Google published three changes to its Search campaign testing and planning tools on August 20, 2026: a budget and target test that spans several campaigns at once, experiments that tolerate brand and location settings, and a forecasting tool that can now push its own suggestions into live campaigns without an export step.

The announcement arrived on the Google Ads & Commerce Blog in a post by Brandon Ervin, Director of Product Management, Search Ads. It runs to four short paragraphs. Two of the three changes are described as already available. The third carries a rollout month and no date.

That brevity conceals a fair amount of machinery. The underlying help documentation for both products was rewritten around the same window, and read together the two pages describe a testing system with a narrower set of eligible campaigns than the announcement implies, and a forecasting tool that has been contracting in scope since March.

What changed on August 20

The headline item is a cross-campaign experiment. According to the post, advertisers will be able to test "different budgets and ROI targets across multiple Search campaigns" inside a single A/B test, with the capability rolling out in September. The stated purpose is to show what happens to overall results when campaigns are scaled up rather than adjusted individually.

The second item removes a restriction rather than adding a feature. AI Max experiments can now run with brand and location settings switched on, which means advertisers operating separate branded and non-branded campaign structures no longer have to strip those controls out to run a test.

The third concerns Performance Planner. The tool has always produced forecasts and suggested budget or bid adjustments. What is new is the route from forecast to account. Suggested changes can be reviewed and committed to live campaigns from inside the planner itself.

Dan Taylor, Vice President, Global Ads at Google, amplified the announcement on LinkedIn the following day, framing it as a response to advertiser requests: "We heard the feedback that advertisers want even more ways to direct AI Max", his post opens. The post repeats the three items and adds one detail absent from the blog, describing the multi-campaign A/B test as rolling out in September rather than simply arriving then.

How AI Max experiments actually work

The experiment format Google is extending is not a conventional split test. According to the Google Ads help documentation for AI Max experiments, the mechanism "diverts traffic and budget within an existing campaign instead of creating a copy". A percentage of an existing Search campaign runs with the AI Max toggle off and forms the control arm. The remainder runs with it on and forms the trial arm.

That design produces three stated advantages over the older custom experiments route, all of them derived from the single-campaign structure. Results arrive faster because the traffic is not split across two campaign entities. Setup errors and synchronisation problems are reduced because both arms are updated in the same place. The learning period is shorter because the traffic never leaves the campaign.

Setup runs through the Experiments section of the Campaigns menu, then the plus button, then Campaign features and settings, then AI Max for Search campaigns. By default the experiment activates both AI Max components at campaign level: search term matching, which governs which queries a campaign becomes eligible for, and asset optimisation, which governs what the resulting advertisement says. Search term matching can be disabled at ad group level and asset optimisation at campaign level, with those choices applying only to the trial arm. URL exclusions can be set at campaign level and URL inclusions at ad group level.

Results appear under the Experiments tab with an expanded summary view. Three routes exist for adopting the outcome: an option selected at setup that applies the changes automatically if results are favourable, a manual Apply action from the Experiments table after the test ends, or enabling AI Max directly from campaign settings.

Google also documents the difference between this flow and the older one. The new experiments start automatically; custom experiments have to be started by hand. Custom experiments remain the only option for Search campaigns targeting the Display Network, and they expose reporting through the report editor and the campaigns page as well as the experiments platform. Trial campaigns created through the custom route may need a ramp-up period; the in-campaign version is described as reaching a readable result sooner.

The eligibility gates nobody announced

The same help page lists six conditions that block an AI Max experiment from being created through the new flow. A campaign cannot have text customisation enabled, the setting formerly labelled automatically created assets. It cannot target the Display Network, sit inside a portfolio bidding strategy, use shared budgets, use bidding exploration, or already have an active experiment running.

Five of those six also apply to AI Max experiments built through the custom experiments route. Only the Display Network restriction is specific to the new flow.

Two of the exclusions deserve attention alongside the announcement. The shared budget restriction sits awkwardly next to a feature whose stated purpose is testing budgets across multiple Search campaigns; the documentation does not explain how a cross-campaign budget test interacts with a rule barring campaigns on shared budgets from the experiment flow. And the text customisation exclusion has a September consequence attached to it, because campaigns running automatically created assets are the exact population scheduled for automatic conversion to AI Max on September 1, 2026.

A separate note on the same page complicates the brand controls story. Some Search campaigns carry legacy configurations that limit experiment features, with the documentation giving the example of a brand list exclusion applied to a non-branded campaign. Where that happens, the documented remedy is to build a fresh campaign without the legacy features and run the experiment there. The August 20 post describes brand and location settings as no longer an obstacle to testing. The help page describes at least one brand configuration that still is.

Performance Planner moves from forecast to execution

Performance Planner arrived in Google Ads as a planning aid that identified spend levels likely to drive incremental conversions, distributing a fixed monthly budget across campaigns to maximise conversions. Its forecasting mechanics have not changed much since. The tool draws on billions of search queries, refreshes roughly every 24 hours, and simulates relevant auctions over the preceding seven to ten days, incorporating seasonality, competitor activity and landing page signals. Forecast accuracy is checked by comparing projections against eventual campaign performance.

Two behaviours in the documentation are worth isolating. The planner can propose a budget of zero for a campaign or portfolio, an outcome the documentation attributes to the campaign not contributing to efficient spend distribution, with pausing described as the implied consequence. And the tool now includes conversion delay estimates for Search and Performance Max impact estimates, accounting for the interval between a click and the completed conversion action, which changes the traffic and conversion projections that bidding and return targets are set against.

Capacity is stated precisely: up to 10,000 campaigns per plan, with support for manager accounts.

The implementation step is the part that changed. Once a forecast is generated, an Apply suggested changes control appears in the top right of the interface. Selecting it surfaces the proposed adjustments campaign by campaign, budget or bid, with the option to deselect any campaign meant to be left alone. Confirming updates the live campaigns immediately. The resulting changes are visible, and reversible, through the Bulk actions interface.

That last detail is the substantive one. A planning tool that exported recommendations left a human decision between the model and the account. A planning tool with an apply button removes that step for anyone who takes it.

Who is allowed to plan

Performance Planner eligibility is set out in three separate tables, and the thresholds are low but exclusionary.

Search, standard Shopping and Performance Max campaigns qualify if they have not changed bid strategy in the past ten days, have recorded at least three clicks and at least ten impressions, have produced at least one conversion or conversion value, and have spent above zero in the past seventeen days. Portfolio campaigns must all be linked to a shared budget covering every campaign using that portfolio bid strategy.

Demand Gen campaigns run on a seven-day clock instead of ten, with a fourteen-impression minimum for target CPA, maximise conversions and maximise clicks strategies, and a five-conversion minimum over seven days for target ROAS and maximise conversion value. Several structural conditions apply: no product feed, no flighted scheduling, no mixed bidding strategy, and no participation in an experiment. Key metric selection is constrained by bid strategy, with clicks unavailable to conversion-based strategies and conversions unavailable to maximise clicks.

App campaigns need ten days of runtime, at least ten conversions in the last ten days, an install or engagement bidding strategy using target cost per install or target cost per action, and a status other than limited by budget.

The ineligibility list repeats three of those conditions in plain form. Campaigns that changed bid strategy in the last ten days are out. Portfolio campaigns not linked to a qualifying shared budget are out. And campaigns that are part of an experiment are out.

That third exclusion is the one that sits oddly against the rest of the August 20 announcement. Google presented experimentation and planning as complementary capabilities in the same post. The documentation states that a campaign in an experiment cannot be forecast in Performance Planner. A campaign running an AI Max test is, for the duration of that test, outside the planner.

A forecasting tool with a shrinking remit

Performance Planner is also smaller than it was. According to the help page, the tool "no longer supports planning for display and video campaigns" as of March 9, 2026, along with any plan built on impression share, top impression share or absolute top impression share as the primary metric. Existing plans meeting those descriptions cannot be viewed or edited. That removal was documented quietly in the Help Center rather than through a product announcement, and it narrowed the tool to campaign types organised around conversion outcomes rather than reach.

The same page carries advisory language about market volatility, recommending weekly rather than monthly or quarterly planning cycles when markets are unsettled, and noting that the seasonal model has been updated for current conditions. Forecasts reflect the seven to ten day window, so any market disruption inside that window is already priced into the projection.

Where this sits in the AI Max sequence

The August 20 post is the latest entry in a product line that has been assembled in public over sixteen months, mostly by adding controls after the fact.

AI Max for Search entered global open beta on May 6, 2025, with an implementation guide claiming 14% more conversions or conversion value at a similar cost per acquisition or return on ad spend. Independent testing complicated that number quickly. Analysis shared in August 2025 found that 99% of impressions across roughly 30,000 AI Max search terms produced no conversions. A study of more than 250 retail campaigns published that November found AI Max delivering conversions at approximately 35% lower return on ad spend than traditional match types in the same campaigns.

Controls followed. Text guidelines, capping term exclusions at 25 and messaging restrictions at 40 per campaign, were introduced in September 2025 and reached all advertisers globally in February 2026. A native branded search toggle appeared inside AI Max in June 2026, offering three settings covering all relevant searches, brand inclusions and exclusions, and unbranded queries only. AI Brief, a natural-language instruction layer, was announced on April 30, 2026, the same day AI Max reached standard Shopping campaigns.

Experimentation infrastructure arrived on a parallel track. The September 10, 2025 blog post announcing global AI Max availability described "new one-click experiments, built directly into your campaign flow", and was written by the same author under a different job title, then Director of Product Management, Google Ads. On May 13, 2026, Google Ads API v24.1 added ADOPT_AI_MAX and ADOPT_BROAD_MATCH_KEYWORDS as formal experiment types, alongside direct retrieval of experiment statistics. The API documentation describes ADOPT_AI_MAX as an intra-campaign workflow that creates no additional campaign, matching the mechanism the help page now describes for the interface.

Why the timing matters

Three separate changes land on the same calendar page, and the testing tools announced on August 20 are the newest of them.

Google began rolling out its bidding target optimisation change on August 17, 2026, three days before the announcement. That change pushes budget-limited campaigns running Target CPA or Target ROAS toward their stated targets across Search, Shopping, Performance Max, Demand Gen and Travel, and it reaches targets set at ad group level as well as campaign level. The rollout is staged over several weeks rather than applied at once.

On September 1, 2026, campaigns running automatically created assets or the campaign-level broad match setting convert automatically to AI Max for Search. Google stopped accepting new campaign-level broad match configurations ahead of that date. The five-month reprieve granted to Dynamic Search Ads in June, which moved that automigration to February 2027, does not extend to the other two settings.

The multi-campaign A/B test arrives in the same month.

For an advertiser trying to read a test result in September, three variables move at once: bidding behaviour recalibrating under a staged rollout, a population of campaigns converting to a different targeting model, and a new experiment format measuring budget changes on top of both. Attribution of any performance movement to any one of those causes becomes an exercise in inference rather than measurement. The window also closes roughly eleven weeks before fourth-quarter peak trading.

Measurement pressure is not hypothetical. Invalid traffic detection firm Lunio published a study on August 12, 2026 covering more than 414 million clicks recorded between October 2025 and June 2026, finding that retail search campaigns running AI Max were exposed to 72% more invalid traffic than search campaigns without it, and that AI Max accounted for 68% of invalid clicks detected across its Google search dataset. A test that measures conversion outcomes without accounting for click validity measures something, but not necessarily incrementality.

What the announcement leaves unstated

The August 20 post contains no performance figures. No conversion uplift is claimed for the cross-campaign test, no accuracy figure for the Performance Planner forecasts, and no adoption number for AI Max experiments. That is a departure from the pattern of AI Max announcements, which have generally carried a percentage: 14% at launch in May 2025, and 7% in the April 15, 2026 post that declared the format out of beta and set the migration schedule.

Nor is geographic or language scope stated. AI Max reached global availability across Google Ads, Google Ads Editor, Search Ads 360 and the Google Ads API in September 2025, but several subsequent features have shipped English-first, including AI Brief. Nothing in the August 20 post addresses whether the multi-campaign test follows that pattern.

The September rollout has no date attached, and no statement covers whether it is staged or simultaneous.

What is stated, and what practitioners will encounter first, is a button. Performance Planner forecasts now end in an action rather than a file. For teams managing budget across hundreds of campaigns, the distance between a model's recommendation and a live account has shortened to a single confirmation step, with the Bulk actions interface as the only record of what moved.

Timeline

Summary

Who: Google, in a post by Brandon Ervin, Director of Product Management, Search Ads, amplified on LinkedIn by Dan Taylor, Vice President, Global Ads at Google. The changes affect advertisers running Search campaigns in Google Ads, and in particular those managing budgets across multiple campaigns or testing AI Max.

What: Three changes to testing and planning tools. A single A/B test able to span multiple Search campaigns and compare different budgets and return targets. AI Max experiments that can run with brand and location settings enabled. Performance Planner suggestions that can be applied to live campaigns in one click, reviewable and reversible through the Bulk actions interface. Supporting help documentation sets out an in-campaign experiment mechanism that splits traffic between control and trial arms, six conditions that block experiment creation, and eligibility thresholds excluding campaigns already running an experiment from Performance Planner.

When: The announcement was published on August 20, 2026. The brand and location capability and the Performance Planner apply step are described as available. The multi-campaign A/B test is scheduled for September 2026, with no specific date given.

Where: Inside Google Ads, through the Experiments section of the Campaigns menu for AI Max tests and through the Performance Planner interface for forecasts. No geographic or language restrictions were stated.

Why: Google presents the changes as a response to advertiser requests for more direction over AI Max and less guesswork in campaign optimisation. The timing places them alongside two other changes: a staged bidding recalibration that began on August 17, 2026 for budget-limited campaigns on target-based strategies, and the September 1, 2026 automatic conversion of automatically created assets and campaign-level broad match campaigns to AI Max for Search. For media buyers, the practical consequence is that September will supply test results drawn from a period in which bidding behaviour, campaign targeting models and the testing apparatus itself are all in motion, roughly eleven weeks before fourth-quarter peak trading.