Every account I transitioned to Google AI Max this year went through the same difficult period. CPAs spiked. Google Performance Max campaigns that had been running well started degrading at the same time. Stakeholders wanted to know what had gone wrong. The honest answer was: the platform behaved exactly as it should. The problem was three specific mistakes I made in sequence, and not fully understanding the mechanics before I started.

I am writing this because I have not seen a clear explanation of those mechanics anywhere. Google's documentation describes what Google AI Max does. It does not describe what it does to the rest of your account when you turn it on, or why the impact looks so different depending on the size and category of the account.

What Google AI Max actually is

Google AI Max is Google Performance Max technology applied on search campaigns. That single sentence contains two things worth understanding separately.

First: it is Google Performance Max technology. That means it is not a smarter version of your existing search campaigns. It is a different campaign type with a different optimisation engine, and it competes for the same traffic your existing campaigns are already bidding on.

Second, and this is the part most operators miss: Google AI Max uses keywords as themes, not as exact triggers. When you give Google AI Max a keyword, you are not telling it to match that specific search term. You are giving it a signal about intent and topic. It uses that signal to find users within that intent space, which is a much wider mandate than traditional keyword matching. 

This is why the match type you use matters so much. Give it broad or phrase match keywords and the theme it infers is too wide. It pulls in traffic from across a loosely related space, most of which will not convert. Give it exact match keywords and the theme stays narrow and focused. The algorithm still has latitude to find users within that intent, but the guardrails keep it on topic.

I set up Google AI Max on exact match keywords first. Not because exact match is stricter in the traditional sense, but because it gives Google AI Max the tightest possible theme to work from. Once the account has sufficient data and stable performance, you can evaluate whether broader match types make sense. But exact match is where I start every time.

What happens to your Google Performance Max campaigns

Google Performance Max campaigns have historically drawn the majority of their impressions from search. In most accounts I manage, that figure sits between eighty and ninety percent. Google Performance Max is, in practice, largely a search campaign under a different label.

When you switch on Google AI Max, the search component of your Google Performance Max traffic begins migrating toward AI Max. The algorithm identifies that Google AI Max is now the designated search tool and starts routing search intent traffic accordingly. Google Performance Max loses its primary traffic source and enters a learning phase to find other placements and signals. Google AI Max is simultaneously in its own learning phase as it figures out your account, your audience, and your conversion signals.

Both campaigns are learning at the same time, drawing from the same budget. This is the source of the disruption. And it is why what happens next depends so heavily on the volume of the account.

Why account size changes everything

This is the nuance that does not get discussed enough.

For a large account with significant daily conversion volume, the learning period is short and largely self-correcting. The algorithm has enough data to learn quickly, and both campaigns stabilise within a week or two. In these accounts the disruption is manageable, and the transition is worth doing without heavy intervention. Ring fencing and elaborate negative keyword structures may actually limit performance here, because you are constraining an algorithm that would learn its way to the right answer faster than you can manually guide it. 

For a small or mid-market account where conversions are measured in tens or low hundreds per month, the same transition looks very different. Both campaigns learning simultaneously against limited conversion data creates a compounding problem. CPAs stay elevated for longer. For a business that cannot absorb several weeks of degraded performance, this is a real commercial risk, not a temporary inconvenience.

The question I ask before every Google AI Max transition is: can this business absorb two to four weeks of elevated CPAs and reduced conversion volume while the algorithm learns? For large accounts, yes. For small accounts, the answer often determines the timing more than anything else.

A separate consideration for regulated categories

If your business operates in a regulated category where text optimisations and landing page optimisations cannot be switched on, the dynamics change again. Google AI Max has less latitude to self-optimise when those features are restricted. The self-correction that happens naturally in unrestricted accounts takes longer or does not happen to the same degree.

In these categories, being deliberate about ring fencing and negative keyword management is not optional caution. It is how you compensate for the reduced optimisation surface the platform has to work with.

The negative keyword question I do not have a definitive answer to yet

Once a Google AI Max account has stable activity and the learning phase is complete, I believe the algorithm will eventually not need negative keywords at all. Google AI Max should, over time, learn which searches belong to which campaign intent and route accordingly without manual negatives enforcing the boundary.

I do not have enough accounts that have run long enough at stable volume to confirm this with confidence. What I have observed is that negatives are essential during the early learning phase to prevent cross-pollination between campaigns with similar themes. Whether they remain necessary indefinitely is a question I am still watching.

What I do know is that once an account has settled and is performing consistently, it is worth systematically testing the removal of negatives to see whether Google AI Max has learned to self-regulate. Remove them one category at a time with close monitoring, not all at once. If performance holds, the negatives were no longer doing necessary work. If performance deteriorates, you have your answer and you put them back.

I would rather run that test deliberately than leave negatives in place indefinitely and potentially limit what Google AI Max can do in an account that has genuinely matured. Over-constraining a capable algorithm is its own kind of mistake.

The three mistakes I made, and what I do differently now

I am naming these specifically because vague advice about sequencing does not help anyone. Here is exactly what went wrong in order.

Mistake one: I changed the conversion action the campaign was optimising toward.

I switched the conversion action without giving the campaign time to resettle under the new signal. A campaign that changes what it is optimising toward needs time to relearn. The rule of thumb is thirty conversions under the new signal before the account is stable enough to trust. I did not wait for that. I moved to the next step while the campaign was still in learning from the first change.

What I do differently now: any change to a conversion action is a standalone phase. I make the change, monitor until the campaign has accumulated thirty conversions against the new signal, confirm performance is stable, and only then consider what comes next. No stacking changes on top of an account that is still learning.

Mistake two: I consolidated the campaign structure without waiting for it to resettle before switching on Google AI Max.

Consolidation is its own disruption. Merging campaigns and ad groups means the algorithm is relearning which traffic belongs where and how to distribute budget. That relearning takes time and conversion volume. I moved to Google AI Max while the consolidated structure was still settling, which meant I had two sources of instability running simultaneously on the same account.

What I do differently now: consolidation and Google AI Max are two separate phases with a stabilisation period between them. I consolidate, wait until the account has reached stable performance and accumulated the required conversion volume, and only then introduce Google AI Max. Never both at once.

Mistake three: I switched on Google AI Max without thinking through the negatives question first.

I did not ask whether the account needed negative keywords to prevent cross-pollination between campaigns before Google AI Max went live. Because Google AI Max treats keywords as themes rather than exact triggers, campaigns with overlapping themes will bleed into each other if you do not actively prevent it during the learning phase. I found this out after the fact, which meant fixing it reactively while the account was already in disruption.

What I do differently now: before Google AI Max goes live, I map every campaign's theme and identify where overlap exists. For smaller accounts and regulated categories, I build the negatives before the first Google AI Max impression is served. For large volume accounts, I monitor closely in the first two weeks and add negatives reactively only if cross-pollination is visible in the search term report.

The timing is yours to decide

Google's account managers will encourage you to switch on Google AI Max. They are not wrong that it represents where search is heading. They are also not the ones managing the commercial consequences if the transition goes badly.

I pushed back on a Google AI Max transition for one account this year after watching what had happened elsewhere. The business was in its best performance period in years and could not justify the disruption. That decision, and the conversation behind it, is its own story. I have written separately about why I move slower with smaller accounts.

Google AI Max will be the default eventually. The question is not whether to transition. It is when, in what order, and with what preparation.

The transition to Google AI Max is one part of a larger platform shift that started showing up in December 2025. If you noticed your portfolio bid strategies quietly missing targets around the same time, that story is connected.