Two exclusion controls surfaced inside Performance Max in the final week of July 2026: household income exclusions, spotted July 24 and reported by Search Engine Land, and a network setting labeled “Partners (Alfa)” that lets advertisers switch off the Search Partner Network and the Google Display Network independently, surfaced July 27 and reported by PPC Land. Neither has an official Google announcement or product blog entry, so both should be treated as limited rollouts rather than general availability.
The industry read is that the PMax black box has cracked open. That framing will cost Michigan advertisers money, because the two controls are not equally useful and one of them is far more dangerous than it looks.
What are the new Performance Max exclusion controls?
Two separate settings that arrived within three days of each other. The first excludes modeled household income tiers at the campaign level. The second excludes entire ad networks — Search Partners and Display — from a PMax campaign for the first time since the format launched in 2021.
The household income exclusion exposes the same brackets Google has long used in Search, Display, Demand Gen and Video: top 10%, 11–20%, 21–30%, 31–40%, 41–50%, lower 50%, and a separate Unknown household income bucket. Until this sighting, those brackets were unavailable in Shopping and Performance Max — the two campaign types carrying most of the budget for Michigan retailers.
The network setting is narrower and, for most accounts, more valuable: two independent checkboxes inside a “Partners (Alfa)” panel. Earlier PMax concessions — channel-level reporting, campaign-level negative keywords, brand exclusions — told you where your money went. This is the first one that lets you stop it going there.
How does Google actually decide your household income tier?
It does not know anyone’s income. It infers a probable bracket from aggregated signals, the heaviest of which is location — the modeled income of the area a user is associated with, blended with behavioral data. The output is a probability assigned to a cohort, not a fact about a person.
Two consequences follow, and understanding them lets you predict how the setting behaves in cases no article covers. First, because the inference leans on location, an income exclusion in a market with sharp geographic income gradients functions as a crude, indirect geographic exclusion. Second, the Unknown bucket is large — VPN users, thin signal histories, shared devices, and much of the privacy-restricted traffic that has grown every year since ATT. Exclude the lower 50% but leave Unknown on, and many of the users you meant to remove stay in the auction. Exclude Unknown too, and you cut impressions in a way that has nothing to do with income.
Why does this land differently for Detroit advertisers?
Because metro Detroit has unusually steep income variance across very short distances, modeled household income here is more tightly correlated with geography than in most metros — which makes the exclusion redundant with tools you already have, and worse at the job.
Bloomfield Hills and Pontiac share a border. Birmingham, Royal Oak, Ferndale and Hazel Park stack up along Woodward in descending order over about eight miles. A marine dealer on Lake St. Clair, a cosmetic dentistry practice in Rochester Hills, a custom builder in Northville — all already express that gradient more precisely through location targeting, radius bidding and first-party lists than a seven-bucket national model ever will.
The failure case is specific. A Harrison Township boat dealer excludes the lower 50% and Unknown to “focus on qualified buyers.” The households that actually finance a used 24-foot pontoon in March are often dual-income families in Chesterfield or Macomb Township the model places in the 41–50% band, plus a slice sitting in Unknown because they searched from a work device. He has removed a large share of his real buyers and will read the resulting CPA increase as seasonality. It is the same diagnostic trap covered in our Google Ads coverage for Michigan lead-gen advertisers: the setting hides its own damage inside a metric that moves for ten other reasons.
Should you turn off Search Partners and Display in PMax?
If you have the alpha and your account is lead-gen or considered-purchase retail, test the Search Partner exclusion first and leave Display on until you have channel-level numbers that justify killing it. This is the control worth having.
PPC Land cited industry research putting Search Partner Network return on ad spend roughly 37% below Google Search proper. Treat that as directional third-party aggregate, not a number describing your account — averages this wide usually conceal a bimodal split rather than a typical case. Broad-appeal ecommerce with strong creative often does fine on partner inventory. Fitment-critical catalogs and B2B lead-gen usually do not, because partner placements carry a very different intent profile than a search on google.com.
Michigan’s industrial base sits squarely in the second group. A Tier 2 supplier in Warren selling stamped components, an industrial distributor in Livonia, a marine parts catalog where the wrong impeller means a return and a chargeback — long research cycles, unforgiving product matching. Partner traffic arrives earlier and looser in that cycle, and PMax’s blended reporting has made it hard to prove. The exclusion is the proof.
How should a small team roll this out?
Eight steps, executable by a two- or three-person team inside a week. Sequence matters more than speed.
- Check whether you have either setting: open a PMax campaign, go to Settings, and look for a “Partners” panel and for household income under demographic exclusions. Most accounts will not see them yet.
- Before changing anything, pull 90 days of channel-level PMax reporting and export conversions and conversion value by channel. You cannot evaluate an exclusion without a baseline you captured yourself.
- Segment that export by your top five products or services. Aggregate channel performance almost always hides one SKU or service line carrying the partner network.
- If Search Partner ROAS in your own data runs more than 25% below your Search figure across two consecutive months, exclude Search Partners on one campaign only. Leave a comparable campaign untouched as a control.
- Leave household income exclusions off. If you have a documented reason — a genuine luxury price point with verified purchase data by income tier — exclude only the lower 50% and keep Unknown enabled.
- Log the change with campaign ID, setting and date. Smart Bidding needs roughly two weeks to re-stabilize, and without a log you will misattribute the transition period.
- Set a reminder for 21 days out to compare test against control on conversions, conversion value and CPA — not impressions or CTR, which move immediately and mean nothing.
- Update Google Ads Editor to version 2.13, released the week of July 27 with 17 new features, so bulk changes do not have to be made one campaign at a time in the web UI.
What deadlines interact with these changes?
Several, and they compound. Avoid making network or income changes in the same window as a forced migration, or you will not know which one moved your numbers.
- September 2026 — Campaigns using Automatically Created Assets and the campaign-level broad match setting begin auto-upgrading to AI Max, per Google’s Ads Developer Blog. This is separate from the DSA timeline.
- October–December 2026 — Q4. Freeze structural experiments. Michigan retail demand and winter-service lead volume both spike, making a mid-quarter exclusion test unreadable.
- January 2027 — The ability to create new Dynamic Search Ads campaigns is removed.
- February 2027 — DSA auto-migration begins; remaining DSA campaigns upgrade to AI Max or Performance Max. Google pushed this back from September 2026 on June 11, 2026, citing advertiser feedback and Q4.
The clean window for exclusion testing runs from now through roughly the third week of September. After that, seasonality and forced migrations make attribution guesswork. If you are also rebuilding conversion tracking ahead of those migrations, our Detroit PPC and paid search resources cover the measurement groundwork that belongs in place first.
Frequently asked questions
Is the household income exclusion officially available in Performance Max?
No. It was spotted in a European Performance Max campaign on July 24, 2026 and reported by Search Engine Land, but as of early August there is no entry on the Google Ads Announcements page or the Google Ads Product Blog confirming it. Treat it as an unconfirmed limited rollout. Availability may change or be withdrawn without notice.
Why does excluding “Unknown household income” cause problems?
Unknown is not a low-income bucket — it is a no-signal bucket. It holds users whose income Google cannot model: privacy-restricted browsers, VPN traffic, shared and work devices, and thin-history accounts. Excluding it removes a slice of your audience that has no consistent income profile at all, so the volume loss is real while the qualification gain is largely imaginary.
Will excluding Search Partners reduce my Performance Max reach too much?
It will reduce impressions, sometimes noticeably. Whether that matters depends on whether those impressions converted. Run it on one campaign against an untouched control for at least 21 days and judge on conversion value and CPA. If reach loss alone worries you, you have not yet measured the thing that decides the question.
Do these controls mean Performance Max is finally transparent?
Not really. Two exclusions in an alpha narrow the black box rather than open it. You still cannot see individual placements, control budget allocation across channels, or set channel-level bids. The fair read is that Google is releasing the controls that cost it least while advertiser pressure on PMax opacity keeps building.
The judgment call
One thing to take from the last two weeks: the network exclusion is a real instrument and the income exclusion is a blunt one. The first removes inventory you can identify and measure. The second removes a modeled cohort that, in a metro where income tracks a zip code boundary this closely, you already target more accurately by geography and first-party data.
The advertisers who get burned will be the ones who read “PMax finally has audience controls” and switch on four exclusions the same afternoon. Change one setting, keep a control campaign, wait three weeks, write the date down.
Comments are closed