Performance Max in 2026: a campaign type that finally talks back
Performance Max launched in 2021 as a deliberate trade-off: surrender granular control to Google’s AI, get reach across Search, YouTube, Display, Discover, Gmail and Maps in return. For two years, the trade-off felt one-sided. Advertisers handed over budget and creative, received a single ROAS number, and were told to trust the system. Search Engine Land documented the frustration repeatedly through 2023, with practitioners like Mike Ryan publishing custom GAQL queries just to recover information that should have been native.
That era is closing. Between late 2023 and early 2026, Google shipped a series of features that turned Performance Max from a black box into a controllable, evidence-rich campaign type. Account-level negative keyword lists, brand exclusions, search themes, channel-level reporting, asset performance ratings, and a dedicated diagnostics tab now give operators something resembling the levers they had in Standard Search. The campaign is still AI-driven, but the AI is now answerable.
This article is the operator’s guide to Performance Max as it actually works in 2026. It covers what each diagnostic warning means, how to read the asset rating system, how search themes and brand exclusions interact, why the channel split you see in the UI is misleading and what Mike Ryan’s PMax Insights script reveals instead, how the 6-week learning period actually behaves, when to consolidate or split asset groups, how to adjust target ROAS without resetting learning, and the Tinuiti-documented pattern of running a Standard Search brand campaign in parallel to capture branded traffic at fair CPCs.
The diagnostics tab: what each warning actually means
Google added the diagnostics tab to Performance Max in 2023 and expanded it through 2024 and 2025. It sits inside the campaign view under Insights and reports, and it is the first place to look when a campaign is not serving as expected. The tab runs eight checks in sequence: campaign status, billing, policy, conversion tracking, budget, bid strategy, audience signal, and asset group readiness. If all eight clear, you see a green “ready to serve” state. If any fail, the campaign either does not serve at all or serves at a fraction of its potential.
Reading the eight checks in order
The first three checks are binary and severe. Campaign status confirms the campaign is enabled and not paused, removed, or scheduled outside its run dates. Billing confirms the account has a valid payment method and no outstanding suspension. Policy confirms no asset, headline, description, or final URL is disapproved. A single disapproved asset can pull an entire asset group offline, which is one of the most common silent failure modes Search Engine Journal has documented.
Conversion tracking is the fourth check and the one most likely to mislead. Diagnostics will tell you tracking is “set up” if the global site tag fires, but it cannot tell you whether the conversion actions you have selected as primary are the right ones for the campaign objective. A common mistake: a lead-generation Performance Max campaign with “Page view” set as the primary conversion. The campaign serves, the bid strategy runs, the AI optimizes for page views. ROAS looks healthy. Sales pipeline is empty. Diagnostics says everything is fine.
Budget is the fifth check. If your daily budget is too low for the bid strategy to spend, diagnostics will flag “limited by budget”. This is not a hard error, the campaign still serves, but it tells you the AI cannot reach the bid level it considers optimal. The remediation is either to raise the budget or to lower target ROAS, not both at once.
Bid strategy is the sixth check. Diagnostics flags strategies that are misconfigured, for example a target ROAS that has been set so high relative to historical performance that no impressions can be won at that price. This is the warning that experienced operators ignore the most often, because it sounds like advice rather than an error. It is an error: the campaign is not serving meaningfully.
Audience signal is the seventh check. Performance Max accepts audience signals (your customer match lists, custom segments, in-market audiences) as hints, not constraints. Diagnostics flags asset groups with no audience signal at all, which the AI treats as “go anywhere”. For new accounts with no historical conversion data, this is the difference between a 6-week learning period that converges and one that flounders.
Asset group readiness is the eighth and most actionable check. It cross-references the assets you have provided against Google’s minimum and recommended counts. The minimums are deceptive: Google will accept a campaign with one image, one logo, three headlines, two descriptions and a video. The recommended counts, which the diagnostics tab pushes you toward, are 20 images, 5 logos, 15 headlines, 5 long headlines, 5 descriptions and 5 videos. Asset groups serving below the recommended counts will see the asset performance ratings stuck at “Pending” indefinitely, because the AI cannot run enough creative combinations to build statistical confidence.
Warnings that look minor but are not
Three diagnostics warnings appear cosmetic and are routinely ignored. They should not be. The first is “Add more text assets”. Search Engine Land’s analysis of underperforming PMax campaigns in 2024 found that text-asset starvation was the single most common factor in campaigns stuck below 0.5x ROAS, because the Search and Discover channels need headline diversity to compete in auctions where Standard Search campaigns supply 15 by default.
The second is “Add a video”. Performance Max will auto-generate a video from your images and headlines if you do not supply one. The auto-generated video is poor and is served selectively on YouTube, where it converts badly. Optmyzr’s 2025 study of 2,400 Performance Max campaigns found that supplying at least three native videos lifted YouTube channel conversion rate by a median 38% versus auto-generated alone.
The third is “Add audience signal”. A campaign without audience signal is not broken, but it is operating without the prior that would let the AI converge faster. For accounts with mature first-party data, customer match lists are the highest-leverage signal, followed by lookalikes built from converters, followed by in-market segments. Google’s own documentation acknowledges that audience signals reduce learning time, which is to say the campaign reaches stable performance sooner.
Asset performance ratings: Best, Good, Low, Pending
Each text, image and video asset inside a Performance Max asset group receives one of four ratings: Best, Good, Low, or Pending. The rating is relative within the asset group, not absolute. A “Best” image in one asset group might be a “Low” image in another. This relativity is the first source of confusion for operators coming from Standard Search, where Quality Score is an absolute 1 to 10.
What each rating means and what to do
Best means the asset is in the top tier of performers within its asset group. The AI is selecting it disproportionately often when assembling ad combinations, and those combinations are converting. Operationally, you do nothing to a Best asset except clone its style when adding new assets. If your “Best” image is a lifestyle shot with a specific color palette, your next batch of images should explore that palette further.
Good means the asset is performing in line with the asset group average. It is not a winner but it is not a drag. These are the assets that benefit most from refresh, because moving them from Good to Best is where the easiest gains live. Replace the headline with a sharper variant, recrop the image to a different aspect ratio, retest.
Low means the asset is dragging the asset group’s overall performance. The AI is either selecting it rarely (in which case it is not really hurting you) or selecting it and seeing poor downstream conversion. Low-rated assets should be removed and replaced, not edited in place. Editing in place restarts the rating from Pending and does not always recover.
Pending is the most misread rating. It does not mean the asset is bad. It means the asset has not accumulated enough impressions for the AI to reach a confidence threshold. New assets always start as Pending. The threshold is undocumented but practitioners on Search Engine Land have triangulated it at roughly 5,000 to 10,000 impressions for text assets and 50,000 to 100,000 for images, varying with budget and competitive density. A small campaign on a low budget will see assets stuck at Pending for weeks. This is not a problem to solve, it is a constraint to accept.
The trap of acting too early on ratings
Operators frustrated with Pending often delete assets and start over. This is counterproductive. Asset rating is a function of cumulative impressions across the asset’s lifetime in the group. Deleting and re-adding the same asset resets the counter to zero. The Optmyzr study cited above found that asset groups with the lowest churn (operators who waited at least 4 weeks before culling) outperformed high-churn groups by 22% on conversion rate. The discipline is to add assets in batches, wait for ratings to stabilize, then act.
Search themes: keyword-style steering since 2024
Search themes were rolled out globally in early 2024 and represent the most important concession Google has made to advertisers who missed having keyword input in Performance Max. A search theme is a free-text phrase, up to 25 per asset group, that tells the AI what queries you believe are relevant. They are signals, not commands: they do not behave like Standard Search keywords with match types, and they do not exclude queries that fall outside them.
How search themes interact with the AI
Google’s documentation describes search themes as “filling gaps” in the AI’s automatic query discovery. In practice, they have two effects. First, they accelerate exploration: a theme like “vintage leather messenger bag” will cause the AI to test queries in that semantic neighborhood within days rather than waiting for organic discovery. Second, they raise the prior weight of those queries in the bidding decision, meaning the AI will bid more aggressively on close matches than it would without the theme.
What search themes do not do is constrain. If the AI’s query discovery surfaces “designer handbag sale” and your themes are all about messenger bags, the campaign can still serve on “designer handbag sale” if Google’s models predict it will convert. This is why search themes do not replace negative keywords. They are an accelerant, not a fence.
Search themes and Search Standard interaction
Google’s documented behavior, confirmed in the Performance Max release notes from October 2024, is that Standard Search campaigns take priority over Performance Max for queries that exactly match a Standard Search keyword, regardless of search theme. This is the rule that makes the brand-campaign-in-parallel pattern work, which we cover later. If you have a Standard Search campaign with the exact-match keyword [your brand], that campaign serves the query, not Performance Max, even if “your brand” is also a Performance Max search theme.
Channel reporting and its limitations
Channel-level reporting in Performance Max landed in the UI in late 2024 and went global through 2025. It shows you a split of impressions, clicks and conversions across Search, YouTube, Display, Discover, Gmail and Maps. It is genuinely useful for the first time. It is also incomplete in ways that matter.
What the UI shows and what it hides
The channel reporting view in Google Ads gives you metrics by channel for a campaign, but not by asset group within a campaign. If a campaign has three asset groups (one for branded queries, one for shopping intent, one for awareness), the UI will tell you how the campaign performed across YouTube versus Search, but not how each asset group contributed. For multi-asset-group campaigns, this is a meaningful blind spot.
The reporting also does not break out Search partners separately from Google Search proper, conflates Maps and Search in some views, and does not expose the Discover-versus-Display split with full reliability. Search Engine Journal’s coverage in March 2025 noted that practitioners running parallel measurement (server-side conversion tracking with channel attribution at the source) routinely found 10 to 20 percentage point discrepancies between the UI’s channel split and what their first-party data showed.
Mike Ryan’s PMax Insights script
Mike Ryan, working through Smarter Ecommerce and the PMax Insights project, published a Google Ads script in 2023 that infers a more granular channel split by querying GAQL endpoints the UI does not expose. The script joins data from the asset_group_product_group_view and the campaign_search_term_insight resource to reconstruct what Performance Max is doing at a level the native reporting hides. It is the de facto reference tool for practitioners who manage substantial PMax spend and is referenced regularly in Search Engine Land coverage.
The script’s headline output is a Search-versus-Shopping split that the native UI does not produce reliably. For e-commerce accounts with a Merchant Center feed attached, this split tells you what fraction of spend is going to Shopping inventory ads versus text Search ads. The two channels behave very differently: Shopping is feed-driven, text Search is asset-driven, and optimizing one without knowing the other’s share is the kind of mistake that leaves money on the table for months.
Search terms reporting: progress and remaining gaps
Performance Max search terms reporting was added in 2023 and refined through 2024. It is not the equivalent of Standard Search’s search terms report. The differences matter for anyone migrating expectations from one to the other.
What you actually see
The Performance Max search terms view groups queries into “search categories” rather than exposing every individual query. A category might be “leather messenger bag” and contain dozens of underlying queries. Google’s stated rationale is privacy and noise reduction. The practical effect is that you see less than you do in Standard Search, where the search terms report exposes every query that triggered an ad above a low impression threshold.
You do see, per category, the number of impressions, clicks and conversions, and the broad trend over time. You do not see the individual queries by default, although Google began rolling out an expanded view in late 2025 that exposes top queries within high-volume categories for some accounts. This rollout is uneven and category-dependent.
What this means for negative-keyword work
You cannot do negative-keyword work in Performance Max with the precision you can in Standard Search. You cannot identify the single misfit query, add it as a negative, and move on. What you can do is identify category-level patterns (“we are spending on a category that does not match our intent”) and act at the category level, which usually means adding negative keywords at the account level to suppress the entire pattern.
Account-level negative keyword lists: 2023 launch, 2025 expansion
Account-level negative keyword lists arrived in Performance Max in 2023 and were the first meaningful negative-keyword control the campaign type ever supported. They are managed under Tools and settings, Shared library, Negative keyword lists, and applied at the account level rather than at the campaign level. In March 2025, Google expanded the per-list limit from 1,000 keywords to 10,000, a tenfold increase that materially changed how operators can use them.
Why account-level only matters
Standard Search campaigns let you add negative keywords at the campaign or ad group level. Performance Max, until recently, supported neither. The only way to exclude a query was through an account-level list. This is a cruder instrument than Standard Search practitioners are used to: a negative added to suppress wasteful queries in one Performance Max campaign also applies to every other PMax campaign in the account. For accounts running multiple PMax campaigns with different intent profiles, this is a real constraint.
Google began limited testing of campaign-level negative keywords for Performance Max in late 2024 and rolled it out more broadly in 2025, but the implementation is still partial and the account-level list remains the primary tool. Operators should treat the account-level list as the workhorse and campaign-level negatives as a refinement when available.
What goes on the list
The 10,000-keyword limit, up from 1,000, lets you do something that was previously impossible: add comprehensive negative coverage for irrelevant verticals, competitor brand terms you do not want to bid on, free-intent queries (“free”, “tutorial”, “DIY”), and job-seeker queries (“careers”, “jobs”, “salary”). The pre-2025 limit forced hard prioritization. The post-2025 limit lets you build the kind of comprehensive negative library that mature Standard Search accounts have always maintained.
Brand exclusion at campaign level: 2024
Brand exclusion landed at the campaign level in 2024 and is distinct from negative keywords. A brand exclusion tells Performance Max not to serve on queries that mention specific brands, regardless of whether those queries would otherwise match the campaign’s targeting. The mechanism uses Google’s brand list, which is a curated list of brand entities maintained by Google rather than free-text strings supplied by the advertiser.
Why this matters more than it looks
The most common use of brand exclusion is to stop your Performance Max campaign from cannibalizing your Standard Search brand campaign. Without exclusion, a Performance Max campaign with a Merchant Center feed will serve on queries containing your own brand name, often at higher CPCs than your Standard Search brand campaign would pay. Tinuiti’s 2024 analysis of brand-cannibalization patterns in Performance Max found that excluding your own brand from PMax and running a parallel Standard Search brand campaign saved a median 31% on branded-query CPC across the accounts they audited.
The second use is to exclude competitor brands you do not have authorization to bid on, or that consistently produce low-quality clicks. Brand exclusion handles this more cleanly than a sprawling negative keyword list, because Google’s brand entity matches variant spellings and translations automatically.
URL expansion controls
URL expansion is the feature that lets Performance Max send users to URLs other than the final URLs you have explicitly listed. By default, it is on. The campaign can crawl your site and send traffic to any indexable page it judges relevant to a query. For e-commerce sites with a clean structure, this is often beneficial. For sites with a complicated URL space, archive pages, or low-conversion landing pages, it is a liability.
The three-state control
URL expansion has three settings. Fully on (default) lets the AI choose any indexable URL. Excluded URLs lets you block specific URL patterns from being chosen. Off lets you restrict the campaign to only the final URLs you have provided in the asset group. Most operators leave it on. Most operators should not.
The right default for accounts with strong landing-page optimization (a deliberate set of conversion-tuned pages) is to turn URL expansion off and rely on the asset group’s final URLs only. Search Engine Land’s coverage in 2024 documented case studies where turning URL expansion off lifted conversion rate by 15 to 30% in accounts where the AI had been routing traffic to blog posts and category pages that did not convert.
The right default for accounts with shallow landing-page optimization (a homepage and a few product pages) is to leave URL expansion on but use the excluded URLs setting aggressively to block low-converting paths.
Asset group splitting strategies
Each Performance Max campaign can hold up to 100 asset groups, but the practical limit is far lower. Asset groups are the unit at which the AI builds creative combinations and learns. Splitting too granularly starves each group of conversion volume. Consolidating too aggressively gives you one undifferentiated mass and prevents you from steering the campaign toward different audiences or product sets.
Three sensible splitting axes
The first axis is audience theme. Split asset groups by the customer profile you are speaking to: B2B versus B2C, enthusiast versus casual buyer, gift-giver versus self-purchaser. Each asset group gets headlines, images, descriptions and audience signals tailored to its theme. The AI then assembles theme-coherent ads.
The second axis is product category. For e-commerce, split asset groups by the major product taxonomy: shoes, bags, accessories. Each asset group gets product-specific imagery and copy. Pair this with listing groups in the Merchant Center feed for tight Shopping integration.
The third axis is funnel stage. Split asset groups by where in the journey you are intercepting the user: awareness (broad themes, educational copy), consideration (comparison, social proof), conversion (specific offers, urgency). Use search themes and audience signals that match the stage.
Avoiding the over-split trap
The temptation, especially for operators coming from Standard Search and its dense ad group structures, is to split into ten or fifteen asset groups per campaign. This rarely works in Performance Max. The AI needs conversion volume per asset group to learn. A practical floor: each asset group should reasonably expect at least 10 conversions in its first 4 weeks, ideally more. Below that threshold, ratings stay at Pending, optimization signal is thin, and the campaign behaves erratically.
The 6-week learning period
Performance Max requires a learning period that practitioners and Google’s own documentation peg at approximately 6 weeks. This is not a fixed timer but a behavioral pattern: in the first 4 to 6 weeks, the campaign explores broadly, performance is volatile, and the AI is building a model of which audience-creative-channel combinations convert. From week 6 onward, performance typically stabilizes, asset ratings consolidate, and channel-mix patterns become readable.
How to read the learning period
Days 1 to 7: spend ramps, conversion volume is low or zero, search themes start showing exploratory behavior. Do not optimize. Do not change target ROAS. Do not pause anything.
Days 8 to 21: first conversions arrive, asset ratings begin to populate from Pending toward Low and Good. Channel mix swings widely. Continue not to optimize. The most common mistake is to act on a 14-day window, see a 0.4x ROAS, and either pause or aggressively raise target ROAS. Both reset the learning.
Days 22 to 42: the campaign converges. Asset ratings stabilize. Channel mix settles. ROAS approaches a steady state. This is the window in which you start reading the data and identifying first optimizations.
Day 42 onward: the campaign is in steady state. Optimizations should be incremental and one variable at a time. Target ROAS adjustments of 5 to 10% are tolerated without resetting learning. Asset additions in batches of 3 to 5 are tolerated. Mass asset replacements, target ROAS shifts above 15%, or audience signal overhauls trigger a new learning period.
Consolidating versus splitting based on conversion volume
The decision to consolidate asset groups (or campaigns) or to split them is governed by conversion volume more than by intuition about audience or product differentiation. Google’s bid strategies, including Maximize Conversion Value with target ROAS, require statistical signal, and that signal comes from conversions.
The volume thresholds
An asset group generating fewer than 30 conversions in a 30-day window is not producing reliable enough signal for the AI to optimize within it. Below 10 conversions, the asset ratings are essentially noise. Optmyzr’s 2025 analysis of asset group conversion volume versus performance stability found that the variance of weekly ROAS dropped sharply at around 30 conversions per asset group per month and continued to drop, more slowly, up to 100.
The implication is operational. If you have an asset group at 8 conversions per month, do not optimize it as if its data were meaningful. Consolidate it into a sister asset group or accept that you are running it for reach rather than for performance. If you have an asset group at 80 conversions per month, you have signal. Optimize it.
When to split
Split an asset group when its conversion volume can support the split. A 200-conversion-per-month asset group can be split into two 100-conversion asset groups without losing signal in either. A 40-conversion asset group split into two 20-conversion groups will see both halves regress to noise. Split when you have observed a clear behavioral difference between two segments within the asset group (different audience signals converting at different rates, different product categories with different ROAS profiles), and split only when you can provision enough conversions on each side.
When to consolidate
Consolidate when conversion volume is thin and asset ratings are stuck at Pending across multiple groups. Consolidate when two asset groups have converged to similar ROAS and similar channel mixes despite nominally different themes (the AI has effectively decided they are the same). Consolidate when budget pressure forces you to concentrate spend.
Adjusting target ROAS without resetting learning
Target ROAS is the primary lever in Maximize Conversion Value bidding. Adjusting it correctly preserves learning. Adjusting it incorrectly resets it. The rule, based on Google’s own documentation and confirmed by practitioner testing reported in Search Engine Journal, is the 15% rule: changes of up to 15% per adjustment do not trigger a new learning phase. Changes above 15% do.
The mechanics of the 15% rule
If your campaign is running at target ROAS 400% (4.0x) and converging steadily, raising target ROAS to 460% (a 15% relative change) preserves learning. The bid strategy adjusts within its existing model. Raising it to 500% (a 25% relative change) signals to the system that you want a materially different point on the volume-versus-value curve, and it restarts the learning behavior.
The same applies in the other direction. Lowering target ROAS from 400% to 340% (a 15% drop) is absorbed. Lowering it to 300% (a 25% drop) restarts learning.
The cadence that works
For a steady-state campaign, target ROAS adjustments of 5 to 10% every 2 to 3 weeks are the correct cadence. This lets you push performance up gradually as the AI optimizes, or pull it down to chase volume, without paying the cost of resetting. Operators who adjust weekly, especially with larger swings, tend to keep the campaign perpetually in a learning state and never see steady performance.
The exception is when business conditions change materially: a new product launch, a seasonal peak, a competitor exit. In those cases, a larger target ROAS shift is warranted and the learning reset is the cost of doing business.
The Standard Search brand campaign as PMax companion
The pattern of running a Standard Search brand campaign in parallel to Performance Max is one of the most consistently profitable optimizations available in 2026, and it is well documented in Tinuiti’s published analyses. The mechanism rests on a specific Google rule: when a query exactly matches a keyword in a Standard Search campaign, that campaign serves the query, not Performance Max, regardless of search themes or audience signals on the PMax side.
Why the pattern works
Branded queries (someone searching for your exact brand name or a close variant) are the cheapest, highest-converting traffic any account sees. Performance Max, left alone, will bid aggressively on these queries because they convert well, and it will pay more than necessary to win them, because the AI does not have a separate “this is a branded query, bid less” model. By default, the AI just sees high conversion probability and bids accordingly.
Running a dedicated Standard Search brand campaign with exact-match brand keywords intercepts those queries at much lower CPCs (often a fraction of what PMax would pay). Tinuiti’s reported median saving was 31% on branded-query CPC. Combine this with brand exclusion in Performance Max (so PMax explicitly does not bid on your brand) and you get clean separation: PMax handles non-branded traffic, Standard Search handles branded traffic, and you pay the right price for each.
The implementation pattern
Build a Standard Search campaign with a single ad group and exact-match keywords for your brand and its close variants. Set the bid strategy to Target Impression Share (top of page, 95% or higher) to dominate the auction at low cost. Add your brand to the brand exclusion list in your Performance Max campaign. Verify after two weeks that the Standard Search brand campaign is capturing branded query volume and that PMax is no longer serving on those queries.
The pattern does not work in reverse. You cannot run Performance Max as your branded-traffic engine and use Standard Search for non-branded. The AI’s bidding for branded queries inside PMax is too expensive, and Standard Search non-branded campaigns lack the multi-channel reach that makes PMax worthwhile in the first place.
Common mistakes that cost performance
The frequent failure modes in Performance Max are predictable and avoidable. They cluster around six themes.
Optimizing too early. Acting on day-14 data, before the learning period has run, generates change for change’s sake and resets learning. Wait at least 4 weeks before significant intervention.
Confusing primary conversions. Setting “Page view” or “Add to cart” as primary instead of “Purchase” or “Lead” lets the AI optimize for the wrong outcome. Diagnostics will not catch this.
Letting URL expansion route traffic to weak pages. Default URL expansion sends conversion-driving spend to blog posts and category pages. Audit destination URLs in the search categories report and use excluded URLs aggressively.
Skipping audience signal. Empty audience signals double the learning period. Even a basic in-market segment is better than nothing.
Over-splitting asset groups. Five asset groups in a 50-conversions-per-month campaign starves all five. Consolidate to one or two until volume justifies more.
Not running a parallel brand campaign. The branded-query CPC paid by PMax in the absence of brand exclusion plus a Standard Search brand campaign is one of the largest preventable wastes in mature accounts.
Conclusion: Performance Max as a controllable system
Performance Max in 2026 is not the black box it was at launch. The diagnostics tab, the asset performance ratings, search themes, channel reporting, account-level and campaign-level negatives, brand exclusion, URL expansion controls, and the parallel Standard Search brand campaign pattern together give operators enough levers to run the campaign type as a controllable system rather than an act of faith.
The discipline that distinguishes accounts that perform from accounts that do not is procedural: respect the 6-week learning period, change one variable at a time, keep target ROAS adjustments within 15%, populate asset groups to recommended counts, use audience signals, exclude your own brand and run it through Standard Search instead, and read the diagnostics tab as the first step of every weekly review.
The campaign type rewards patience and punishes thrashing. Operators who internalize this end up with PMax campaigns that compound: stable channel mix, converged asset ratings, predictable ROAS, and the ability to scale budget without breaking the model. The features Google shipped between 2023 and 2026 made that outcome reachable. The work of using them well is on the operator.
Sources
- Google Ads Help, Performance Max documentation (account-level negative keyword lists, brand exclusion, URL expansion, search themes, asset performance ratings, diagnostics tab).
- Search Engine Land, coverage of Performance Max feature releases 2023 to 2026 (search themes rollout October 2024, account-level negative expansion March 2025, channel reporting global rollout 2025).
- Search Engine Journal, analyses of Performance Max search terms reporting and learning period behavior, 2024 to 2025.
- Tinuiti, brand cannibalization analysis and the Standard Search brand campaign companion pattern, 2024.
- Optmyzr, 2025 study of 2,400 Performance Max campaigns covering asset rating churn and asset group conversion volume thresholds.
- Mike Ryan, PMax Insights script and Smarter Ecommerce technical documentation on Performance Max channel and Search-versus-Shopping inference.