Is Performance Max Worth It? When It Pays, When It Cannibalizes

Performance Max (PMax) is worth it when it reaches buyers the rest of your media would not. It hurts when it takes space that was already yours: terms your Search campaign covered, branded queries, and visitors who would return on their own. The table below maps each account situation to what you check and the decision each signal points to.

No single row gives a verdict. The table shows where the risk sits, and the decision comes from the test described further down.

The aggregate studies we gathered measure overlap and attributed performance. None measures how much of PMax revenue is incremental. That is why we do not give a benchmark percentage for revenue cannibalization. The number that counts is your account's, measured in the test.

In February 2025, Optmyzr matched PMax search terms against the keywords in Search campaigns across 503 accounts¹. In 91.45% of them, PMax served on terms with text identical to a Search keyword¹. The overlap showed up in every match type, including exact. On performance, most cases came out even.

The result contradicts the rule Google documents. Under that rule, when a query is identical to an exact match keyword, the Search campaign takes priority over PMax. In practice, having the keyword on exact match did not prevent the overlap.

The study has three limitations, and the authors flag all of them. First, differences under 10% were treated as insignificant¹. Second, there was no CPA or ROAS comparison, because Google did not report cost per PMax search term at the time. Third, Search may have been ineligible due to budget or targeting, and that possibility was not analyzed.

Our read: the study measures where the two campaigns compete for the same term. It does not measure whether the sale is new. When conversion rates tie, the question becomes who gets the credit and whether you pay twice for the same buyer. This is a hypothesis, and it is the one the geo test answers.

The same company's larger study, with 9,199 accounts and 24,702 campaigns, points the same way². In it, 82% of advertisers run PMax alongside other campaign types, and PMax campaigns performed worse when they shared the account with Search and Shopping². The study included accounts with at least 90 days of history and monthly spend between $1,000 and $5 million². Both studies measure correlation. Neither uses a control group.

For branded search, containment relies on brand exclusions and campaign-level negative keywords. For remarketing, it relies on the new customer acquisition goal. All three tools limit where PMax serves. None of them tells you how much of the blocked revenue was incremental. That answer comes from the test.

Brand variations are the weak spot. Misspellings, brand plus product name, and brand plus city have no identical keyword in the brand campaign and can fall into PMax. Even so, 97% of advertisers in the Optmyzr study do not use brand exclusion lists². In the same study, accounts using exclusions performed about the same as accounts without them. That is correlation and says nothing about the effect on incremental sales.

For non-brand terms, PMax now accepts campaign-level negative keywords. The limit started at 100 and rose to 10,000 per campaign, matching Search campaigns³. Use negatives for terms Search already converts well and that showed up in the match-up from the previous section. Avoid over-negating early on, because it cuts the campaign's learning volume.

On remarketing, the available data is weaker. No source in this piece measures, with a real sample, how much PMax overlaps with remarketing. What exists is indirect: in the Optmyzr study, the large majority of advertisers use audience signals, and those accounts did worse on every metric except CTR and ROAS, which held steady. You cannot say the signals cause the result.

The slice at stake in remarketing is large. Repeat purchases can account for 5% to 30% of a brand's sales, according to a MarTech article on CRM segmentation for paid media⁴. If PMax serves to that base without distinction, part of that revenue shows up in the dashboard as its conversion.

Audience signals are a starting point and do not restrict reach. PMax can serve to current visitors and customers. If the configured conversion is "purchase" with no distinction, the algorithm hunts for the cheapest conversion, which tends to be someone who had already decided. This is a hypothesis. The new customer acquisition goal fixes part of this, but it depends on the customer list you upload: with a stale list, an old customer gets counted as new.

The PMax channel performance report shows impressions, clicks, cost, and conversions split by Search, Shopping, YouTube, Display, Gmail, Discover, and Maps. It shows you where the budget went and who got credit under the account's attribution model. That is correlation. The report does not answer whether the sale would have happened without the campaign.

Use the report for what it does. It shows whether spend is concentrated in Search and Shopping, where overlap with your other campaigns is most likely, or spread across visual channels. Search terms show how much of the volume is brand. Both inherit the account's attribution model. If the model gives credit to the last paid click, a branded search that would have closed through organic shows up as a PMax win.

Dashboard ROAS misleads at exactly this point. In the Optmyzr study, 51% of advertisers put more than 50% of budget into PMax, and those accounts had the highest ROAS, at 652.03%³. Conversion rate and CPA were mixed. The same study points to a possible conversion and CPA advantage when PMax sits between 10% and 25% of budget³. Both readings are attributed ROAS. The more budget PMax gets, the more conversions it can claim, including ones that would have come through other paths.

Before-and-after comparisons mislead too. When the category grows or a peak season hits, the period with PMax looks better even if the campaign adds nothing. A geo control group solves this, because both halves of the sample go through the same seasonality.

The test splits the country into two groups of similar regions. In one, PMax stays on. In the other, which works as the control group, it is switched off. The rest of your media stays identical in both. The difference in revenue or new customers between the groups, measured in your CRM, is the campaign's incremental effect.

1. 01
   Define the decision metric outside Google Ads. Use revenue or new customers by region, pulled from your CRM or ERP. The conversion the platform reports does not work as a decision metric.
2. 02
   Confirm you have enough volume. In the Optmyzr study, accounts below the 60-conversion threshold performed worse on nearly every metric². The study does not state the time window for that threshold. If your test group lands well below it, add regions.
3. 03
   Build the region groups. Split states, metros, or DMAs into two groups with similar sales history. Check the historical series to see whether the two groups' curves move together. If they do not, redo the split.
4. 04
   Turn off PMax only in the control group. Do it through location targeting. Search, Shopping, and remarketing stay the same in both groups.
5. 05
   Freeze the account. During the test, do not change budget, target CPA or ROAS, assets, lists, or negatives. Every change shifts what is being measured and blends the campaign's effect with the effect of the tweak.
6. 06
   Audit assets and policies before you start. A disapproval or suspension mid-test contaminates the result. If you advertise financial services, track Google Ads financial services verification: the program already covers 42 countries, with expansion to the US and Canada planned for early 2027⁵. The announcement does not list which countries are covered, so confirm your markets before you start.
7. 07
   Run for at least one full purchase cycle for your account. The material behind this piece does not offer a number of weeks that fits every account. Use the average time from first click to sale shown in your CRM. Do not start in the middle of a peak period such as Black Friday.
8. 08
   Compare the test group with what it would have done without PMax. The projection comes from the historical relationship between the two groups. The effect is the gap between actual and projected.

Google Ads also offers Conversion Lift studies. They measure the lift in conversions tracked by Google itself, have eligibility requirements, and do not read revenue from your CRM. Use the study as a second opinion and keep the CRM as the decision criterion.

Once the test closes, you have three numbers: the test group's result, the projected result without PMax, and PMax spend in the test regions. With them, you calculate incremental customer acquisition cost (CAC) and incremental return on ad spend (ROAS). Those two values decide the budget.

• Incremental new customers = new customers in the test group − projected new customers without PMax.
• Incremental CAC = PMax spend in the test group ÷ incremental new customers.
• Incremental ROAS = incremental revenue ÷ PMax spend in the test group.
• Incrementality rate = incremental conversions ÷ conversions PMax attributed to itself in the test group.

The incrementality rate measures cannibalization. Close to 1, what the dashboard shows is mostly new sales. Well below 1, the gap is what PMax took from Search, organic, or natural repeat purchases.

The final decision compares incremental CAC with the maximum CAC your margin can support. To find that ceiling, multiply average order value by contribution margin and by the average number of purchases a customer makes over your planning horizon. If incremental CAC lands below the ceiling, PMax gets budget. If it lands above, pull brand, terms Search already covers, and remarketing out of the campaign and rerun the test. If it stays above even with the exclusions, shut the campaign off.

There is one limit you have to accept. If the difference between the groups is smaller than the weekly swing they already showed historically, the test decided nothing. In that case, the result is inconclusive. Increase the group size or the test period before touching the budget.

Before building the geo test, find out how much of PMax could be cannibalization. You need four reads that already exist in your account and your CRM. If they show a heavy share of Search terms, brand terms, or returning customers, the test is a priority. If they show the opposite, you still need the test, but the risk of paying for sales that were already yours is lower.

1. 01
   Match PMax search terms against the keywords in your Search campaigns. Flag terms with identical text and add up their volume.
2. 02
   Open PMax search terms insights. Add up the volume of the categories that contain your brand name and its variations.
3. 03
   Match PMax conversions against the CRM. Separate new customers from returning ones using the key that ties click to sale.
4. 04
   Check the campaign setup. Confirm whether brand exclusions and negatives are active and whether the new customer acquisition goal uses an up-to-date customer list.
5. 05
   Log the numbers in a spreadsheet. They are the baseline for choosing the region groups and for reading the test later.