AI Max is Google's AI layer for Search campaigns — it broadens matching past your keyword list and, through text customization, writes its own headlines and descriptions from your site, ads, and landing pages. That means the ad copy running in your account is increasingly machine-generated, and the sensible reaction is not to trust it blindly or refuse it outright but to test it. On August 20, 2026 Google shipped the tools to do exactly that: AI Max experiments now run as an A/B test inside a single existing campaign — diverting traffic between an AI-Max-off control and an AI-Max-on treatment rather than cloning the campaign — and they can now hold brand and location controls active while they run. Alongside them, a September rollout adds multi-campaign A/B testing for budget and ROI-target changes, plus a Performance Planner that forecasts those changes and applies them in one click. This guide explains what an AI Max experiment actually measures, what's new in the August update, how to run one cleanly, and the honest limit of the whole exercise: AI Max only tests the paid-search text inside Google's walls. The creative that decides whether the click converts — the video, the organic proof, the on-brand landing experience — is produced outside it, and that is the half this page connects back to.
AI Max is Google's AI optimization layer for Search campaigns — a setting you switch on inside an existing campaign rather than a new campaign type. When it is on, it does two things that change what your account is doing. Its search term matching goes past your keyword list, using broad-match and keywordless technology to show your ads for queries it judges relevant by intent. And, if you enable text customization, it generates its own headlines and descriptions on the fly, built from your domain, landing pages, existing ads, and the ad group's keywords, tailored to the context of each search. Final URL expansion can also route a click to whichever page on your site best fits the query. All of it leans on conversion-based Smart Bidding to work.
That combination is powerful and also unnerving, because it hands the machine control over two things advertisers have historically owned: which searches you pay for, and what your ad says. An AI Max experiment exists to answer the obvious question that raises — does turning this on actually help my campaign — with data instead of faith. It is an A/B test, but a specific kind. Rather than copying your campaign and running the copy alongside the original, it keeps everything inside one campaign: a share of the campaign's traffic runs with the AI Max toggle off as the control, and the remaining share runs with it on as the treatment. You let it run, then compare the two arms on conversions, cost per acquisition or ROAS, and — the metrics that matter most with broadened matching — the quality of the search terms it brought in and the quality of the resulting leads.
The single-campaign design is the important detail. Because a traditional experiment clones the campaign, it splits budget and bidding history across two campaigns, restarts the learning period, and opens the door to subtle setup drift between the original and the copy. Keeping the test inside one campaign avoids all of that: budget and history stay consolidated, learning does not reset from scratch, and the only deliberate difference between the two arms is the AI Max toggle. Google's pitch for the format is faster, cleaner results with fewer of the errors that make experiments hard to trust. For advertisers, the practical translation is that testing AI Max is now cheap enough and clean enough that there is no good reason to flip it on blind.
On August 20, 2026, Google announced a set of new testing and planning tools built around this workflow. Three pieces are worth separating clearly, because they solve different problems and one of them is not strictly about AI Max at all.
The most consequential change for AI Max specifically is that experiments can now run with brand and location controls enabled. Previously, testing AI Max often meant choosing between running the experiment and keeping the restrictions your business depends on — brand-term controls that stop your ads showing on competitors' or unrelated brand queries, and location settings that keep spend inside the geographies you serve. Now you can hold those controls active while the experiment runs, which means the test reflects the settings you would actually use long term, and far more advertisers who could not previously loosen their guardrails can test at all. A test that only produces a valid answer under settings you would never ship in production is not much of a test; this change closes that gap.
Starting in September 2026, Google is adding the ability to test budget and ROI-target changes across multiple Search campaigns in a single A/B experiment. This is the tool for the question 'what happens to my bottom line if I scale up' — instead of nudging budgets or targets campaign by campaign and trying to read the noise, you test a change across a group of campaigns at once and measure the aggregate effect. It shipped alongside the AI Max news but is a distinct capability: it is about validating spend and target changes at portfolio scale, not about AI Max's matching or creative. Treat them as two tools that happen to have launched together, not one feature.
The third piece is an enhanced Performance Planner that forecasts how changes such as bidding or budget targets would affect your existing campaigns, and then lets you apply Google's suggested changes with one click. Where an experiment measures a change after the fact, the planner estimates it beforehand. Used together, the pattern Google is steering toward is: forecast a change in the planner, validate the ones that look promising in an experiment, then roll out. The one-click application is a convenience with a caveat worth stating — a forecast is a model's estimate, not a measured result, so the discipline of confirming the meaningful changes in an experiment before trusting them is exactly what the rest of this guide is about.
It is easy to read AI Max experiments as just another campaign setting to A/B test. The framing that makes them matter is narrower: with text customization on, AI Max is generating the ad copy itself. The headlines and descriptions running against a given search were not written by you — they were assembled by a model from your site and assets, per query. That is AI-generated creative going live in a paid channel, at the scale of every impression. The experiment is how you find out whether the machine's copy and the machine's expanded matching beat what you had, for your account, on the metrics you actually care about.
This is the same discipline that already applies to AI-generated creative everywhere else — you generate at scale, then you test what works rather than shipping the model's first output on trust. The principle is covered for the organic side in the guide on A/B testing social creatives, and the broader shift of ad-creative generation moving inside the ad platforms themselves is the subject of AI ad generation inside ad platforms. What Google's August tools do is bring that test-before-you-trust loop into Search, with a clean single-campaign mechanism and guardrails that stay on. The reason to care is not that a new experiment type exists; it is that the copy in your account is now partly written by AI, and Google finally gave you an honest way to check whether that is helping.
Being precise about the boundary matters, because it is exactly the boundary this guide's final section is about. In a Search campaign, AI Max generates text: headlines and descriptions, tuned to each query, from material you already have. It does not generate video. It does not generate images or scenes. It does not build a landing page or the organic content that warmed the searcher up before they ever typed the query. Its creative surface is the paid-search ad unit, and within that unit, the words.
That is not a criticism — text customization for Search is a genuinely useful capability, and the experiment framework around it is a real improvement. But it defines a hard edge. Everything that happens before and after the paid click — the short-form video that built awareness, the carousel that established the proof, the persona and voice that make a brand recognizable across a feed, the landing experience the click lands on — sits entirely outside AI Max. AI Max optimizes the moment of the auction. It has nothing to say about the creative that makes that auction worth entering, or the content that converts the click once it arrives. An experiment can tell you the machine's headline beat yours; it cannot tell you the click had nowhere good to land.
The mechanics are simple; the discipline is where results are won or lost. A few rules keep an AI Max experiment honest.
The value of the single-campaign design is that the only intended difference between control and treatment is the AI Max toggle. Preserve that. Set your brand and location controls to what you would actually run in production — now that experiments support keeping them on, there is no excuse to test under settings you would never ship — and do not change bids, budgets, or assets mid-flight. If you also want to test a budget change, that is what the September multi-campaign tool is for; do not fold it into the AI Max test, or you will not know which lever moved the number.
Broadened matching and generated copy need time to accumulate enough conversions to compare, and Smart Bidding needs traffic to learn against. Ending an experiment early, on a few days of thin data, is how advertisers convince themselves of a result the numbers do not actually support. Let it run across a representative window — long enough to cover your normal buying cycle and clear the initial learning noise — before you read it as real.
AI Max's biggest risk is not usually cost per click; it is relevance drift. Broad, keywordless matching can bring in volume that converts on paper but is the wrong audience — cheaper conversions of lower value, or leads that never close. So look past conversions and CPA to the actual search terms the treatment matched and, where you can measure it, the downstream quality of the leads. A treatment that raises conversions while lowering their quality is a loss disguised as a win, and only the query- and lead-level view catches it.
When the experiment gives a clear answer, apply it — or don't — and keep watching. AI Max keeps generating and matching after you commit, so a good experiment result is a license to proceed, not a finished decision. The same is true of the Performance Planner forecasts feeding this loop: confirm the meaningful ones with a real test rather than one-click-applying on the estimate alone.
Put the two previous sections together and the shape of the problem is clear. AI Max, tested well, tells you the best possible version of your paid-search text and matching. It says nothing about whether anyone was primed to search for you in the first place, or whether the click had a reason to convert. Those are creative and content questions, and they live entirely outside Google Ads: the awareness content that seeds branded demand, the video and social proof that make a searcher choose you over the other three AI-generated headlines in the auction, the on-brand consistency between the ad and the page it opens.
This is where most of the leverage actually sits, and it is the part AI Max's experimentation tools structurally cannot reach. You can perfect the paid unit and still lose, because the paid unit is the narrowest slice of the journey. The creative that decides the outcome — produced at the scale and consistency that a modern funnel demands — is a separate production problem, and the one worth solving alongside a well-run AI Max test rather than instead of it. For the paid-social side of the same creative-generation shift, see AI ad creative generation across social platforms; for the disclosure obligations that now attach to AI-made ad creative on Google specifically, see the guide on Google Ads AI content disclosure.
Kompozy does not run your Google Ads, and it does not try to — AI Max experiments are the right tool for tuning paid-search matching and copy, and this guide's advice is to use them well. Kompozy owns the other half: producing the on-brand video, image, and text creative that feeds the funnel AI Max only optimizes the auction inside. It is an AI content generation and multi-platform publishing engine, so where AI Max generates headlines for one ad unit, Kompozy generates the surrounding content library — the awareness and proof layer that decides whether a search ever happens and whether the click converts.
Concretely: from one source idea, Kompozy produces captioned Persona Shorts and longer avatar video that build branded demand, brand-exact Carousel Posts and other output buckets rendered through HyperFrames, quote graphics, photo posts, plus blog articles and email newsletters — all governed by a written Persona Brief so the voice and face stay consistent from the top-of-funnel video down to the landing page the paid click opens. That consistency is precisely what the isolated AI Max test cannot give you: a brand a searcher already recognizes, and a click that arrives somewhere coherent. Autopilot then schedules and fans that library across the eight social platforms plus blog and email, with a per-post human review gate before anything publishes — the same test-before-you-ship discipline this guide urges for paid copy, applied to the organic creative Google never sees.
The clean division of labor is the point. Let AI Max experiment its way to the best paid-search text inside Google's walls; that is a real and useful job, and the August 2026 tools make it easier to do honestly. Use Kompozy to produce and distribute the creative that makes those auctions worth winning — the video, the social proof, the consistent brand identity — at the scale and cadence a funnel needs. One optimizes the moment of the click. The other builds everything on both sides of it. Testing the first without producing the second is optimizing the narrowest part of the journey and leaving the widest part to chance.
Google Ads AI Max experimentation is the test-before-you-trust mechanism for AI Max — an A/B test run inside a single Search campaign that diverts traffic between an AI-Max-off control and an AI-Max-on treatment, so you can measure the real effect of broadened matching and machine-generated ad copy before committing account-wide. The August 20, 2026 update made it materially more usable: experiments can keep brand and location controls active, a September rollout adds multi-campaign budget and ROI testing, and Performance Planner now forecasts changes with one-click apply. Used with discipline — one variable, a full cycle, and a hard look at search-term and lead quality — it is the right way to adopt AI Max. Its honest limit is its scope: AI Max only tests the paid-search text inside the auction. The creative that seeds the demand and converts the click — video, social proof, a consistent brand across every surface — is produced outside it, and that is the half worth building in parallel, not the half to leave to chance.
An AI Max experiment is an A/B test that lets you compare a Search campaign with AI Max turned off against the same campaign with AI Max turned on, before you commit to it account-wide. Unlike a traditional experiment, it does not clone the campaign — it diverts a share of the existing campaign's traffic and budget to an AI-Max-on treatment and keeps the rest as the control. You then read conversions, CPA or ROAS, and the quality of the search terms and leads AI Max brings in, and decide whether to apply it.
Three things. AI Max experiments can now run with brand and location controls kept active, so you no longer have to strip your guardrails to test. Starting in September, multi-campaign A/B testing lets you test budget and ROI-target changes across several Search campaigns in one experiment. And Performance Planner gained the ability to forecast how bidding or budget changes would affect existing campaigns, with one-click application of Google's suggested changes. The experiment and planning tools shipped together to make validating AI Max before rollout easier.
For Search, yes — but text creative, not video or images. AI Max's text customization setting generates headlines and descriptions on the fly from your domain, landing pages, existing ads, and the keywords in the ad group, tailoring the copy to each query's context. Its search term matching also broadens which queries you show for using broad-match and keywordless technology. So the running ad copy becomes partly machine-written, which is exactly why an experiment matters: you are testing generated creative, not just a setting.
A standard campaign experiment creates a copy of your campaign and splits traffic between the original and the copy, which restarts learning and can introduce setup drift between the two. An AI Max experiment keeps everything inside one campaign and toggles AI Max on for part of its traffic. Google's stated advantages are faster results, fewer experimentation errors, and a shorter learning period, because budget and history stay consolidated in a single campaign rather than being split across a duplicate.
You can, but the whole point of the new experiment type is that you no longer have to guess. AI Max changes both which queries you match and, if text customization is on, what your ads say — two levers that can lift or dilute performance depending on your account. Running an experiment for a full conversion cycle, with brand and location controls set the way you would actually use them, tells you whether it helps your specific campaign before you make it permanent. Test, read the search-term and lead quality, then decide.
A Google Ads AI Max experiment is an A/B test that compares a Search campaign with AI Max off against the same campaign with AI Max on, without cloning it — it diverts part of the existing campaign's traffic to an AI-Max-on treatment and keeps the rest as control. Announced with new tools on August 20, 2026, it can now run with brand and location controls active, and lets you validate AI Max's broadened matching and AI-generated ad copy before rolling it out account-wide.
Get started → · ← All guides · Compare Kompozy vs other tools