Google "AI Max": When One Switch Makes Three Decisions – and Who Looks Before the Budget Increase

You have probably already flipped a switch in your Google Ads account that decides at the same time which searches your ad appears for, what the ad says and which page the click leads to. Or someone flipped it for you. Do you remember when you last checked? And what it costs you if the machine gets it wrong?

Google calls the switch "AI Max for Search". It is not a new product: Google introduced it in May 2025, and since April 15, 2026 it has, according to the source we rely on, been out of beta. What is new is something else, namely that existing campaigns were migrated automatically in September 2026 and more will follow. Anyone using Google Ads is therefore very likely affected, whether they decided so consciously or not.

This article is an evergreen topic, not a news flash. It asks a basic question: how do you recognize in your own ad account that rule-based automation has turned into an autonomously deciding agent? AI Max is a vivid example. We assign the topic to Daniel, our analytics agent, whom we describe on our website with the sentence "Daniel tells you whether it adds up." The article ends with the difference between an agent with limits and an approval step and automation that simply runs.

What "AI Max for Search" does: three decisions under one switch

Let us start with what is confirmed. Google's own launch post of May 6, 2025 is titled "Unlock next-level performance with AI Max for Search campaigns" and was written by Brian Burdick, Senior Director of Product Management at Google Ads. PPC Land, a trade publication for online advertising, published a detailed analysis on October 25, 2026. Both sources describe the same three building blocks.

1. Search term matching. Your existing keywords are expanded using broad match and "keywordless" technology. Google says it expands on existing keywords with these means. In practice, your ad can also appear for searches that no longer directly match any of your stored keywords.

2. Text customization. This feature used to be called "automatically created assets". The system generates new headlines and descriptions based on your landing page, your existing ads and your keywords. According to PPC Land, these generated texts are reviewed and refreshed at least every 48 hours. That is an upper limit for the interval: at the latest after two days, a change on your website can reach the ad text.

3. Final URL expansion. Users are sent to the page on your domain that Google considers most relevant. According to PPC Land, this feature requires text customization, and when it is active, pinned headlines are not taken into account.

The point is the bundling. These used to be three separate decisions a human made in the account: which search terms, which text, which landing page. Now they run bundled under one switch, without each individual one being confirmed by you beforehand.

Automation follows rules you have set. An agent decides by itself within a frame. With AI Max, the question is not whether it can be useful, but whether you know the frame.

Timeline: what has already happened and what is still coming

PPC Land gives the following key dates, which we reproduce here:

• On May 6, 2025, Google announced AI Max. The global beta launch was at the end of May 2025, according to the source.
• On April 15, 2026, Google took AI Max out of beta. The source does not use the term "generally available", but in substance the statement matches it.
• Between September 1 and September 30, 2026, affected campaigns were migrated automatically to AI Max. Whether your campaigns were among them, you can see only in your own account.
• The automatic migration for Dynamic Search Ads was postponed on June 11, 2026 to February 2027. If you use Dynamic Search Ads, note this date.

What the source does not name: country-specific exceptions. The article lists none. That means none are named, not that we can confirm availability for your account in Germany, Austria or Switzerland. The source also says nothing about your specific account, budget and industry.

What Google promises and what independent analyses show

The numbers on AI Max contradict each other, which is exactly why they are instructive. We put them side by side and name the source and its limits each time.

Google's statements. In the launch post, Google writes that advertisers who activate AI Max typically see "14% more conversions or conversion value at a similar CPA or ROAS". For campaigns that mainly use exact and phrase keywords, Google cites up to 27%. A footnote refers to internal Google data from 2025 and is limited to non-retail advertisers whose conversions come more than 70% from such keywords. That footnote is attached to the sentence about the 27% figure; whether it also applies to the 14% is not clear from the post. Philipp Schindler, Alphabet's finance chief, cited an average uplift of 15% in the quarterly call on July 22, 2026, according to PPC Land.

Independent findings. Set against this are analyses that PPC Land reports:

• The agency Smarter Ecommerce found, in an analysis from November 2025 across more than 250 retail campaigns, an ROAS about 35% lower for AI Max compared with other match types.
• In a four-month test, the cost per conversion was 100.37 US dollars for AI Max and 52.69 US dollars for exact match. The test is from Xavier Mantica.
• According to Smarter Ecommerce, in July 2026 about 29% of exact-match impressions in a sample of 383 million impressions from the EMEA region had been broadened by AI Max.

An important caveat is in the article itself: none of these figures comes from a controlled, published study, including Google's own values. They are vendor statements, agency analyses and individual tests. They are defined differently, cover different industries and periods and cannot be added up into one number. We deliberately draw no conclusion of the kind "AI Max is good" or "AI Max is bad".

The reliable conclusion is a different one: whether AI Max works for your account cannot be answered for you by Google or by an agency analysis. Only your own account shows that. And for that, you have to look.

What you can control, and what not

PPC Land describes the controls, and they are unevenly distributed.

• Search term matching: can be switched off for individual ad groups. Brand inclusions and exclusions exist at campaign and ad group level.
• Text customization: the article describes no separate opt-out, only text guidelines. According to the source, these allow up to 25 term exclusions and 40 message restrictions per campaign.
• Final URL expansion: the article likewise describes no opt-out. It is steered through URL inclusions and exclusions.
• Negative keywords: according to the article, they apply only to literal strings and are therefore incomplete.

Google's launch post also names controls: locations of interest at ad group level, brand controls at campaign and ad group level, removal of generated assets and URL controls. The post names no explicit opt-out option. An opt-out per component can therefore not be substantiated. What is confirmed: only search term matching can be switched off for individual ad groups, according to PPC Land.

How to tell whether automation has become an agent

A simple rule of thumb we use for decision makers: for every function in your ad account, ask who makes the decision and who sees it beforehand.

Rule-based automation: you set a rule ("Pause ads with costs above X"). The machine executes it. Every decision can be traced back to your rule.

Agent: you set a goal and a frame. The machine decides by itself how to reach the goal and picks means you did not specify individually. With AI Max, these are the search queries, the texts and the landing pages.

Both can make sense. But with an agent, the question to you changes: not "Did I set the rule correctly?" but "Does what the machine decided still fit what I wanted?" That can only be answered by looking.

Three checks for your account

Here is the concrete part. These checks need no special knowledge, only time and access.

1. Check the search terms report. Look at which actual searches your ad appeared for. After activating functions with "keywordless" technology, actual targeting can change independently of your keyword maintenance. Whoever controls their keyword list no longer fully controls when the ad appears.
2. Read the generated texts. Do they fit your brand, your offers and your legal requirements? If a text makes a statement you would not make that way, it is your statement, because it appears under your name.
3. Check the chosen landing pages. Does the click lead where your plan intended? If Google sends users to a different page on your domain, your conversion rate changes without you having changed anything.

With and without an approval step

The difference can be shown well with AI Max.

Without an approval step, keyword expansion, text generation and page switching run bundled and immediately. You see the result only in the report, that is, after budget has been spent. If a generated text does not fit or a landing page converts worse, you pay for it before you notice.

With an approval step, a human looks at which search queries, texts and landing pages the automation has meanwhile chosen itself, before the budget is increased. The approval step does not prevent the automation from working. It ensures that someone looks before more money flows in.

That is how we describe the work of our agents too: our website says critical actions need your approval. Daniel analyzes performance metrics, reporting and anomalies across all channels and shows where something does not add up. The decision on what follows stays with you.

Three objections we often hear

"Google knows better than we do." Google knows the search data, but not your margin, your delivery capacity and your brand rules. A figure like "14% more conversions" says nothing about whether those conversions are profitable for your company. That is not criticism but the reason someone who knows your numbers has to look.

"We never switched it on." That may be true. The automatic migration in September 2026 means accounts can be affected in which nobody consciously agreed. That is exactly why a look at the account is worthwhile.

"Then we switch everything off." That is also an option, but according to the source only partly: opt-outs are not described for every building block, according to PPC Land. It usually makes more sense to set the frame and check regularly.

What you can do this week

1. Note the date: if you use Dynamic Search Ads, the automatic migration is due in February 2027 according to PPC Land. Check now whether and how you are prepared.
2. Take a sample: open the search terms report for the last 30 days and read 50 rows. You need no tool for it, only attention.
3. Set one approval rule: write down who must see the automatically chosen search queries, texts and landing pages before a budget increase.

Summary: automation with a frame

AI Max for Search bundles three decisions that used to be made separately by humans. That can save effort and, according to Google, bring better results. Independent analyses also show clearly worse figures, and none of the figures cited comes from a controlled study. The responsible way to deal with it is not yes or no but a frame: you know which switches are active, you regularly check what the automation does with them, and a human looks before more budget flows in.

That is the difference between an agent with limits and an approval step and automation that simply runs. One works for you as long as you control it. The other works whether you look or not.

Sources

• PPC Land, Luís Rijo: "AI Max", ppc.land/ai-max/, October 25, 2026 (primary source for timeline, controls and independent analyses).
• Google, Brian Burdick: "Unlock next-level performance with AI Max for Search campaigns", blog.google, May 6, 2025 (functions, 14% and 27% statements).