Key takeaways
- AI Max experiments now support more realistic testing because brand and location controls can stay in place.
- From September, advertisers will be able to test budget and ROI target changes across multiple Search campaigns in one experiment.
- The main risk is false positive uplift, where conversion volume rises but lead quality, margin or non-brand economics get worse.
- UK advertisers should define commercial pass marks before launching tests, including CPA, ROAS, qualified lead rate and margin thresholds.
- Performance Planner one-click changes speed up implementation, so campaign-level review must happen before recommendations are applied.
AI Max experiments now matter because Google is giving PPC teams a safer way to test automation before it rewrites account economics. That is the right direction, but it also raises the standard for how UK advertisers judge uplift. A weak experiment now creates a false sense of confidence faster than a manual bid change ever did.
The accounts most exposed are already using broad match, Smart Bidding and AI-driven Search expansion. If that sounds familiar, treat this as a testing discipline issue, not a feature announcement. Our earlier piece on AI Max auto upgrade for PPC teams explains why the shift matters: once automation has more permission to find traffic, your control moves from daily bid tweaks to experiment design, conversion quality and campaign structure.
What has changed in AI Max experiments
Google Ads is adding more controlled testing options for Search campaigns and AI Max. From September, advertisers will be able to test budget and ROI target changes across multiple Search campaigns in one A/B experiment, rather than running isolated tests campaign by campaign.
These experiments can also now keep brand and location controls enabled. That matters because many serious Search accounts do not run without those safeguards. A test that forces you to remove brand exclusions or geographic restrictions is not a fair test of how the campaign will run after launch.
Performance Planner is also gaining one-click implementation for budget and bidding plans. Advertisers can review the proposed campaign-level changes, deselect campaigns they do not want to change, apply the plan, and undo it through Bulk Actions if needed.
Why Search and AI Max testing now changes the money
The real value is not that Google has added another experiment button. The value is that PPC teams can now test account-level commercial decisions against a control, especially budget scaling and target changes. That is where the money moves.
Here is the mechanism. If you raise budgets across a group of Search campaigns without a control, you cannot separate genuine incremental demand from cannibalised conversions, brand leakage, query drift or normal market movement. Spend rises, conversion volume rises, and the board assumes the plan worked. Then CPA creeps up over the next month and nobody knows whether the problem is the budget increase, seasonality, auction pressure or Smart Bidding chasing weaker conversions.
Multi-campaign experiments give you a cleaner answer. You can test whether loosening a ROAS target across a portfolio actually produces profitable extra sales, or whether it simply buys more expensive clicks from the same pool of demand. For lead generation, you can test whether a higher budget brings more qualified enquiries, not just more form fills. That distinction is everything.
Budgets are the obvious starting point. In our accounts we regularly see the median account losing roughly a fifth of its search clicks to capped budgets, with a large share of accounts losing more than 10%. That is an opportunity, not a spend more nag. The point is knowing which missing clicks are worth chasing, not raising budgets blindly.
This is where AI Max experiments become useful for grown-up PPC management. If a campaign is budget limited, automated expansion will find more auctions when released. Some will be profitable. Some will be softer, broader and further from the original keyword intent. A proper experiment tells you which side wins after conversion value, lead quality and margin are included.
Brand and location controls stop tests becoming fantasy
The brand and location change is bigger than it looks. UK advertisers often have strict trading areas, store coverage, franchise boundaries, regional pricing, compliance wording or call-handling limitations. A test that ignores those settings produces results you cannot operationalise.
Keeping brand and location controls inside these experiments means PPC teams can test the feature under the same commercial rules the live account must follow. A local service business can keep its catchment area intact. A retailer can prevent AI expansion leaning on brand demand. A multi-location advertiser can stop a national-looking uplift hiding regional waste. Our take on the Performance Max brand leak problem covers why this matters for budgets.
That gives you a better read on incrementality. The question is no longer, did AI Max increase conversions? The question is, did AI Max increase conversions that the business can serve, at a cost and value mix that beats the current setup?
PPC Geeks’ View
The specific problem advertisers will face is false positive testing. Google will make it easier to launch bigger experiments, and weaker teams will treat the experiment result as proof without checking the quality of the traffic behind it. That is how an account scales into worse economics whilst still showing a tidy conversion graph.
We see this most often in lead-gen accounts running broad match with Smart Bidding and thin offline feedback. The experiment variant gets extra latitude, it finds more queries, form volume improves, and the CRM later shows the sales team had more poor-fit enquiries. Google Ads records the win. The business feels the waste.
Do not test AI Max against a vague conversion goal. Test it against the commercial outcome you actually want Smart Bidding to learn.
— Rory Bettany, Senior PPC Account Manager, PPC Geeks
For ecommerce accounts, the trap is slightly different. The experiment looks good on ROAS because brand, returning users or high-margin hero products carry the numbers. Unless you split reporting by brand, new customer value, product margin and query intent, these experiments become a blended average that hides where the extra spend went.
The practical takeaway is simple. Before you launch a test, define the failure condition. For example, reject the variant if non-brand CPA rises by more than your agreed tolerance, if offline qualified lead rate falls, or if spend shifts towards products with weak contribution margin. This is exactly the type of issue we look for in a free Google Ads audit, especially where automation, tracking or campaign structure is changing performance.
What advertisers should do next with AI Max experiments
Start by building an experiment map this week. List every Search campaign that shares budget logic, target logic or the same commercial objective. Do not test one campaign in isolation if the decision you want to make affects a group. A portfolio ROAS decision needs a portfolio test.
- Separate brand from non-brand before the test. If brand activity sits inside the same campaign group, split reporting before you start. If structure prevents that, add brand exclusions or dedicated brand controls where available. A test that wins by harvesting your own demand is not a Search growth strategy.
- Lock the conversion goal set. Go into Google Ads conversion goals and remove soft events from campaign optimisation if they do not represent value. Newsletter sign-ups, page views and weak micro-conversions will teach automation to buy cheaper actions, not better customers.
- Set query and landing page review dates. Pull search terms and final URL performance at day 7, day 14 and at the end of the test. You are checking whether AI Max is finding adjacent intent or drifting into traffic your landing pages do not satisfy.
- Define the commercial pass mark before launch. Write down the CPA, ROAS, qualified lead rate, revenue per lead or margin threshold that makes the variant worth keeping. If the result misses that threshold, do not let a higher conversion count talk you into rolling it out.
If offline lead quality feeds Smart Bidding, fix imports before any major test starts. Our guide to campaign data import validation explains the checks that stop bad CRM data being treated as bidding truth. A clean experiment with broken conversion imports is still a broken experiment.
Use Google’s own setup guidance to avoid procedural mistakes. The official documentation on how Google Ads experiments work sets out how a test compares against an original campaign, which matters when you explain results to a board or finance team. For planning changes, validate the forecast against the Performance Planner forecasts help page before applying one-click recommendations.
Treat the September multi-campaign rollout as a trigger to clean your test backlog. The Search Engine Journal report on the new tools confirms the Search testing capability covers budgets and ROI targets across multiple campaigns. Build your first test around a decision you already need to make, such as whether to relax ROAS targets in one category, increase budget in non-brand Search, or test AI Max on campaigns currently constrained by exact and phrase match.
If target changes are already on your roadmap, read our practical framework for competitive bidding before deciding which metric should govern the test. A lead-gen account with poor sales feedback needs a different pass mark from an ecommerce account with clear product margin. If you need senior support designing the control, our Google Ads agency team can help structure the test so the result is commercially usable, not just statistically tidy.
What this means for your campaigns
Google is making Search automation easier to test, and that is a good thing only if advertisers raise their own standards at the same time. AI Max experiments give you better tools to answer the questions that matter: which budget caps deserve more money, which targets are too tight, which match-type expansion creates real uplift, and where automation is simply buying cheaper-looking noise.
The wrong response is to accept every forecast and roll out every winning variant. The right response is to test fewer things, define success more clearly, and force each result through brand, location, query, conversion quality and margin checks. That is how UK advertisers turn Google’s new tools into better decisions rather than faster waste.
If you’re unsure how exposed your campaigns are, a free Google Ads audit will surface the practical gaps quickly. For the detail behind this, see Google’s own announcement on new AI Max testing and planning tools.
Frequently asked questions
What are AI Max experiments in Google Ads?
AI Max experiments let advertisers test AI Max features in Search campaigns against a control before rolling changes out. They are useful for judging whether automation adds incremental conversions or simply shifts spend into broader traffic.
Why do brand and location controls matter in AI Max tests?
Brand and location controls keep the test close to the way the campaign will actually run. Without them, the variant can win by leaning on brand demand or traffic outside the advertiser’s trading area.
Should UK advertisers increase budgets after a successful experiment?
Only if the variant passes the commercial thresholds set before launch. Extra conversions do not justify extra spend unless qualified lead rate, ROAS, margin or revenue quality also hold up.
How should lead generation accounts test AI Max?
Lead generation accounts should use offline qualified lead data, not only form fills. A test that increases enquiries but lowers sales acceptance rate should be rejected.
What should ecommerce advertisers check before rolling out AI Max?
Ecommerce advertisers should split results by brand, non-brand, product margin, new customer value and query intent. Blended ROAS hides where the extra spend has gone.
