Google has rolled AI Max for Search across Australia and New Zealand and from September 2026 it auto-upgrades remaining legacy campaigns by default. The 27% conversion lift on the pitch is real but conditional and self-measured, and auto-enrolment hands most advertisers to the system without the holdout that would tell them whether it worked. The one clean Australian win so far came from Telstra running a proper A/B test with its agency.
When the platform spends your money and reports its own results, its number and your truth are not the same thing.
The Take: Google has switched AI Max on across Australia and New Zealand, and from September 2026 it starts auto-upgrading the legacy search campaigns it has not already converted. The 27% conversion lift on the billboard is real for a thin slice of accounts and close to invisible for the rest, because it is a figure Google measured on its own system with its own definition of a win. Take the upgrade if the maths works. Do not take the result on faith.
Does a 27% conversion lift settle the AI Max question?
No. Google's own documentation puts the typical activation uplift at 14% more conversions, not 27%. The 27% is the top of the range, reserved for a narrow set of accounts.
The number has done its job before you finish reading it. It anchors the decision. It makes the upgrade feel like found money. Google rolled AI Max out at Google Marketing Live in Sydney and put an average of 27% more conversions against manually managed search at the front of the pitch. Then read the fine print.
What Google actually measured
In Google's own announcement the 27% is the higher end of the range, specific to non-retail campaigns where more than 70% of conversions already came from exact and phrase match keywords. The figure it quotes for simply activating AI Max is 14% more conversions or conversion value at a similar cost. Two different numbers, one press release and the bigger one on the poster.
Independent PPC analysis of accounts running AI Max in 2026 reports the majority see neutral or negative results, with the strong outcomes concentrated in accounts that already meet the prerequisites.
Why auto-enrolment is the real story
This should worry you more than the headline. From September 2026 Google auto-upgrades the remaining legacy campaigns to AI Max by default. Most Australian advertisers will be moved onto the system without ever running the one test that tells them whether it helped.
The Australian ad market grew 5.2% to $28.9 billion last year. We have scored hundreds of Australian businesses at New Rebellion and most of them cannot tell you which half of a search budget is actually producing sales. That read comes from how we score Australian businesses, a scan and benchmark method of our own rather than a platform dashboard, run on a rolling sample so it is directional not a fixed count. Hand that same operator an automatic upgrade and a self-reported 27%, and the upgrade books itself.
Forecast Australian search ad spend for 2026, up 9.5% on the year (WPP Media)
That is the pool being quietly moved onto a new set of rules. Around 12% of Australian businesses reported using AI at all in 2024-25, so most of the people about to run an AI bidding and matching system have never run one before. They will not know what a normal week looks like, which means they will not notice when it changes.
How did Telstra get a clean win?
By doing the work. Telstra and its agency OMD ran a controlled A/B test of AI Max in the prepaid SIM market and recorded a 74% lift in click-through rate, a 52% lift in SIM orders and a 36% cut in cost per order. The result is believable precisely because it came from a holdout, not from the platform's read on itself.
A holdout is just a slice of your audience the new system never touches, kept back so you have something honest to compare against. Read that back. The only Australian AI Max number worth trusting so far is the one a brand measured against itself.
Google measured the 27% on itself. Telstra measured the 52% against a group AI Max never touched. Only one of those has a control.
What I would do about it
Do not refuse the upgrade. Turning down automation on principle is how you fall behind. Put a scoreboard around it before September instead.
None of that is exotic. It is the difference between running an experiment and being one.
The upgrade is coming either way
My view is simple. AI Max will help some accounts and quietly bleed others, and the auto-upgrade guarantees most advertisers find out which one they are the expensive way. The platform gets to spend your money and report the score. Your job is to keep your own. That is the work we do inside NR Studio: building the holdout, the baseline and the scoreboard that tell you whether the 27% ever showed up in your bank account.
Frequently asked questions
What is Google AI Max for Search?
It is Google's package of AI-driven keyword matching, creative and targeting for search campaigns. From September 2026 Google auto-upgrades remaining legacy campaigns to it by default across Australia and New Zealand.
Is the 27% conversion lift real?
It is real as a measured average on a specific set of Google's own accounts. Google's typical quoted uplift is 14%, and independent testers find most accounts see neutral or negative results.
Should I turn off AI Max?
Not on principle. Run it with a holdout or geo-split and your own conversion definition. If it wins against a proper control, keep it. If it does not, you will have the evidence to roll back.
How do I test AI Max properly?
Hold back a region or campaign the upgrade never touches, record your baseline cost per order and conversion rate first and compare the two over at least four weeks.