conversion terms and schedule a two-week review"
    },
    "audit_asset_groups_and_landing_control": {
      "check": "count asset groups per campaign, read asset ratings and asset-level conversions, list search themes per group, and read url contains rules and final url expansion status",
      "flag": "fewer than 3 or more than 6 asset groups, any group with all assets rated low after 30 days, final url expansion on with no url rules, or fewer than 10 search themes on a group",
      "recommend": "consolidate or split groups by message, replace low-rated assets, add url rules or run a final url expansion experiment, and fill search themes from the search terms report"
    },
    "audit_new_customer_setup": {
      "check": "confirm the new customer acquisition goal and mode, the customer match list feeding it, its refresh date and size, and whether existing customers are excluded from prospecting campaigns",
      "flag": "no new customer goal on a prospecting campaign, a customer list older than 30 days, or no customer list exclusion",
      "recommend": "set the goal with high value mode, refresh the buyer list through data manager, and exclude existing customers from the prospecting campaign"
    },
    "monitor_channel_mix_and_overlap": {
      "check": "monthly, read channel performance and placement by network for each performance max campaign, brand search impression share, and any standard shopping campaign sharing products",
      "flag": "search plus shopping below 80 percent of spend on a retail campaign, brand search impression share down more than 10 points since the performance max launch, or product overlap with standard shopping at different targets",
      "recommend": "exclude low-value placements, restore brand to its own campaign, and give overlapping campaigns different products or the same target"
    }
  }
}

---

the bottom line

performance max in 2026 is steerable: brand out, negatives in, url rules on, one campaign per goal above the conversion floor, three to six asset groups by message, and targets set from numbers you reconciled yourself.

the campaign did not get honest on its own. the controls to make it honest shipped. use them.

---

built by orbit - the performance marketing intelligence layer. questions? jose@orbitllm.com