AI Max is Google’s optimisation layer for Search campaigns. Turn it on, and Google expands your keyword matching with broad match and keywordless technology, generates ad copy drawn from your own site, and routes clicks to whichever landing page it predicts will convert best. It is not a new campaign type. It applies on top of the Search campaigns you already run.
Here is the part most write-ups bury: it works, but the efficiency is close to a coin toss unless you control it tightly. Independent testing tends to land near Google’s own claim on volume, a lift in roughly the low-teens range, but at a noticeably higher cost per conversion, and with ROAS sitting around flat on average. The real catch is the spread: results swing widely either side of that average, comfortably up for some accounts and well down for others.
So the real question is not “is AI Max any good?” It is “can you steer it.” This guide is for advertisers running mature Search campaigns who want the additional reach without ceding control of the spend.
Three features carry most of the weight.
This is the engine. AI Max expands beyond your keyword list using broad match and keywordless technology; it learns from your existing keywords, ad copy and landing pages to match queries you never added. Google’s own description is that it expands on existing keywords with broad match and keywordless technology. Turn AI Max on for the campaign, and search term matching switches on at the ad group level by default. You can toggle it off there if you want the rest of the suite without the matching.
Formerly Automatically Created Assets. AI Max generates headlines and descriptions from your site content and existing ads, tailored to the specific search. Useful at scale. Generic in practice, it pulls product names and features competently, but it does not grasp your positioning or seasonal angle.
Google sends a click to the page on your domain it predicts will perform best for that query, not only the URL you set. It fires only on pages themed to your ad group. One detail worth knowing before you enable it: when FUE selects a more relevant URL, your pinned RSA assets are ignored because they may not match the page the user lands on. FUE also requires text customisation to be on, so the ad copy and the landing page stay aligned.
Around those three sit the controls that make AI Max defensible: locations of interest (geographic intent at ad group level), brand inclusions and exclusions, URL inclusions and exclusions, negative keywords, and reporting upgrades, including a new “AI Max” match type and a match source column that tells you whether a query came from broad match expansion or keywordless matching.
AI Max is not one switch. It is a panel. The reach features (search term matching, FUE) are where the volume comes from, and the control features (negatives, brand lists, URL exclusions) are how you prevent that volume from turning into waste.
This is the question every experienced PPC manager asks first, and the honest answer is partly.
Pull the report yourself: cross-reference your keyword match types against the search term match types. In practice, the bulk of the expansion lands on exact and phrase match keywords, and it barely touches broad match, because broad is already broad. Your exact-match keyword still wins when the query is identical. Keyword priority has not changed.
But “just broad match” undersells it. Check the match source column, and you will usually see broad match expansion and keywordless matching each contributing roughly half the weight. Keywordless is the genuinely new arm; it matches on landing page content and ad group context the way DSA and Performance Max do, without you seeding a keyword at all. That is the real shift here: away from keyword syntax, toward intent matching.
Most of the confusion comes from overlap. These tools do similar work in different containers.
|
AI Max for Search |
Standard Search |
Performance Max |
|
|
What it is |
Feature layer on a Search campaign |
Traditional Search campaign |
Separate cross-channel campaign |
|
Where ads run |
Google Search + Search Partners |
Google Search + Search Partners |
Search, Display, YouTube, Gmail, Discover, Maps |
|
Keywords |
Optional, AI expands past your list |
Required |
None |
|
Search term visibility |
Yes, marked “AI Max” with a source |
Full |
None |
|
Ad copy |
You write it; AI generates extra variations |
Manual |
AI generates everything |
|
Landing page |
Optional URL expansion |
Fixed per Ad |
AI decides |
|
Negative keywords |
Supported |
Fully supported |
Limited (brand lists only) |
|
Control |
Moderate, keywords and terms stay visible |
High |
Low |
The distinction that matters: AI Max enhances a Search campaign. Performance Max replaces the structure entirely and trades your visibility for reach across every Google surface. If you want automation but are not willing to give up the search term report, AI Max is the version that lets you keep it.
Yes, eventually. Google has confirmed the long-term plan is to fold Dynamic Search Ads into AI Max for Search, aiming for the same use cases and parity with PMax on Search. There is no firm date, but Google’s past deprecations tend to run about a year from the sunset start.
So there is no urgent deadline. There is a clear strategic signal. If a portion of your spend still runs through DSA ad groups, the considered move is not to migrate them blindly to PMax; it is to begin activating keywordless AI Max features inside your existing Search campaigns and wind DSA down on your own terms. That is where you retain cleaner budgeting, measurement and control. Control is the priority throughout.
Google’s headline claim: campaigns that enable AI Max typically see low-double-digit growth in conversions or conversion value at a similar CPA or ROAS, and for campaigns still relying on exact and phrase keywords, Google puts the uplift higher again, up into the mid-to-high twenties.
Read that closely and a few things stand out.
The higher of those figures is aimed squarely at exact and phrase-match loyalists, and for them, it is plausible. Smart Bidding plus broad match expansion genuinely finds volume that an over-curated account is leaving on the table. That is a real, cost-efficient opportunity for advertisers who have avoided automation until now.
For an already well-optimised account, a double-digit volume lift at the same efficiency quietly defies the law of diminishing returns. And independent testing supports the scepticism: the extra volume tends to show up, but the cost per conversion usually creeps up alongside it. When advertisers asked Google’s own Ads Liaison why AI Max CPA looked worse than their other match types, the answer was, in plain terms, the cost of incrementality: the next marginal conversion costs more than your hand-picked baseline keywords. Which is exactly how performance marketing works. It is simply not what the sales slide says.
One further detail Google places in a footnote: the uplift figures do not cover retail. There is no official claim for e-commerce yet.
A flat average on ROAS sounds like “no change.” It is not. The outcomes are spread wide rather than clustered in a tidy bell curve; some accounts come out well ahead, others well behind. Enabling AI Max without controls is genuinely close to a coin flip on efficiency. The task is to shift those odds in your favour.
Four failure modes appear repeatedly. None of them is reasons to avoid AI Max. There are reasons to monitor it.
The broad match arm should not expand on keywords that are already broad, yet in some accounts it ends up matching against existing broad match far more often than it should. The usual culprit is legacy Broad Match Modified keywords that report as broad but behave like a phrase. The fix: clean up the legacy match types, then re-check the source column.
AI Max seeks out high-intent traffic, and competitor brand queries fit that profile precisely. Left unchecked, it can pour a large share of its expansion into competitor terms — exactly the traffic many advertisers deliberately avoid for brand-safety or truce reasons. The fix: add competitor terms as negatives, set brand exclusions, or disable search term matching for the affected ad groups.
Google provides a rich report unifying the query, headline and landing page, but it can balloon to an unworkable number of rows, most of them irrelevant. Heavy on data, light on action. The fix: segment rigorously. Filter to the combinations with real spend and conversions, and set aside the long tail of noise.
AI Max normally sends only a small share of impressions to Search Partners. Occasionally, it scales into that inventory without warning, and because the intent there is generally much weaker, you can burn through impressions for very little return. The fix: audit placements, and if SPN intent is that weak, switch it off entirely.
Avoid building a fresh AI Max campaign from scratch. Upgrade an existing Search campaign that already holds conversion history; Google’s models need that data to behave.
The cleanest test is a Draft + Experiment: split the budget evenly between the campaign as it stands (your control) and the same campaign with AI Max enabled (your trial). Traffic stays inside one campaign, so the learning period is shorter, and the read is cleaner. It will not run if the campaign sits in a portfolio bid strategy, a shared budget, or already has an active experiment, worth checking first.
A rough timeline that holds up: leave settings unchanged for week one, monitor closely in week two, begin adjusting search terms, negatives and targets at the ad group level from week three, then compare a full four-week window against the prior period. The learning phase itself runs about two weeks.
Two cautions. Be careful directing AI Max at brand campaigns; it has a tendency to lean into branded searches that were already covered. And when a Google rep pushes a minimum daily budget so the model has enough signal, treat it as a data point rather than a rule. Reps have their own targets. The only reliable judge is your own data, so test, validate, and let the numbers decide.
This is where expectations need managing, particularly with clients.
AI Max optimises within one campaign. It does not reallocate budget across campaigns. It will not notice your brand campaign cannibalising non-brand, will not restructure your account, will not build a keyword strategy, and has no awareness that you are running a promotion or launching in a new market unless you tell it manually. It does not know your Meta spend or your email calendar. And its reporting, while better than PMax, still will not explain why its models made a given matching or audience decision.
So it is a sharper lever, not a strategist. The cross-campaign decisions. The ones that actually move your blended CPA or ROAS still require a person looking at the account as a whole. AI Max is one input into that work, not a replacement for it.
It comes down to where you are starting from.
If you are an exact match loyalist, you are Google’s prime target, and the upside it is dangling is real. A controlled test is warranted, even if it runs against instinct.
If you are a hybrid running a mix of match types, expect a low-double-digit lift at best, with inconsistent efficiency. Commit a modest budget under strict testing rules. If it works, scale it.
If you are an automation-heavy account already running DSA, PMax and broad match aggressively, the uplift will be modest, but you should be winding down legacy DSA now, ahead of the migration, rather than scrambling when the date arrives.
For most advertisers with mature campaigns, solid conversion tracking and comprehensive negative lists, AI Max is worth enabling. Build the controls first and keep the search term report open daily for the first fortnight.
The keyword has not died, and it might never. But the era of intent matching is here, and it is likely to remain. It is better to develop the competence to steer it now, on a controlled budget, than to inherit the feature unprepared when DSA is retired.
Considering a second review before enabling AI Max across live campaigns? Contact us and we will pressure-test your negatives, match types and tracking first, so the test produces a meaningful result.
Why is AI Max sending traffic to a broken landing page (404 errors)?
The cause is almost always a clash between your tracking template and final URL expansion. AI Max swaps your set URL for a dynamic landing page, and a static tracking URL or a non-standard LPURL tag breaks the result. Before enabling AI Max, confirm a dynamic landing page resolves, and use the supported patterns: {lpurl}, {lpurl}?, {lpurl}& or {lpurl}#.
Does AI Max serve ads in AI Overviews and AI Mode?
It can, but you do not need AI Max to appear there. Broad match, Shopping and Performance Max already serve in AI Overviews and AI Mode. More to the point, AI Max settings apply to all your Search traffic, not only the AI surfaces. Do not enable it out of FOMO about AI placements; judge it on overall performance instead.
Is AI Max available in the Google Ads API and Editor?
Not yet. AI Max is not supported in the Google Ads API or Editor, so activating or deactivating it can throw errors for API requests managing text customisation and brand settings. If your team manages campaigns programmatically, tell them before you toggle AI Max on or off, or you risk silent breakages in those workflows.
What bidding strategies work with AI Max?
AI Max requires Smart Bidding. Eligible strategies are Target CPA, Target ROAS, Maximize Conversions and Maximise Conversion Value. Manual CPC campaigns cannot run it. AI Max does not change your bidding logic; it expands reach within the goal you have set, so you also need active conversion tracking with enough volume for the models to learn.
Does AI Max work for e-commerce and retail?
Unclear, and Google has been quiet on it. The headline uplift figures exclude retail in a footnote, and there is no official performance claim for ecommerce yet. That does not mean it fails; it means you test it yourself, watch search term quality and ROAS closely, and trust your own data over the absent benchmark.
How long before I can judge AI Max results?
Give it six weeks. The learning period runs about two weeks, and you want roughly four more weeks of stable delivery before reading anything into the numbers. Then compare that window against the prior period on conversions, CPA and ROAS. Judging AI Max inside the learning phase is how good campaigns get switched off too early.
