AI Max Doesn't Reset When You Switch It Off
Every AI Max conversation I get pulled into starts with the same question: on or off. That framing is why owners keep getting surprised by their own account a month later. Turning AI Max on buys you a test, and turning it back off does not buy you a refund, because the campaign keeps the buying habits it learned while the test ran. Both halves of that decision have a price, the entry and the exit, and you should know the second one before your card is already in the machine.
The exit is the part nobody quotes you a price for
Search "is AI Max worth it" and you get three flavors of the same page: what the feature is, what Google says it does, and a verdict on whether to enable it. All of that assumes the decision is reversible at zero cost. It is not. The setting flips back in a second; the account does not.
What actually happens is that Smart Bidding spends the test period learning a new definition of a good click. Broader queries, cheaper traffic, a different mix of people. When you switch the expansion off, that learned behavior does not vanish with the toggle. The campaign works through a shedding phase while it drifts back toward the tighter targeting you originally set, and the work of pulling it back is manual: negative keyword lists to rebuild, match types to tighten, bidding to retrain on clean data. Dii Pooler of PPC.live named this directly.
Read that as an invoice, not a technical footnote. The test period costs you whatever the traffic mix costs. The unwind costs you a second block of spend at degraded efficiency plus somebody's hours to do the cleanup, and if that somebody is your agency, you are paying for those hours too. Ten years of running Google Ads accounts across 300-plus businesses in medical, ecommerce, local services and fintech taught me one pattern I now plan around: owners almost never budget for the unwind, because nobody sells them the exit. I do not have a published number for how long the drift lasts, and neither does anyone else I can cite.
If you want the mechanism behind why the expansion pulls in that traffic in the first place, I argued that case separately in my breakdown of why AI Max behaves like broad match wearing a new badge. This piece assumes you already suspect that and want to know what the experiment will cost you.
Cheaper clicks are not cheaper customers
The dashboard makes the test look like an early win. Pooler documented exactly that in a single-account test.
“When AI Max was turned on, click volume nearly tripled and average CPC dropped 59%. On paper, that looks like a win. In reality, that was just vanity improvement because conversions fell 38% and CPL almost doubled to $850.”
Dii Pooler, PPC.live, April 14, 2026 (updated June 22, 2026)
Three times the clicks at 59% less per click is the kind of screenshot that gets forwarded to a business owner with the word "working" in the subject line. Then the leads report arrives and the picture inverts.
| What moved | How it looked in the dashboard | What it actually meant |
|---|---|---|
| Click volume | Nearly tripled | More traffic, not more demand. The extra clicks came from queries the account never chose to bid on. |
| Average CPC | Down 59% | Cheap clicks are only good news if they convert at the same rate. These did not. |
| Conversions | Down 38% | Roughly a third fewer real enquiries reaching the business, while activity charts pointed up. |
| Cost per lead | Almost doubled, to $850 | The only number attached to the bank account got worse by roughly 2x. |
Source: single-account AI Max test reported by Dii Pooler, PPC.live, April 14, 2026 (updated June 22, 2026).
Run that against your own economics for a second. If you close one lead in five and a closed job is worth $1,200 to you, an $850 lead means every sale costs you more than it brings in. You would be paying to lose money faster than before, with a prettier chart. That is one account, one test, and I am not presenting it as the average outcome. It is a documented example of how a metric can improve and a business can get poorer at the same time.
Only 22% of campaigns landed near their target, and you cannot pick which group you are in
One account proves a mechanism. A larger sample tells you the odds. Smarter Ecommerce analyzed more than 250 AI Max campaigns, and Brooke Osmundson reported the findings in Search Engine Journal in March 2026: median conversion value rose 13%, and median cost per acquisition rose 16%. More value coming in, each unit of it costing more to buy. For a business already at the edge of its margin, that trade is a loss with extra steps.
Consistency is where the study gets hard to argue with.
A four-in-five chance of missing your own target is not a feature rollout, it is a coin flip with worse odds than a coin. Plan the budget accordingly.
The same research found where the expansion enters your account. Across more than a million impressions, AI Max launched its expansion off exact-match keywords 80.11% of the time, phrase match 19.52%, and broad match 0.38%. Your most disciplined, most carefully chosen keywords are the springboard, not your loosest ones. Owners assume the tidy part of the account is protected because it is tidy. It is the opposite: exact match is precisely where the drift begins, which is why the cleanest accounts often feel the change fastest.
Google's own figure for the feature is around 14% more conversions at a similar cost per acquisition, rising to 27% for campaigns weighted toward exact and phrase keywords. That is Google's number, produced by Google, about a Google product. Hold it next to a median CPA that rose 16% in the independent study and decide for yourself which one to plan around. I would plan around the one that was not published by the company selling the inventory.
That spread is the whole argument for treating this as a funded experiment rather than an optimization. A tool with a 77-point range of outcomes is not something you switch on across an account that is currently paying your staff. It is something you test in a corner, with money you have decided in advance you can afford to lose.
Switching it off starts the second bill
Most people plan the test and skip the unwind, then discover the unwind by watching their cost per lead stay bad for weeks after they "fixed" it. The sequence below is ordered the way I would work through it on a client account.
Step four is where most of the invisible cost sits, and it is the same risk I have written about when you hand budget decisions to Google's automated bidding without a hard guardrail. Bidding models are trained on what recently happened in your account. Feed them a month of low-intent conversions and you have not just wasted the month, you have taught the system that low-intent traffic is what you want.
Run the test inside a box you can close
None of this means never touch it. It means test the way a business owner tests, with a capped downside and a written exit, not the way a dashboard invites you to test. The forced migration from Dynamic Search Ads into AI Max moved from September 2026 to February 2027, according to Common Thread Collective's coverage of Google's June 2026 update. That delay is your runway. You now have months to run a controlled test on your terms instead of waking up inside someone else's timeline, and if you have not confirmed how the change is set to land in your account, that is what the forced switch checklist covers.
If an agency runs your account, this is the whole conversation in one question: "Show me the baseline you saved, the budget cap, and the exact steps you will follow to unwind it." A good operator answers without hesitating. A vague answer tells you they were planning to test with your money and improvise the recovery, which is the expensive version of every one of these stories.
Watch the number that pays you, not the number that flatters you
Cheaper clicks and rising volume are the two metrics that will move first and mean least. In the PPC.live test, both looked excellent while conversions dropped 38% and the cost of a lead nearly doubled to $850. If your weekly report leads with traffic, impressions or average CPC, you are reading a report designed to make the platform look good rather than one designed to make you money.
Track the weekly three below on a fixed rhythm. The third item is the one owners skip, and it is the fastest way to see what the expansion decided you sell. Owners who do this see the drift while the budget is still recoverable. Owners who wait for the monthly report see it after the money is gone.
Lead quality is the one that never shows up in the interface at all. Your front desk knows when the enquiries get worse, and that signal arrives before any of the platform metrics confirm it. When I booked a London ADHD clinic solid for three straight months, volume was never the question the business asked me about. Fit was. Volume that the business cannot serve profitably is not growth, it is a staffing problem you paid Google to create.
“Treat AI Max as a controlled expansion layer, not a replacement for the foundation of your Search campaigns.”
Brooke Osmundson, Search Engine Journal, March 2026
Controlled is the operative word, and it means what it means in any other part of your business: a defined budget, a defined window, a defined way out. Everything in the data says the tool can add volume. Almost nothing in the independent data says it reliably adds efficiency, and the 22% figure says you cannot know in advance which side you land on.
Frequently Asked Questions
Is AI Max worth turning on if I only spend a small amount each month?
The smaller the budget, the worse the risk-to-reward trade, because a small account cannot absorb a bad month and a slow recovery at the same time. In the SMEC study of 250-plus campaigns reported by Search Engine Journal in March 2026, ROAS outcomes ranged from 42% better to 35% worse, and only 22% of campaigns landed near their original target. If losing your test budget entirely would change a decision you need to make this quarter, do not run the test yet. Fix conversion tracking, search term hygiene and landing pages first, since those improvements do not require an unwind.
What actually happens when I turn AI Max off?
The setting turns off, but the campaign does not return to its old behavior on its own. Dii Pooler of PPC.live describes a shedding phase where the campaign works through what it learned, and the account may need rebuilt negative keyword lists, tightened match types, and time for Smart Bidding to relearn what a real conversion looks like. Plan on doing that cleanup by hand rather than waiting for it to self-correct. Keep watching cost per lead weekly until it returns to whatever it was before you started.
How long should I run an AI Max test before I decide?
Set the window before you flip the switch, and make it long enough to collect a meaningful number of leads for your business rather than a fixed number of days that ignores your volume. A clinic getting forty enquiries a month can judge faster than a specialist firm getting six. Judge on cost per lead and lead quality, and stop early if cost per lead moves clearly in the wrong direction, since there is no prize for completing a losing experiment. The recovery window is the one number nobody has published, so treat the cleanup as an open-ended cost rather than a fixed one.
AI Max in 2026 is a volume lever with an unpredictable efficiency outcome and a cleanup bill that arrives after you think you have stopped. That does not make it useless. It makes it an experiment, and experiments get budgets, end dates and exit plans, the same as anything else you would spend real money on.
If you want a second pair of eyes on your account before you run the test, or if you already ran it and the numbers have not come back yet, book a call and we can look at the search terms together. Bring your last 90 days. The setting you can switch off in a single click is not the same thing as a decision you can take back.
