Which Campaign Should Own a Search When Performance Max Runs Beside Search or Shopping

How to decide it, how to check the campaign you picked is serving that search, and how to tell added sales from sales you'd get anyway.

Flat graphic of one line splitting into two branches that reach a pair of joined seats. A blue rounded block fills the left seat and an amber circle fills the right.

Last verified October 2026

Google says an eligible exact match keyword identical to a search takes priority over Performance Max. So you add [corvane cv-220 pump seal] as an exact keyword, wait a month and open the Performance Max report. The search is still there, still spending, still selling.

Nothing broke. Segment that search's row by Ad format and it says Shopping ads. Google's priority rules don't apply to Shopping formats, a fact it lists among the exceptions on a different page from the one you probably read. Your keyword took the text ad, which is all it could ever take. A Shopping ad from Performance Max sat down beside it.

One row is a small thing until it becomes a budget decision. Someone reads it as proof that Performance Max is stealing from Search. Someone else reads a 600% ROAS and wants $4,000 a month moved out of Search by Friday. Both are reading reports that show who served a search and who took the credit. Neither report shows whether the sale would have happened anyway.

This guide is for people who manage real spend in an established account running Performance Max beside Search or Standard Shopping, in ecommerce or lead generation. By the end you'll be able to read which campaign actually served a search, decide which campaign should own each kind of search, move one and confirm it moved, and work out whether Performance Max is adding sales or collecting them, including which test your account has enough conversions to read. If this is new to you, read from the first section; it builds the model everything else uses. If you came with one question, start here.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

If you're here because… Go to
Performance Max reports better results and someone wants to move budget, or you need a fair test Is Performance Max adding sales?
You added an exact keyword and Performance Max still shows Why Performance Max still shows
You're deciding which campaign owns a group of searches or products Which campaign should own which searches
You're deciding whether to exclude brand Who should own brand searches
You've moved a search, or Search is Limited by budget Move a search and confirm it moved
Spend has moved to YouTube or Display, or channel prioritization has arrived When spend moves
A rep or client is pushing back Answering the rep

The short version. If you read nothing else, read these nine lines.

  1. One search can show two of your ads, one text and one Shopping. Treat them as two seats with different rules. Before you judge any overlap, segment the Performance Max search terms view by Ad format.
  2. In the text seat, one claim holds: an eligible exact keyword identical to the search. Plurals and synonyms aren't identical. Everything weaker is contested.
  3. In the Shopping seat, keywords have no say. Ad Rank decides between Standard Shopping and Performance Max on the same products with overlapping settings, so ownership runs through products.
  4. Either rule applies only to campaigns still eligible. Check budget, keyword status, approvals and targeting before you trust an owner.
  5. Excluding a campaign takes it out of a search without handing the search to anyone. Never route a search with an account-level negative, which blocks the owner too.
  6. Claim a search only when you can name what would change: the ad, the page, the bid, the budget or a campaign setting. Otherwise leave it.
  7. Reports answer who served a search and who got the credit. Whether the sale would have come anyway takes a readable holdout, or stays a labeled estimate. Set targets from affordable and achieved numbers, and re-measure after moving searches.
  8. Performance Max needs this share of its credited sales to be real: your floor, the lowest ROAS the business can accept, ÷ its reported ROAS. Uplift doesn't support campaigns with a feed. Before an Uplift or geo test, check it can see the lift that would pay. If it can't, redesign or skip it, decide on that share and call it an estimate.
  9. Channel prioritization, Google's High and Low spending tiers, doesn't separate Search from YouTube inside Performance Max. A Target CPA or ROAS campaign not limited by budget generally sees no change; untargeted Maximize spends on High first.

The row in the opening belongs to the account this guide follows, which is made up. Larkfield Pool Supply sells pumps, filters, chemicals and parts online, carries a pump brand called Corvane, and takes trade quote requests from pool contractors. Its floor, the lowest ROAS it can accept on retail, is 400%. It can afford $50 per quote request and $200 per qualified quote. These are its last 28 days, with conversions finished arriving:

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Campaign Type Spend Result Reported
Brand Search $1,800 $36,000 value 2,000%
Generic Search $9,000 $45,000 value 500%
Catalog Standard Shopping $6,000 $27,000 value 450%
Store Performance Max with a feed $14,000 $84,000 value 600%
Trade Search Search $2,500 40 quote requests $62.50 each
Trade Leads Performance Max, no feed $3,000 120 quote requests $25 each

Larkfield's Google rep has looked at that table and suggested the obvious move: Store is the star and Generic the laggard, so shift $4,000 a month from one to the other. It sounds like arithmetic. Larkfield's manager opens Store's search terms first and finds corvane cv-220 pump seal, which Generic holds as an exact keyword. Either Store is taking Generic's sales, which would make the rep's table unfair to Generic, or something else is going on. The $4,000 waits on that row.

Most writing on this subject pictures one prize. A search arrives, two campaigns want it, one gets it, and cannibalization means the wrong one won. Hold that picture and the row is a broken promise.

Picture seats instead. One search can show a text ad and a Shopping ad from the same account: Google's own example has a Shopping ad appearing "alongside the ad for your exact keyword" on the same search (ad group and asset group prioritization). That gives you two seats, each with its own rule for who sits down. "Seat" is our word, not Google's, and it leaves out other inventory that doesn't bear on this decision. What it gets right is the thing that matters: Generic's keyword and Store's Shopping ad were never competing for the same place.

Diagram of one search, corvane cv-220 pump seal, passing through a band labeled Still eligible: budget, keyword status, approvals, targeting. It then reaches two seats. The text seat is a ladder of four rungs: 1, exact keyword identical to the search, marked as the one claim you can rely on; 2, identical phrase or broad keyword, or search theme; 3, most relevant ad group or asset group; 4, highest Ad Rank. The Shopping seat has no ladder: Standard Shopping and Performance Max, with the same products and overlapping settings, go to highest Ad Rank, and keywords don't reach this seat.

Figure 1. One search, two seats, and both can be filled at once. Each seat has its own rule, and both rules apply only to campaigns still eligible.

The text seat has a ladder

Only one of Larkfield's keywords or search themes can enter the auction for a search, and Google's prioritization page gives the order. Here it is for the search pool pump seal:

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Rung What qualifies Larkfield's example
1 An exact keyword identical to the search [pool pump seal] in Generic
2 A phrase or broad keyword, or a Performance Max search theme, identical to the search The search theme pool pump seal in Store
3 The most relevant ad group or asset group, picked by Google's AI An ad group about replacement parts, over one about new pumps
4 The highest Ad Rank among what's left Whichever remaining candidate scores higher

Rung 1 is the only claim you can rely on, and the only rung reserved for Search. Below it the ground is contested. Google's shorter page on Performance Max and other campaigns goes straight from an exact keyword to Ad Rank, so don't build a plan on finer distinctions further down.

"Identical" is a narrower word than "exact", and this catches a lot of people. An exact keyword can serve searches with the same meaning or intent, including plural forms. Priority asks for more: spelling corrections count as identical, and plurals and synonyms don't. So [pool pump seal] can serve pool pump seals all month and hold no claim on it.

Rung 2 has a trap for anyone who adds search themes for reach. Give Store the theme pool pump seal and it beats Generic's phrase keyword "pump seal" on that search, because the theme is identical and the keyword isn't. Used on purpose, a theme is the one claim Performance Max can make, but it still loses to an identical exact keyword, and it only adds: narrowing stays the job of exclusions and negatives.

The Shopping seat has no ladder

Google says the priority rules don't apply to Shopping campaigns or other Google inventory, so no keyword reaches this seat. When Standard Shopping and Performance Max target the same products with overlapping locations, languages, hours and other settings, Google's Shopping FAQ says Ad Rank decides which one serves. Standard Shopping's campaign priority sorts out Shopping campaigns promoting the same product. Given that FAQ, we read it as having no say against Performance Max.

That makes this seat something you own through products. Catalog and Store both carry Larkfield's pumps, so Ad Rank is choosing between them for every pump search, whatever Generic's keywords say. To change who sits there, change which campaign carries the product, or take a campaign out of that search.

If you run lead generation with no feed, you have one seat, and you can skip to the next heading.

Both rules apply only to campaigns still eligible

A rung is only as good as the campaign standing on it. Google lists why a keyword you expected might not serve: a campaign limited by budget, a keyword with low search volume status, all ads or landing pages disapproved, or targeting that leaves the searcher out. Searches on Lens, AI Mode, AI Overviews or autocomplete are a further exception: the term in your report is an approximation, and it isn't technically identical to any keyword, however identical it looks.

Larkfield's Brand campaign is the plain case. It's limited by budget on Saturdays, so it holds its own name during the week and drops it at the weekend. The same search can sit in both reports in one month with the ladder working exactly as described, which is why a headline overlap figure can't settle anything. Optmyzr found Search keyword text appearing as a Performance Max search term in 91.45% of 503 accounts in February 2025, and it says it didn't analyze eligibility.

A part number or odd model name can reach low search volume status if few people search it worldwide. Read the keyword's Status column: the status is judged on twelve months of worldwide searches and rechecked automatically. Larkfield's [corvane cv-220 pump seal] shows as eligible, so it's doing its job.

What you can now read from a row

Back to the $4,000. Store's row isn't evidence that Store took Generic's sale. Here's what each kind of row means, in the Performance Max search terms view with Ad format showing:

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

In the Performance Max view Ad format What it means What to do
Any search, including one identical to your exact keyword Shopping ads Different seat. Normal, unless you've decided to remove Performance Max from that seat Leave it. Your keyword was never competing for that place
A search identical to one of your exact keywords, dated after the keyword existed Text ads The keyword may not have been eligible then, or the row came from inventory outside the priority rules Check the row's dates, as in When the owner can't hold a search
A plural or synonym of your keyword Text ads Nobody claimed it Add that form only if you'd change something for it
A search your phrase or broad keyword also fits, without being identical Text ads Normal. Contested ground nobody owns Claim it only if you'd change something for it
A search none of your keywords fits Text ads Performance Max found it Usually what you're paying it for. Leave it

You can't tell from the view which rows came from Lens, AI Mode, AI Overviews or autocomplete. So one remaining Text row on an identical term proves nothing. Expect some, and judge the pattern.

Your levers follow from the seats. An identical exact keyword claims the text seat. An exclusion or negative takes a campaign out, and the search goes to whoever is left, which may be nobody. Budget and targeting keep an owner eligible. The Reference has them side by side.

From here the guide follows the order Larkfield works in: read the rows and the credit, decide who should own what, make the moves, confirm they took, and only then ask whether Performance Max is adding anything.

Read who served a search, and who got the credit

Three questions get mixed up in every argument about these campaigns. Each has a different place to look.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Question Where the answer is What it can't tell you
Who served this search? Search terms reports, with Ad format for Performance Max Whether either campaign was needed
Who got the credit? Campaign columns, the channel performance report Whether the sale would have come anyway
Would the sale have come anyway? A readable holdout Which search or channel did the work

Read who served

  1. In the Campaigns menu, open the Search terms report and choose Search terms and landing pages for Performance Max from the dropdown (Google's page on the view). Set dates old enough for conversions to have arrived.
  2. Show the Ad format segment. Google names it for this view and gives no click path, so look for it in the view's own controls. It separates Shopping ads from Text ads. Google's API view lists no such segment, so don't count on one in an export.
  3. Open the Search terms report for your Search campaigns over the same dates, and add the Keyword column through Columns, then Attributes. It's off by default.
  4. List the searches that appear in both, and mark each with its ad format.

You've finished when every overlapping search has a format beside it and you can place it in one of the five rows above. Store visits and store sales aren't in this report, and terms with too little query activity are left out for privacy.

Check whether credit moved

If Search fell while Performance Max rose, compare three dated lines before you call it either way: Search, Performance Max and the account total, on the same conversion action, with conversion delay left out of the dates.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

What the three lines do What you can say What you can't say
Search down, Performance Max up, total flat Credit moved That Performance Max added nothing. Only a holdout can say so
Search down, Performance Max up, total up Something grew That Performance Max caused it
Search down, Performance Max up, total down Something's wrong beyond credit That Performance Max is to blame. Date it first

Lead generation needs one more comparison, because cheap conversions are where these reports mislead most. Larkfield's sales team marks which quote requests are real, and the table changes shape:

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Campaign Quote requests Cost per request Qualified quotes Cost per qualified quote
Trade Search 40 $62.50 16 $156
Trade Leads 120 $25 24 $125

Trade Leads looks two and a half times cheaper per request and is 20% cheaper per qualified quote. Both are true, and only the second is about the business. If you see any two of a soft conversion, a jump without extra spend, mostly non-Search conversions, or other campaigns on the same demand not improving, check the CRM before funding. Korf Digital reported that pattern in one account, with lead quality unverified. Cascader's answer on form fills and CRM sales covers which stage to count.

Which campaign should own which searches

You now have rows and you know what they mean. The decision is whether to change anything, and most of the time the answer is no. Ownership buys control, so a claim is worth making only if you can name one change it would let you make: to the ad, the page, the bid, the budget, or a campaign setting such as hours, location or conversion goal. If the answer is "I'd have this term in my keyword list", leave it. Cascader's piece on when a term deserves its own keyword goes through that test row by row.

Published data won't decide this for you. Same-search comparisons favor Search on click and conversion rates more often than not (Adalysis, non-retail), the one study that sized the gap, Optmyzr's, found most differences under 10%, and none measured profit.

Build your own table

  1. Take your overlapping rows and your biggest Performance Max rows. Give each the first kind below that fits, reading from the top.
  2. Write the named change beside any row that has one. No named change, no claim.
  3. Decide each seat separately.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Kind of search Text seat Shopping seat
Outside the offer A negative reaching every campaign that shouldn't serve the search and no campaign that should, written with the Negatives guide The same negative
Your own brand Brand Search, if it's eligible. See the next section A separate choice
Competitor or partner brand you don't bid on or sell Brand exclusion in Performance Max if it's listed, otherwise the Negatives guide The same exclusion, unless you allow Shopping ads
You'd change the ad, page, bid, budget or a setting Identical exact keyword in an eligible Search campaign. Check that campaign's AI Max state: text customization can change the ad, and Final URL expansion, on by default if you turn AI Max on by hand, can change the page and doesn't respect pins Stays with whichever campaign carries the product
Part number, very rare search or carried brand Leave it open Open. The product listing does this work
Anything else, including searches a phrase or broad keyword fits and searches you couldn't have named in advance Performance Max or broad Search. Leave it Whichever campaign carries the product

In lead generation with no feed you have one seat, so ignore the Shopping seat entries.

A claimed keyword with low search volume status won't serve until searches rise. If a listing can carry the sale, leave it to the listing. If the change only works in a text ad, use a broader match type and give up rung 1. Otherwise wait. That order is our judgment.

Larkfield's table comes out like this. Own-brand Text rows are 16% of the spend in Store, its Performance Max campaign, and 34% of its credited value, so they move to Brand. Five money terms earn a claim, because Larkfield has a sizing page, a rebate page and a seal-fitting page that Store's ads don't reach. One is a Corvane part. Generic, Larkfield's non-brand Search campaign, already holds [corvane cv-220 pump seal] for the same reason: both go to the fitting page. The rest of Corvane stays open, and everything else stays with Store, including most of what it found by itself. No term moved because it looked untidy.

When the same products sit in Standard Shopping and Performance Max

If you need a specific owner, take the product out of one campaign. Keywords and campaign priority don't decide this seat. In Performance Max, a product leaves the campaign's listings only when it's excluded in every asset group that includes it, and that takes out its listings, not necessarily Text ads. The usual reason is a limit Performance Max can't honor, such as a per-click ceiling.

If you don't need a specific owner, leave both only if you accept Ad Rank deciding, and read them as one Shopping total. Larkfield has no reason to name an owner today, so it reads Catalog and Store's product results together and stops comparing them.

If you want one search out of Performance Max with the products left in both, claim it in Search first. That moves the search in the text seat. It doesn't move the budget, and it doesn't reach the Shopping seat, because Standard Shopping can't claim a search. A Performance Max campaign negative is the second resort, on the terms in Make the claim.

Who should own brand searches

Google's position has two parts. It recommends brand exclusions when you manage brand traffic in a separate campaign, and it warns that brand settings limit traffic and performance may decrease. It also says Performance Max may filter navigational or brand-specific queries so that dedicated Search campaigns serve them, with no frequency or conditions given. So don't assume the filter is doing your work.

Evidence on the outcome is split. Across Haus's 2024 customer geo tests, excluding brand won 50% of experiments on revenue. Lower new-customer CAC showed up in every test that reported it, a subset of unstated size. Caraway's Performance Max with brand generated more new-customer revenue, and some low-transaction, high-AOV brands did better including brand. The whole-account question, whether brand ads add anything, is in our review of the brand search studies.

Take the first row that fits.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Your situation What we'd do
Carried brand Not your brand. Leave it open and decide by seat
No brand Search campaign, or one that's limited by budget, has unapproved ads or doesn't match targeting Build or fix it and fund it first. Don't exclude, whatever your KPI
Low transaction volume and high AOV, with a repeat or blended goal Skip the default. If the two sums before a test say it can see break-even, test: a geo split between regions with brand in and regions with brand excluded, with break-even resting on the spend difference between the arms. Otherwise leave brand in, or run a dated exclusion with a reversal date and label it an estimate
New-customer volume or revenue is your main KPI Leave brand in. The evidence is mixed, and Caraway did better with it in. That's our judgment
New-customer CAC is your main KPI Exclude brand. It's the better-supported default
Anyone else Exclude brand. That's the default in our judgment: Google recommends brand exclusions if you manage brand traffic in a separate campaign, and by this row yours is funded and eligible

Wherever the table says exclude, do it in Performance Max, decide the Shopping option, then re-price the target. If your own brand isn't in Google's list, request it through the option in Google Ads and use the Negatives guide's route meanwhile.

The Shopping option is a second decision. In Performance Max, brand exclusions apply to Search and Shopping inventory, and Allow Shopping ads on searches that mention excluded brands keeps Performance Max in the Shopping seat. We'd allow it unless a Standard Shopping campaign carries the same products and you want that campaign to have the seat, because with Allow off the seat goes to whoever is left. That's our judgment, and the steps are in the same guide.

Larkfield moves its name

Brand was limited by budget on Saturdays, and an exclusion takes Store, its Performance Max campaign, out without handing the search to anyone, so Larkfield raises Brand's budget first. Then it excludes Larkfield from Store, with Shopping ads allowed, and waits for a matured window. With Allow on, Text rows should stop and Shopping rows may stay. With Allow off, new rows of both formats should stop. Non-brand uses of the words and hidden terms can remain.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Campaign Spend Credited value Reported ROAS
Brand $1,800 → $2,900 $36,000 → $63,500 2,000% → 2,190%
Generic $9,000 → $9,000 $45,000 → $45,000 500% → 500%
Catalog $6,000 → $6,000 $27,000 → $27,000 450% → 450%
Store $14,000 → $11,800 $84,000 → $55,500 600% → 470%
All retail $30,800 → $29,700 $192,000 → $191,000 623% → 643%

Store lost $28,500 of credited value and Brand gained $27,500. Total credited value barely moved, and spend fell by $1,100.

Now read the rep's comparison again, the one that wanted $4,000 moved from Generic, non-brand Search, to Store. Store reports 470% and Generic reports 500%. The star and the laggard have swapped places, and Generic didn't do anything. No campaign got better or worse at selling pool pumps. A share of Store's credit went back where it had been coming from, and the ranking that justified the $4,000 went with it. Store has a worse month on paper and the account is slightly better off, which is the rare kind of bad news worth wanting.

Store's 550% target was set when the campaign reported 600%, and it's now stricter than anything Store achieves. Left alone, it starts buying less for a reason nobody chose. After a brand move or any large block leaves Performance Max, re-measure achieved on matured data, re-price the target with our Target CPA and Target ROAS guide, and check in the channel report where the budget went. Larkfield re-prices between the 470% Store now achieves and the 400% it can afford, with a margin on the business side. Never do this during a test.

Move a search and confirm it moved

A term deserves a keyword only if you can name one change it would let you make: to the ad, the page, the bid, the budget or a campaign setting. What's specific to Performance Max is how to claim the search and how to tell the claim worked.

Make the claim

  1. Add an exact keyword with the term's exact text. Priority covers that form and its spell-corrected forms only. A plural, synonym or reordered form has no priority of its own, so add it only if you can name a change for it too.
  2. Put it in an eligible campaign, and make the change. Build the ad, page, bid or setting that justified the claim, or you've claimed the search and changed nothing. The campaign passes if its ads are approved, its targeting reaches the searcher and it isn't limited by budget. A limited campaign holds the claim only while its money lasts, so read a partial result there as a budget result, not a failed claim.
  3. Protect the keyword. Dismiss any Remove redundant keywords or Remove non-serving keywords recommendation that touches it, and turn that type off under Auto-apply settings, which works account-wide. Google also auto-pauses keywords with no impressions in 13 months, so recheck rare claims.
  4. Note the date. Everything you read next is read against it.

Prefer a claim to a negative. Use a Performance Max campaign negative to move a search only when a claim can't hold it, after confirming the destination has an eligible matching keyword, and accepting that Performance Max's Shopping ads go too. Google calls these negatives highly restrictive. Never do this at account level, because an account-level negative blocks your owner as well.

Confirm it took

After one to two conversion cycles, compare with the same stretch before the date and look for three things:

  1. The keyword gets impressions, and the term shows under it in the Search terms report if it's above the privacy line.
  2. Performance Max's Text rows for the term fall.
  3. The total for that kind of search, across Search and Performance Max, held.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

What you see What it means What to do
Text rows fall, a few remain It took. Some rows come from places priority doesn't govern Leave it
Shopping rows remain Different seat. Expected, unless you've decided to remove Performance Max from that seat Don't add a negative for that reason alone
Keyword has no impressions It isn't serving Read its status. For low search volume, see the choice under the table
Text rows unchanged The owner isn't holding it Run the check below
Text rows gone, but the total fell The move cost sales, or something else changed Date it, and we'd undo the move if nothing else explains the fall

A Text row that stays on an identical term means the owner was out of the running on those dates, or the row came from inventory priority doesn't govern. Budget is the cause you can least afford to misread, so start there.

  1. Budget. For a past period, read Search lost impression share (budget) by date on Search and Shopping target campaigns. On untargeted Maximize, compare budget with spend. If an owner was limited at any point, assume it lost identical searches on those days.
  2. Keyword and ads. Their status columns show today only, so past states come from change history. A blank there doesn't rule a change out, because Google doesn't always list account-level changes or its representatives' changes.
  3. Targeting, experiments and match type. Treat any gap between the owner and Performance Max on location, language, audience, schedule or network as open to Performance Max. Check whether a test was splitting traffic and whether the keyword is truly exact.
  4. The present state. The Ad preview and diagnosis tool shows whether another candidate was identical or more relevant, and Performance Max asset groups can appear in its results. It can't settle a past row. A preview can vanish once the campaign reaches its average daily budget, so a blank late in the day may be the budget again.

If the cause is budget, the Limited by budget status doesn't show that extra spend would pay.

  • If extra spend pays on a Search or Standard Shopping owner, raise the budget, using What a Target Does When the Money Runs Out and our Target guide. On a target campaign, work through "The decision" under "Target campaigns that are Limited by budget" in that guide: the table at its step 4 governs a budget raise.
  • If it doesn't, don't raise it. Move the claimed keywords to a campaign with its own unshared budget, use the fixed-budget choices in "The decision", or accept the loss.
  • If the limited owner is Performance Max, reported results can't show the next dollar pays. If you can read a test, run one and act on its result. If you can't, hold its budget flat and label the conclusion an estimate.

Larkfield's five claims, scored

All five went into Generic, Larkfield's non-brand Search campaign, which is on Maximize conversion value with no target and runs out of budget at weekends.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Claimed keyword Page What the rows showed Verdict
[variable speed pool pump] Sizing Text rows fell, Search shows the term, kind-of-search total held Took
[above ground pool pump] Sizing The same Took
[salt chlorine generator] Rebate The same Took
[inground pool pump] Sizing Text rows fell on weekdays, unchanged at weekends Took on weekdays only
[corvane cv-220b shaft seal] Seal fitting No impressions, Low search volume in Status Won't serve for now. Left to the product listing

Shopping rows stayed on all five, as they should. The fourth is the one that matters. Homeowners price inground pumps at the weekend, and the other three terms are searched mostly on weekdays, so the same limit caught one claim and barely touched the others.

Raising Generic's whole budget doesn't pay, even at a 500% average, because an average doesn't show the next dollar pays. Google's budget recommendation estimates another $1,500 a month would bring about $4,800 of value, which Larkfield divides out to 320%, under the 400% floor. That's reported value, not lift, and your account may not show such an estimate, but it's enough to decline a raise. So Larkfield accepts the weekend loss on [inground pool pump].

Generic's weekend limit also changes how the rep's $4,000 looks. Take that from Generic and it runs out sooner, loses more of its own keywords to Store, Larkfield's Performance Max campaign, and Store reports those sales as its own. No report can say whether those sales would survive. The move would have made its own case look better whether it worked or not.

Is Performance Max adding sales, or collecting sales you'd get anyway?

Campaign numbers report credit, so they can't settle this. If the brand table told you to exclude, do that before you test. Otherwise test with brand in and say so.

Work out what would have to be true

Needed share = floor ROAS ÷ reported ROAS

For leads, use reported cost per lead ÷ affordable cost per lead. Your floor, the lowest ROAS the business can accept, comes from margin: see our Target guide.

A campaign reporting 600% against a 400% floor claims $6 of sales for every $1. If only $4 of that was truly added, it's exactly at the floor. So 400 ÷ 600 means two thirds of its credited sales have to be sales you wouldn't have had otherwise. A comfortable-looking ROAS is really a claim about that share, and the report never says whether it's true.

Store, Larkfield's Performance Max campaign, saw its needed share go from 67% to 85% when the brand credit left. A high share is a reason not to add budget on reported results: above about 80%, in our judgment, don't. A low share shows that current spend could pay, not that the next dollar will.

Choose a test your account can run

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Test Who can run it What it answers The trap
Uplift experiment Lead gen, online sales with no feed, or store goals, judged on a Google Ads conversion Does it add conversions on top of comparable campaigns? Much larger comparable campaigns can bury the result
Standard Shopping vs. Performance Max experiment Feed accounts, with the same products in both campaigns and in no other campaign Which campaign type does better on these products? The experiment's result applies by default, so opt out unless you've accepted the result
Geo go-dark Accounts with sales by region What do total sales do without it? Regions differ, so it's noisier than the sum below suggests
Dated brand exclusion Accounts with a funded brand campaign Was the brand part collected rather than added? No control group, so it's an estimate

With a Merchant Center feed, Uplift isn't open to you, so a retailer's real question goes to a geo go-dark, or to Conversion Lift based on geography if a Google representative offers it.

Check the test can see what matters

Before an Uplift or geo test, work out the break-even lift. The campaign has to add its spend × your floor. Take that from your total sales with Performance Max running, in the same regions and window as the spend, and you have what you'd get without it. Lift is measured from there, so divide the first figure by the second. For leads, it has to add spend ÷ affordable cost per lead at the stage you're judging, out of total leads at that stage. Store has to add $11,800 × 4 = $47,200 out of $320,000 of all-channel sales per 28 days: $47,200 ÷ $272,800, or 17.3%.

Smallest visible lift = 2.8 × √(1/n₁ + 1/n₂)

Here n₁ and n₂ are the conversions in each arm, counting readable weeks only and the outcome you're judging. In a spreadsheet that's =2.8*SQRT(1/A1+1/B1). It's Cascader's calculation, not a Google figure, at 95% confidence and 80% power, and it's a best case. The number is how far apart two things must be before the test can tell them apart. In planning an Uplift or geo test, they're a flat result and break-even. If the number comes out above break-even, a flat result means nothing, because the campaign could be paying and the test wouldn't see it. In our judgment a design within a point or two of break-even fails too.

A before and after has no arms, so don't apply the formula. In a Shopping vs. Performance Max test, base break-even on the spend difference × your floor. The test settles only which type beat the other. If the lift of the type that spent more beats break-even by more than the smallest visible lift, that type won. If that lift falls short of break-even by more, the other type won. If neither did, it can't tell them apart, so keep what you have, in our judgment.

At about 615 orders a week, going dark in a third of its regions for five weeks and setting aside the first week leaves Store 820 orders dark and 1,640 elsewhere: 12.0% against 17.3%. We'd run it, knowing the test can tell a flat result from break-even and can't place a result close to break-even. And 12.0% is a best case: regions vary, and a revenue read is noisier than an order count.

Google leaves out the first 7 days, so a six-week Uplift test of Trade Leads at 50/50 has five readable weeks. Totals count both Trade campaigns.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Counting Break-even lift Smallest visible lift Verdict
Quote requests, 100 in each arm $3,000 ÷ $50 = 60, and 60 ÷ (160 − 60) = 60% 40% Readable
Qualified quotes, 25 in each arm $3,000 ÷ $200 = 15, and 15 ÷ (40 − 15) = 60% 79% Not readable

The test can read the number that's easy to collect and can't read the number that pays the bills.

Set it up so it can't fool you

On a geo go-dark, cut a budget-limited campaign's daily budget to the historical share of spend in the regions left on, or the money moves into them and inflates the baseline. We infer that untargeted Maximize needs the same cut, even with no such status. Record each arm's prior share of sales, because lift can't be read without it.

On Uplift, check comparable campaigns after day one. Keep every campaign that can serve the same searches or inventory as Performance Max, and remove only those that can't. Removing one that shares searches strips out the cannibalization you came to measure.

The rest is in the Reference.

You've done this properly if, before launch, you can write the sentence you'd send at the end for each outcome.

Read the result in money

Read a geo holdout on business totals against your floor. Read a Google experiment on its group results, and work break-even again from the test itself. On Uplift, that's Performance Max's spend in the readable weeks and the total of the group it ran in. Turn on View Results with Comparable Campaigns to see that total, because comparable campaigns are excluded from reporting by default. Count that total in the outcome you're judging: if you're judging qualified quotes, count qualified quotes, not quote requests.

Over Store's four readable weeks, regions left on sold $213,000 and dark regions $92,600, a ratio of 2.30. Their prior shares of sales were two thirds and one third, a ratio of 2.00. So the lift is 2.30 ÷ 2.00 − 1, or 15%, against the 17.3% break-even. Added sales = $320,000 × 0.15 ÷ 1.15 = $41,739, an incremental ROAS of 354% against the 470% Store reports.

On an Uplift or geo test, the smallest visible lift decides again, in our judgment: it's how far apart your lift and break-even must be before the test can tell them apart. If the gap between them is smaller, say inconclusive, not "no effect": you can't say which side of break-even the campaign is on, so hold spend and look again. If the lift beats break-even by more, test more budget in steps. If it falls short by more, this spend isn't paying at the floor. Store's gap, from 15% to 17.3%, is far smaller than the 12.0%, so the test can't say, and the 354% isn't a verdict.

The 354% doesn't go in the target field, and a target result isn't proof of lift.

If your account can't read any test

Many can't, and finding that out took one spreadsheet line, not six weeks of budget. Lengthen the test or widen its scope if you can. Otherwise, use needed share and your table of searches, hold Performance Max's budget flat, and label the conclusion an estimate.

A longer test of Trade Leads would freeze budget and assets across the turn of the pool season, so Larkfield runs no test. Needed share on qualified quotes is $125 ÷ $200, or 63%. It keeps Trade Leads at $3,000 and writes "estimate" beside the number.

"We can't tell" is a legitimate finding, and it's cheaper than the alternative.

When spend moves, and what channel prioritization changes

A week after the geo test ended, Larkfield added a new video to Store, its Performance Max campaign. A month later Store's YouTube share had gone from 10% to 31%, and the question every Performance Max manager asks at this point arrived: what moves spend back to Search?

Read where the money went

The channel performance report answers where the money went, and that's all it answers.

  1. Go to Campaigns, Insights and reports, then channel performance (Google's page). Select one Performance Max campaign to get its detail.
  2. Set dates after June 6, 2025 that leave out your conversion delay. A recent window makes CPA look higher and ROAS lower than they are.
  3. Start at campaign level. Decide whether the campaign is reaching its goal before reading any channel.
  4. In the channel distribution table, read cost by channel. Add Ads using product data through Segment on the table toolbar to see Shopping ads by channel.
  5. Read the Status column for diagnostics. If a channel has no issues and isn't spending, Google says other channels likely offer a higher return at that moment.

You've finished when you can say what share of spend went where, over which dates. YouTube search results count under YouTube, not Search.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Store's channel Spend Share of spend Share of clicks
Google Search, Search partners, Maps $7,000 59% 50%
YouTube $3,660 31% 14%
Display Network $700 6% 30%
Discover, Gmail $440 4% 6%

Display has 30% of the clicks and 6% of the spend. Use spend share to say where the money went. That gap isn't a reason to act by itself.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

What you see What to check What it doesn't justify
A channel's average ROAS looks low Whether the campaign as a whole is reaching its goal. Google warns that channel averages mislead Cutting the channel on that alone. Channel isn't cause
YouTube's result looks high Ad event type (Segments icon, Conversions, Ad event type), which separates engaged-view conversions from clicks Believing it before you've looked
Spend shifted The date, on the time series and against change history Any change made before you've dated it

Store's time series puts the shift on the date of the new video, and most of YouTube's reported conversions turn out to be engaged-view. A budget-limited target campaign has one more date to check: August 17 to 27, 2026, when Google's change to target-based bidding could shift traffic in multi-channel campaigns.

What moves it back

Inside the campaign, nothing does directly. You can't pause channels like YouTube in Performance Max, and no setting separates Search from YouTube. Here's what does exist:

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

You want What exists What it doesn't do
Specific searches back Claim them in Search It moves the search, not the budget. YouTube-type spend may not fall
Less of unwanted content Audience signals and account-level placement exclusions can steer the AI away from it It doesn't set a channel's share
Less Display Network or Search partners, for a reason beyond low average ROAS Set the channel to Low with an allocation, where available. 0% stops spend It does nothing for YouTube or Search
A fresh start Don't rebuild unless the cause is dated and rebuilding addresses it A rebuild can repeat the problem

Larkfield has a dated cause. Asset reporting shows how the video is doing, and Google's asset guidance suggests considering replacing lower-performing assets when an asset group is full, but it doesn't say that moves spend between channels. Larkfield reads that report, keeps its claims where they are, and doesn't rebuild Store. It accepts that the claims didn't pull YouTube's share down, because claims were never going to.

What channel prioritization changes for ownership

Channel prioritization sets a channel to High or Low, and High spends first. A Low channel can also take an allocation, a best-effort percentage of average daily budget. In Performance Max, Search, YouTube, Maps, Gmail and Discover are prioritized together as Google channels, so it can't separate them. Only Search partners and the Display Network take their own priority. It's still rolling out, and you can't set it from a manager account.

Take Generic, which owns Larkfield's non-brand keywords. At 0% on Search partners, its keywords couldn't serve on partner searches, so those searches would be open to Performance Max, which isn't the same as taken. At Low with no allocation, that would happen only when the budget ran short, as it does at weekends. As we read it, that's the only way a tier or allocation changes who holds a search: by stopping the owner spending on the channel the search runs on, directly or by letting another channel take the budget first.

A Target CPA or ROAS campaign not limited by budget generally sees no change from priority. Limited by budget, its High channels spend first, within the target. On Maximize with no target, High spends first and the rest goes to Low. An allocation on a Low channel applies in every case. So recheck:

  • Any campaign with a Low tier while limited by budget.
  • Any campaign on untargeted Maximize with a Low tier.
  • Any campaign with an allocation, including 0% and an inherited account default, on a channel where it owns searches.
  • Any Search campaign with Display on, by February 2027.

Starting in February 2027, Search campaigns with Display on move to these controls automatically. The current spend setup goes away, and spend may shift between Google Search, Search partners and Display, though Google's pages don't say which way, or which settings a migrated campaign lands on. Display will no longer serve on Search campaigns using the Target Impression Share bid strategy. Priorities set on a campaign before then apply right away and replace current Display Expansion budget behavior. Google says Low priority with a 50% daily allocation keeps the current configuration, on a best-effort basis, and What "Low priority, 50%" leaves out covers what that recipe doesn't set.

When the controls reached Larkfield's account, Store was a target campaign not limited by budget, so priority changed nothing. Generic has Display on and runs out of budget at weekends, and by the same reading a shift toward Display in February could take money from Google Search, where its keywords serve. So Generic goes on the February list, with a decision to make before then. After all that, Larkfield hasn't touched a tier.

Answering the rep: one account, start to finish

Here's Larkfield's whole decision in one place, then the sentences you're likely to hear next. Store is its Performance Max campaign, Generic its non-brand Search campaign, and the floor 400%.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

When What Larkfield read What it did What happened
Month 0 Rep suggests moving $4,000 from Generic to Store. Store's row is a Shopping ad Held the budget The row proved nothing
Month 0 Rows by format and kind of search Decided to move own-brand text, claim five terms, leave the rest A short list of changes, each with a named reason
Month 1 Brand limited on Saturdays Raised Brand's budget, excluded brand from Store, allowed Shopping, re-priced Store's target Store 600% to 470%, Generic 500%, credited value flat, spend down $1,100
Month 2 Five claims after two conversion cycles Accepted the weekend loss on [inground pool pump], left one to the listing Three took, one held on weekdays only, one wouldn't serve
Month 3 Needed share 85%. Geo design 12.0% against 17.3% Went dark in a third of regions for five weeks Lift 15%, incremental ROAS 354%, too close to break-even for the test to say which side of the floor
Month 3 Trade Leads can't be read on qualified quotes No test, budget flat, labeled estimate Needed share 63%, reported as an estimate
Month 5 YouTube share up after a video Dated it, reviewed the video, didn't rebuild Claims held, YouTube's share didn't fall

The answer to the rep comes down to two sentences. Store reports 470% and the floor is 400%, and our test landed too close to break-even to say which side of the floor it's on, so we'll hold it and look again next quarter. The $4,000 stays in Generic because nothing shows it would pay more in Store, and moving it would hand Store more of Generic's own keywords, so Store's report would rise whether or not the move worked.

What you'll hear next

A rep's second sentence is usually better than the first. Nothing here needs you to win an argument. Google's own pages and a sum do the work.

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

You hear Check You can say
"Performance Max has your best ROAS" Needed share, and how much of the result is brand "It has the best credit. At our floor it needs 67% of those sales to be new."
"Your exact keywords are protected, so overlap doesn't matter" Ad format. Rung 1 covers text only, and only when eligible "That's true for the text ad. Your own page says the rules don't apply to Shopping."
"Exclusions limit performance" Google does warn of this. Check total brand conversions after the change "Agreed, so we funded Brand first and we're watching the total, not the campaign."
"Search is limited by budget, so it can't compete" Whether extra spend pays, and which identical searches the limit is losing "Limited means it's losing searches. It doesn't mean take more away."
"You just said Generic's extra spend doesn't pay" Which dollars. The estimate priced money above today's budget, and the $4,000 is money inside it "Not at the top of the budget. Nothing we've measured says the $4,000 would earn more in Store."
"The experiment came back favorable" Which test it was, and the bid strategy. On Maximize with no target, favorable means more "It says this campaign type sold more on these products. It doesn't say the sales were new."
"The channel report shows YouTube converting" Ad event type, and the dates "How many of those are engaged-view? Google also says not to judge by channel average."
"The leads are cheaper in Performance Max" Cost per qualified lead, same action, both campaigns "They're 20% cheaper at the stage we can sell to, not two and a half times."
"Channel prioritization will fix this" Whether the campaign is limited by budget "Search and YouTube are one group in Performance Max. It can't separate them."

Reference: levers, tests, reports and where Google's pages disagree

Levers by seat

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Lever Seat What it does What it can't do
Identical exact keyword Text Claims rung 1 Reach Shopping, or hold while ineligible
Identical search theme Text Claims rung 2 for Performance Max. Up to 50 per asset group Beat an identical exact keyword. Ties with an identical phrase or broad keyword go to Ad Rank
Brand exclusion, Shopping allowed Text Takes Performance Max out of Text on excluded brands Hand the search to anyone
Brand exclusion, Shopping not allowed Both Takes Performance Max out of both Hand the search to anyone
Listing group exclusion Shopping Takes a product out, when done in every asset group that includes it Necessarily stop Text ads
Performance Max campaign negative Both Blocks on Search and Shopping inventory Reach Display or Video, where excluded content keywords apply
Account-level negative Both Blocks across Search, Performance Max, App, Shopping, Smart and Local campaigns, up to 1,000 keywords Route a search. It blocks the owner too
Budget, targeting, approvals Both Keep an owner eligible Show that extra spend pays

Tests at a glance

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Item Detail
Smallest visible lift, by conversions in each arm 50: 56%. 100: 40%. 200: 28%. 400: 20%. 800: 14%. 1,600: 10%
Length of a Google experiment At least 4 to 6 weeks, or 1 to 2 conversion cycles with long delay. Ramp-up can run 7 to 14 days
First 7 days Left out of Experiment results. Stats stay in the main campaigns table. Don't count them in the formula
Uplift split Plan 50/50. It can't be changed after scheduling
During Uplift An existing Performance Max campaign serves only 50% of eligible traffic. Existing campaigns in the trial arm are measured running beside it, which captures traffic shifting
Reading Uplift Results compare aggregated group totals on a user-level split, so arms can't be read from a CRM. A qualified-lead test must import that outcome as a conversion. Use View Results with Comparable Campaigns
Comparable campaigns Selected by domain, goal and location. The list populates a day after start, is editable until the end, and is excluded from reporting by default. Recheck after any change to goals, domain or location
Spend between arms A 50/50 split doesn't guarantee equal spend. Google treats a discrepancy under 20% as normal
Auto-apply On by default for new experiments, and you can opt out before the end. Results apply only to experiments that end on their end date. Applying a Shopping vs. Performance Max result gives Performance Max all the traffic and pauses the control. Google says it can be turned off from the Report page or a tooltip on the summary card, only for some experiment types
Favorable With a target: more conversions or value at a better CPA or ROAS. With Maximize and no target: higher conversions or value only
Shopping vs. Performance Max Google recommends the same Target ROAS as the Shopping campaign
During any test No avoidable changes to target, budget, assets or feed. Log what you can't avoid. Run no second experiment. Don't cross a sale or season boundary in a before and after or a geo read
Dated brand exclusion At least four weeks, skip the first, no season boundary, read total brand conversions, target unchanged, and label it an estimate
Geo go-dark steps Target only the regions left on, set Presence, keep an individual budget, and cut it to those regions' historical share of spend when budget-limited. Record each arm's prior share of sales. Google calls the dark regions test geos and the regions left on control geos, the reverse of how many people use those words. Fitness for your account is our inference. Full steps
Conversion Lift based on geography Supports Performance Max. Needs a Google representative, a single-country campaign and a compatible conversion source. Not available to all accounts

How much of the spend you can see

Add up the cost of the term rows and divide by the cost of the Google Search, Search partners and Maps channels for the same dates, from the channel performance report. Report the other channels separately, and call the result an estimate. Google documents cost per term in its API, and Google's page for this view doesn't name a cost column, so check that yours shows one.

Store's term rows add up to $9,100 against $11,200 for those channels, so about 81% of that spend has a visible search. Dividing by its whole $14,000 would report 35% hidden, when $2,800 of it was never a search at all. For what the unlisted searches can be, see "Other search terms".

Where things are in Google Ads

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

To Go to
See Performance Max search terms Campaigns menu, Search terms report, Search terms and landing pages for Performance Max
See Search terms Campaigns, Insights and reports, Search terms. Add Keyword through Columns, Attributes
Read keyword status Keywords in the Campaigns menu, Status column. It shows today
Read ad status Ads in the Campaigns menu, Status column. Add Policy details through the column icon, under Attributes
Read budget status Status column of the campaigns table, which shows today. Search lost impression share from budget is a campaign-level column under Competitive metrics
Check location settings Campaigns, the Settings icon next to the campaign, Locations and Location options
See change history Change history in the Campaigns menu, or select a campaign, then Change History above the table. It covers the past 2 years
Preview a search Tools menu, Troubleshooting, Ad preview and diagnosis
Manage auto-apply Recommendations in the Campaigns menu, Auto-apply settings. Uncheck on the Manage tab and save, or select on the History tab and disable
Add a campaign negative to Performance Max Keywords in the Campaigns menu, Negative keywords tab, plus button, select the campaign
Set channel priorities Account default: Account settings, Channel Prioritization card. It applies to every eligible campaign of that type unless overridden. One campaign: its Settings, expand the card, Use campaign-level settings
Set up Uplift Campaigns, Experiments, plus button, Campaign types card, Performance Max Uplift card, select campaign, Traffic split, Schedule (steps)
Set up Shopping vs. Performance Max Campaigns, Experiments, plus button, Campaign types card, Performance Max vs Search, Display, or Shopping card, Shopping, select campaign, Traffic split, then create or select a campaign (steps)
Open an experiment's auto-apply Experiments in the Campaigns menu, select the experiment (Google's page)

Where Google's pages disagree

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

Topic The disagreement What this guide does
Campaign-level Performance Max negatives One FAQ line on the Shopping onboarding page says they aren't available. Three other pages say they are Treats them as available
Budget by channel The channel report FAQ says it can't be directly controlled. The allocation controls cover some channels Both hold at their scope. Inside Google channels no split exists
Brand exclusion coverage One page adds YouTube search inventory States Search and Shopping, and neither promises nor denies YouTube search
Uplift traffic split The setup page offers a percentage, and the FAQ says split options are unavailable without a feed Plans 50/50
Experiment length One page recommends 2 to 3 weeks where there's no clear winner. The experiments FAQ says at least 4 to 6 Uses 4 to 6

Where to go instead

If columns extend beyond the page, scroll sideways. With a keyboard, focus the table and use the arrow keys.

If you need Go to
Which lead stage to count, or how to import it Should form fills and delayed CRM sales both be Primary? and I made SQL Primary. Why did Search impressions and clicks drop?
To set or reset a target Our Target CPA and Target ROAS guide, and What a Target Does When the Money Runs Out
Creative, auto-generated video or placements Google's asset and suitability documentation. Those are out of this guide's scope
A first campaign on a new account My new Shopping campaign has clicks but no carts

The ownership check to rerun every month

Run this for each account running Performance Max beside anything else. It takes the guide down to the decisions you'll actually repeat.

  1. Read the rows. In the Performance Max view, with Ad format showing, look at searches identical to your exact keywords. Shopping rows are normal. Text rows get the check in When the owner can't hold a search.
  2. Read the owners. Was any owner limited by budget during the period, read by date and not from today's status? Which keywords carry low search volume status? Has a recommendation removed a claimed keyword?
  3. Read the themes. Has anyone added a search theme identical to a search you wanted Search to serve?
  4. Read the credit. Search, Performance Max and account total, dated, on the same conversion action.
  5. Recompute needed share on matured data. If it's high, no budget goes up on reported results.
  6. Read the channels at campaign level. If spend shifted, date it before you touch anything.
  7. Write it down. What you found, what you changed and why, and the next review date. If you can't tell, write that.

Most months at Larkfield, the check changes nothing. Store's report still lists corvane cv-220 pump seal, Ad format Shopping ads, still selling. Generic's keyword has the text ad beside it, and the manager no longer needs to open the row to know.