AI Max for Search Campaigns: What the Early Data Shows, the Risks and a Safe Rollout Plan

The screenshot we've been sent most this autumn looks the same every time: a match type in the search terms report labelled "AI Max", a jump in clicks, a CPA that's crept up, and the message "We never turned this on." Usually they didn't. The campaign used the campaign-level broad match setting or automatically created assets, and in September 2026 Google moved it into AI Max for Search on its own.

That's how a lot of advertisers are meeting AI Max: not as a feature they tested, but as a change that arrived in their account. Some are seeing genuine new demand. Others are paying more for queries their exact keywords used to catch at a lower price. And with Dynamic Search Ads (DSA) scheduled to migrate in February 2027, nearly every Search advertiser will have to decide how much of it they want.

This guide covers what Google has actually changed, what Google's numbers and independent data say, how to tell whether AI Max is helping you, and a rollout plan we'd be comfortable running on a client account.

What is AI Max for Search, and what changed in 2026?

Google announced AI Max for Search campaigns on 6 May 2025. It isn't a new campaign type. It's a switch inside a normal Search campaign that turns on three things at once:

• Search term matching, which expands your keywords using broad match and keywordless technology based on your assets and landing pages. It's controlled at ad group level.
• Text customisation, the old "automatically created assets", which writes headlines and descriptions from your site, ads and keywords. Campaign level.
• Final URL expansion, which sends people to whichever page on your site Google thinks fits the query best. Campaign level, and it needs text customisation on.

Alongside those come new controls: brand inclusions and exclusions under Brand settings, URL inclusions and exclusions, and ad-group-level locations of interest. Google's AI Max help page lists where each one lives.

The migration timeline

The bigger news this year is that AI Max stopped being optional for some setups. Google's April 2026 announcement laid out upgrades for DSA, automatically created assets and campaign-level broad match, and a June update pushed the DSA part back. Search Engine Land published the full timeline on 14 August 2026:

When a DSA campaign is upgraded, Google says dynamic ad groups become standard ad groups with settings and data carried over, all three AI Max features are switched on, and legacy URL controls are kept. In other words, the default after migration is maximum automation. If you want less, you'll have to switch it off yourself.

What does the early AI Max data show?

Two sources matter: what Google says, and what independent analysts with access to many accounts have measured. They don't fully agree.

At launch, Google said advertisers who activate AI Max "typically see 14% more conversions or conversion value at a similar CPA/ROAS", rising to 27% for campaigns still running mostly exact and phrase keywords. Its 2026 DSA post added that using all three features delivers an average of 7% more conversions or value than search term matching alone. All three figures come from Google's internal data and cover non-retail advertisers.

Smarter Ecommerce (smec), a Google Ads technology company that works with retailers, analysed more than 250 Search campaigns using AI Max. As Search Engine Journal summarised in March 2026, the median campaign gained 13% conversion value, close to Google's claim. But the incremental conversions cost a median 16% more, ROAS ranged from 42% above baseline to 35% below, and only 22% of campaigns landed near their original targets.

Where AI Max traffic comes from

The most useful finding for day-to-day management is where the extra matches come from. In smec's sample of one million AI Max impressions, 80% expanded from exact match keywords and almost all the rest from phrase. Broad keywords barely registered, which makes sense: broad already matches loosely.

And that share is growing. A later smec analysis of 383 million impressions from EMEA e-commerce campaigns, reported by PPC Land in August 2026, found that by July 2026 more than a quarter of impressions on exact match keywords were being broadened by AI Max.

Our read: AI Max does find queries you'd miss, and Google's headline number isn't fiction. But a median doesn't run your account. The spread is wide, and much of the "new" volume is traffic you were already buying through exact, phrase, broad or Performance Max.

Is AI Max hurting your account? A diagnostic

If AI Max is already on, by choice or by migration, start here. All of these reports live under the Campaigns menu, per Google's AI Max reporting guide.

The "Search terms and ad combinations" view (previously "Search terms and landing pages from AI Max") is the most useful report Google has added in years. It shows each query alongside the headline and landing page Google chose, so you can see exactly why a strange click happened.

Why AI Max underperforms: the root causes, ranked

This order reflects what we see in audits of Search accounts in the US, UK, Canada and Australia since mid-2025. It's practitioner observation, not a controlled study.

1. The conversion signal is weak or wrong

AI Max finds people who look like your converters. If your conversions include duplicate GA4 imports, newsletter sign-ups or spam leads, it will find more of those, cheaply. Fix measurement first; our conversion tracking guide walks through it.

2. It's cannibalising traffic you already had

The smec data shows expansion mostly comes from exact and phrase keywords, and up to 63% overlap with broad match. Add Performance Max and DSA, and three systems can bid on the same query. Reported uplift looks great inside the campaign while total account conversions barely move. Our Performance Max troubleshooting guide covers the overlap from the PMax side.

3. Final URL expansion picks the wrong page

A query about "returns" lands on the returns policy. A careers page gets clicks for "jobs in [your industry]". In lead-gen, a blog post with no form can soak up budget for weeks. URL exclusions fix most of this, but only if someone reads the landing pages report.

4. Text customisation goes off-message

Google writes headlines from your site. If your site mentions a discount from last year, or a claim your legal team wouldn't approve in an ad, it can end up in one. For finance, health and legal advertisers, that's a compliance problem, not a performance one.

5. Search Partners and loose geography

smec found Search Partners typically took 3–8% of AI Max impressions, and in one campaign half of 500,000 monthly impressions, at a fraction of the conversion rate. Locations of interest can also pull in searches about your city from people who'll never visit.

6. Not enough budget or data

Google's own help page warns AI Max won't be effective when a campaign is limited by budget. Small accounts with a handful of conversions a month rarely give it enough to learn from.

A safe AI Max for Search rollout plan

The sequence is the same for a voluntary test and for DSA migration prep. Set the guardrails, test one feature, measure at account level, then expand.

1. Fix tracking and pick the right goalOnly Primary conversions that represent real value: purchases with revenue, or qualified leads imported from your CRM. AI Max on a "form fill" goal in a lead-gen account is how you get junk.
2. Set brand and URL guardrailsIn the campaign's Brand settings, exclude your own brand from non-brand campaigns and add competitor brands you don't want to pay for. Add URL exclusions for careers, blog, policy, login and out-of-stock pages before you switch anything on.
3. Protect exact matchAdd exact-match copies of your top 20–50 converting queries, especially if you're leaving the broad match campaign setting. Keep shared negative keyword lists current; negatives still block AI Max matches.
4. Run a 50/50 experimentUnder Campaigns › Experiments, create a new experiment and choose "AI Max for Search campaigns" as the variable. Since August 2025 this runs inside the original campaign with a 50/50 budget split. Switch off the "apply automatically" option so you decide.
5. Start with search term matching onlyTurn off text customisation and final URL expansion for the first test. You'll learn whether the extra queries are worth buying without changing ads and landing pages at the same time.
6. Judge it at account levelRun at least four weeks, or two full conversion cycles if your sales take longer. Compare total account conversions and cost, not just the experiment campaign, so cannibalisation shows up.
7. Add features one at a timeIf search term matching passes, test text customisation next, then final URL expansion. Google's own number for the full suite over search term matching alone is a 7% average gain, so don't expect miracles from the last two.
8. Upgrade DSA before Google doesDo the DSA upgrade yourself between now and January 2027, with your URL controls and negatives already in place, rather than inheriting Google's all-features-on default in February.

Mistakes that make AI Max worse

• Switching everything on at once. When CPA jumps, you won't know if it was the queries, the ad text or the landing pages.
• Judging the experiment inside the campaign only. A 20% lift in one campaign with a flat account total means you moved conversions, not added them.
• Deleting exact keywords because "AI Max covers it". In the cases we've reviewed, exact match on proven queries is usually still cheaper.
• Ignoring the landing pages report. Final URL expansion problems are invisible in campaign-level numbers until someone opens it.
• Running AI Max, DSA, broad match and PMax on the same products. Pick which campaign owns which queries and use negatives and brand settings to enforce it.
• Leaving DSA to the February migration. You'll inherit every feature switched on, during a month when your team should be watching budgets.

How to monitor AI Max once it's live

Honestly, AI Max needs more weekly attention than exact match campaigns did, not less. Here's the routine we run:

• Weekly: search terms report filtered to Match type "AI Max", sorted by cost. Add negatives for anything irrelevant, and note good new queries to add as exact keywords.
• Weekly: keywords report totals for AI Max expanded and landing page matches, with cost per conversion against your keyword rows.
• Fortnightly: landing pages report and Expanded final URL assets tab. Exclude any page you wouldn't choose yourself.
• Monthly: ad-level asset report for Google AI headlines; remove anything off-brand or non-compliant.
• Monthly: total account conversions, cost and revenue against the pre-AI Max baseline, plus overlap with PMax search themes.

Worked scenario: a Melbourne furniture retailer after the September migration

Illustrative example. The business and figures are invented to show the method.

A Melbourne furniture retailer that ships across Australia spends AUD 60,000 a month on Search. Its main non-brand campaign used the campaign-level broad match setting and automatically created assets, so in September 2026 Google migrated it to AI Max with search term matching and text customisation on. In the four weeks before migration it delivered 300 online orders at an AUD 120 CPA. In the four weeks after, it delivered 330 orders at an AUD 140 CPA, and the campaign report credits 90 of them to AI Max rows. On the campaign dashboard, that looks like a win.

The account-level view tells a different story. Total orders across all Search and Performance Max campaigns rose from 520 to 540, not by 30 or 90. Most AI Max orders came from queries that exact keywords and PMax had been winning before, such as "oak dining table Melbourne". Worse, Google-written headlines are still promising an end-of-financial-year sale that finished on 30 June, because the old sale banner was never removed from a category page.

The team turns off text customisation, sets brand exclusions, adds exact copies of the 40 best-converting queries, removes Search Partners and sets up a 50/50 AI Max experiment with search term matching only. Six weeks later the test arm shows 4% more account-level orders at a CPA within 5% of the control. That's a modest, real gain. The team keeps search term matching on and plans a separate final URL expansion test once URL exclusions are in place.

Comparing AI Max with automation on the social side? Meta's version of the trade-off is in our Advantage+ campaigns guide, and Google Ads vs Meta Ads helps decide where budget goes. Final URL expansion is only as good as the pages it can choose from, so our landing page examples are worth a look too.