PMax and AI Max: How Much to Hand to Automation

The direction ad platforms have taken in recent years can be summed up in a single sentence: reduce the number of decisions a human has to make by hand. Settings that were once configured one at a time — how much to bid on which keyword, which placement to appear in, what hour of the day the budget opens — can today be tied to one goal and left to the system. On Google's side the broadest form of this is Performance Max, and the newer automation layer added to search campaigns goes by the name AI Max.

In Turkish-language sources the promotional side of these products is more than covered. The same definition, the same feature list and the same "AI optimises your budget for you" sentence are repeated across dozens of pages. What is missing is this: how far is handing over control defensible, what stops being visible in the report once you hand it over, and which accounts are not mature enough to take that step.

The text below is not a product tour. It treats automation not as a list of capabilities but as a trade in which you give something up in return: you give control and visibility, you get scale and speed. Whether it is a good trade depends on the data in your account.

Automation is not a strategy, it is a bet on data quality

Automated bidding and automated targeting rest on the same foundation: extracting a pattern from past conversion data. The system tries to learn which user signal ended in a conversion, and bids higher on users carrying a similar signal. That mechanism has one decisive input — the conversion signal you send it.

Two consequences follow. First, if the signal is wrong, automation accelerates the error. A badly defined conversion — a setup that counts every page view as a conversion, say — tells the algorithm "plenty of traffic is good", and the algorithm delivers exactly that. Second, if the signal is sparse there is no pattern to learn. A pattern extracted from a handful of data points is not a pattern, it is noise.

That is why "shall we move to automation" is less a strategy question than a readiness question. The right order is: verify measurement first, then look at volume, then hand over. In accounts that go in the reverse order, when results fail to arrive the blame usually lands on the product; the product, meanwhile, has followed the signal it was given faithfully.

What Performance Max takes over, and what you get in return

Performance Max is a campaign type that runs across most of Google's inventory from inside a single campaign: search, display, video, shopping and other surfaces all feed from the same budget. You supply the goal, the assets (text, image, video) and a product feed if you have one; the system builds the distribution.

What it takes over is clear: how budget is spread across surfaces, placement selection, audience expansion and bidding. What it gives back is scale and setup speed — one campaign puts you on many surfaces, and you are spared having to build a separate campaign for each.

What you pay in return is a portion of your visibility. In a classic search campaign you can read separately which query triggered which ad, how much each placement spent and what each audience brought in. In automated campaign types some of that detail arrives aggregated. Reporting scope has been widened over time and continues to widen; but rather than assuming what is visible in your own account today, you should open it and look. That is the most practical advice in this section: check what the campaign's report will show you before you launch it, not afterwards.

What AI Max adds to a search campaign

AI Max is presented as a set of features that can be switched on for search campaigns: extending matching beyond the keyword list, generating text automatically from the landing page and from existing assets, and routing a query to the most suitable page are offered together under one heading. In other words it is not a separate campaign type like Performance Max but a layer sitting on top of an existing search campaign.

The practical difference follows from that. Performance Max takes over the whole campaign; an automation layer added to a search campaign stays inside search intent and widens the matching and creative side. When the two are presented as alternatives to each other the decision gets harder, when the real question is not "which is better" but "which control am I ready to let go of".

Honesty is required here: the names of this family of features, their scope and which accounts have them switched on all change quickly. While this article was being prepared it was hard to find a Turkish-language page devoted to AI Max; on the English side, announcement copy and independent assessments were not yet at the same level of maturity. The reliable way to learn what a feature does in your account is to read your own dashboard's documentation and to test it with a small budget.

The preconditions for handing over: a clean signal and enough volume

There are two gates before you move to automation, and both have to be open at the same time.

The first gate is the accuracy of the signal. The conversion counted in the dashboard has to correspond to an action that genuinely maps to a business outcome, land as a single record, and not be counted twice in two different systems. How to verify that, and why this is the first breaking point in most accounts, we covered in detail in our article on ad measurement, consent and modelling.

The second gate is volume. Conversion-based strategies learn when there is a steady flow of conversions. The platform's own documentation carries minimum conversion count recommendations for some strategies; because those thresholds are updated from time to time, reading the figure from your dashboard's current help documentation is more reliable than memorising it. What should be memorised is not the number but the principle: if your account struggles to gather more than a few conversions a week, strategies that target a cost per acquisition will produce volatile results.

If either gate is shut, the job is not to postpone automation but to open the gate. Fixing measurement takes days; accumulating volume takes weeks. Both are cheaper work than fiddling with campaign settings.

What it looks like in the dashboard when the data is not enough

Automated bidding working with insufficient data bases its decisions on very few examples, and swings as a result. The picture is usually this: cost falls one week, rises noticeably the next, and by the third week impressions have almost stopped. The business tries to explain the swing with a story about seasonality or competition; most of the time the system is simply trying a different hypothesis every week with the little data it has.

The second typical outcome is targeting drifting away from intent. Automatic expansion looks for signals that resemble conversions. If the signal set is small the definition of "resemblance" loosens, and part of the spend can drift towards subjects, regions or user groups you do not serve.

The third is that diagnosis gets harder. In a manually built campaign the reason for a bad result is usually visible in a setting. In an automated campaign the same result is hidden not inside the settings but inside the signal — and it is invisible to a team that has not made a habit of looking at the signal. For a reader who wants to strip the problem back step by step, we set out the diagnostic order in our article on ad spend going out with no conversions.

The levers that keep control in your hands

Moving to automation does not have to mean letting go of everything. Even after handing over you keep several levers, and the real work is using them properly.

Conversion definition and value. You set the goal the system chases. Declaring more than one action a primary conversion means handing it a goal with no clear direction. Assigning different values to different actions is the most direct way to raise the quality of automation.

Separating brand from non-brand. Searches on your brand name convert anyway. When the two are collected in one campaign, the strong results from the brand side cover up the weak results on the service side, and automation is pulled — misleadingly — in that direction. Separating them cleans up both the report and the decision.

Negative keywords and exclusions. Negative keyword support in automated campaign types has widened over time; check from your own dashboard at which level you are able to define them. Brand exclusions, subjects you do not serve and regions you do not work in are the first items on that list.

Asset and feed quality. Automation produces combinations from the text and the images you supply. Handing over weak assets and expecting good combinations is like asking a kitchen with missing ingredients for a rich menu. In accounts that sell products, the same role falls to the product feed.

The landing page. The one area automation cannot control is what happens after the click. A promise the page fails to meet is not made good by any bidding algorithm.

Discipline during the learning period

When a significant setting is changed the campaign starts learning again, and results produced during that phase should not be treated as settled. That is not new information, but there is a side of it that goes unapplied: batching the changes.

A workable routine looks like this. Write change proposals into a list over the course of the week, adding the reason and the effect you expect next to each. Once a week, apply them in one go. After applying, do not touch anything until the next review. That way you avoid restarting the learning phase continuously, and you leave yourself a window of time in which the effect of the changes can actually be read.

The most expensive habit is cramming several changes into the same day. When bidding, budget and assets are revised together, even if the result improves the source of the improvement stays unclear; an improvement whose source is unclear cannot be produced a second time.

Winning back the areas lost in reporting

In accounts that move to automation, part of the report becomes aggregated. Not all of it can be won back, but a significant portion can be compensated for through account structure.

The whole of the right-hand column rests on one logic: what cannot be separated cannot be measured. Whatever automation aggregates, you can partly re-separate by splitting up your campaign structure. That has a cost — a divided budget means less data in each piece. So the decision to split also depends on volume, and does not give the same answer in every account.

Automation on a small budget: what order to follow

The most common mistake in accounts with a tight budget is starting the job with the broadest automation available. The reasoning looks sound: there is no time to manage this by hand, let the system handle it. The result is usually this — a small number of conversions is spread across many surfaces and the learning threshold is reached on none of them.

A more defensible order can be built like this. Start with a narrow, high-intent search campaign; the aim here is not revenue but accumulating clean conversion data. Once measurement is verified to be working correctly and a few weeks of steady flow have formed, the bidding side is moved to automatic. If the account passes that stage steadily, only then are campaign types that widen inventory tried — preferably with a separate budget, without closing the existing campaign.

The advantage of that order is that each step uses the data produced by the step before it. The disadvantage is that it is slower. The fast route is always available; it is just a speed that cannot be measured. When deciding which set of metrics to follow, the five items in our article on customer data metrics for small businesses offer a short framework.

A decision table: which account is ready for which level

The table is an order, not a prescription. As you move down it the manual work does not decrease — it moves. In an automated account most of the time goes not into campaign settings but into the conversion definition, the assets and data quality. That shift also changes how scope is defined in an agency relationship; where the models diverge from one another and which clauses to look for in a contract are unpacked in our article on fee models and account ownership.

What we do not know

The way to be honest in this area is to write down what is unknown.

The scope and the names of automation products change quickly; the product descriptions above set out the general framework, not today's feature list. Second, what is shown in reporting can differ from account to account and over time — what appears in your dashboard may not be the same as what appears in someone else's screenshot. Third, the question "above which threshold does automation start to pay off" has no single numerical answer independent of sector; that is why no threshold figure is given here. Any figure given means nothing until it has been tested in your own account.

What does not change is this: signal quality comes before automation, a change that cannot be measured teaches nothing, and performance that cannot be separated cannot be defended. How campaign structure is built around those three principles, and which items arrive in writing in the monthly report, is set out item by item on our ad management service page. If you would like to work out together which row of the table above your account falls on, you can write to us from our contact page.

Frequently Asked Questions

Is Performance Max suitable for accounts with a small budget?

Suitability depends less on the size of the budget than on how regular the conversion flow is. In an account taking few conversions, a campaign spreading across many surfaces at once thins the data out and may not reach the learning threshold on any single surface. On small budgets the more defensible order is to accumulate clean conversion data first with a narrow, high-intent search campaign, and only once measurement and flow have become stable to try campaign types that widen inventory, with a separate budget.

What is the difference between AI Max and Performance Max?

Performance Max is a separate campaign type and runs across many surfaces at once; budget distribution, placement selection and bidding are largely left to the system. AI Max is presented as an automation layer added on top of search campaigns: extending matching beyond the keyword list, generating text and assets, and routing a query to the most suitable page are offered together. In short, one takes over the whole campaign and the other takes over part of a search campaign. Because the scope of both products changes quickly, the current feature list has to be verified from your own dashboard's documentation.

How much conversion data do you need before moving to smart bidding?

There is no single figure independent of sector. The platform's own documentation carries minimum conversion count recommendations for some strategies, and those thresholds are updated from time to time; rather than memorising the number, it is more reliable to read it from your dashboard's current help documentation. The principle worth memorising is this: in an account gathering only a few conversions a week, strategies that target a cost per acquisition will swing, because noise takes the place of the pattern there was supposed to be to learn.

Can I see which search terms I appeared on in an automated campaign?

Partly. In automated campaign types query visibility is not identical in detail to a classic search campaign, and reporting scope changes over time. The practical approach is to check what the report will show you before you launch the campaign, then to export and archive the data you can see at regular intervals. Keeping brand and non-brand traffic in separate campaigns also wins back part of the lost detail through account structure.

How often can I touch a campaign before it disrupts the learning period?

Every significant setting change triggers the learning phase again, so the real issue is not the size of any one change but how frequently changes are made. An account where bidding, budget and targeting are touched every other day never comes out of the learning phase; part of the volatility you see is created by the account's own restlessness. A workable routine is to write changes into a list over the week along with the reason for each, apply them in one batch once a week, and not touch anything until the next review.