AI Max Is On. Is It Paying for Itself or Padding Your Numbers?
Google gave itself a September 1 to 30 window to auto-upgrade Search campaigns using broad match or automatically created assets into AI Max. By late September most of those accounts have made the switch whether the owner noticed or not. AI Max is Google's setting that lets its AI show your ads on searches beyond the keywords you picked, using your landing pages and ad text as clues. Most guides on this tell you to track what percentage of conversions AI Max drives and to compare your results with Google's promised 14% lift. That percentage is the wrong scoreboard for an owner with a fixed budget. The only question that decides whether AI Max stays is whether its slice of your spend wins customers at or below your own break-even cost per customer.
The switch was announced, not negotiated. Search Engine Land reported on August 14, 2026 that Google set the September 1 to 30 window for Search campaigns using campaign-level broad match or automatically created assets. Broad match is the keyword setting that lets Google show your ad on searches related to your keyword, not only the words you typed. If your campaigns have not flipped yet, start with the checks to run before Google moves your ads to AI Max. Everything below assumes the switch already happened and asks what comes next: is it earning its keep?
Google's 14% is an average, and your budget lives in one account
Google's own number sounds reassuring. Brian Burdick, Senior Director of Product Management at Google, made the promise in the Google Ads blog announcement of AI Max. Advertisers who activate it "will typically see 14% more conversions or conversion value at a similar CPA/ROAS," he wrote. The figure rises to 27% for campaigns still running mostly on exact and phrase keywords. CPA is cost per acquisition: what you pay in ad spend for each conversion Google records. The footnotes narrow the promise. The figures come from Google's internal 2025 data on non-retail advertisers, and the 27% applies only to campaigns getting more than 70% of their conversions from exact or phrase match.
An average is the middle of a spread, and SMEC's data shows how wide that spread runs. Mike Ryan, Head of Ecommerce Insights at Smarter Ecommerce (SMEC), analyzed more than 250 ecommerce Search campaigns using AI Max, as reported by Brooke Osmundson in Search Engine Journal on March 5, 2026. ROAS, or return on ad spend, is the revenue you get back for every dollar you put into ads. Across those campaigns, ROAS outcomes ranged from 42% above the original baseline to 35% below it. Only 22% came in close to their original ROAS target.
The other 78% over- or under-performed significantly. A campaign at the bottom of that range earns 35% less revenue per ad dollar than it did before the switch. The headline lift figure gives no warning that you are that campaign. You do not get the average. You get your account.
I have run Google Ads for 10 years across 300+ businesses, and the accounts that taught me to distrust averages were the lead-gen ones. The London ADHD clinic I kept booked solid for three straight months is the standard I hold every lead-gen account to: the result that counted was a full calendar, not a flattering report. In an account like that, a conversion only matters if it becomes a booked assessment. An ad shown on a loosely related search produces a click, sometimes a form, and no appointment. Google counts the click and the form. The clinic's calendar only counts the appointment. Broad expansion into searches that come close to your service without being your service is exactly how a reported lift turns into money spent on people who will never book.
Your blended cost per lead is covering for the AI Max slice
The first mistake is judging the campaign by one blended figure, its overall cost per lead. If that holds steady after the switch, AI Max looks harmless. It is not a fair test. The blended figure averages traffic from keywords you chose, which you have tuned over months or years, with traffic from searches Google chose for you. Strong keywords can quietly carry a weak expansion, and the blended cost per lead will look fine the entire time the expansion burns money.
The second mistake is reaching for the Search Terms report. Armen Sargsyan, co-founder of 405 Ads, pointed out in Search Engine Land on September 23, 2026 how big the "Other search terms" line can get. That line, where Google lumps the queries it does not itemize, can be 40% or even 50% of total search activity in some accounts. In those accounts, close to half the picture is missing. You cannot work out AI Max's share from a report like that, and if you try, you end up judging the feature on whichever searches Google decided to show you.
The third mistake is treating a conversion as a customer. Google's conversion column counts whatever action you told it to count, and for most service businesses that is a form submission or a phone call. A form from someone who typed a loosely related query is still a form in the report. It is not a patient, a signed client, or a paid invoice, and your bank account knows the difference even when the dashboard does not.
Sargsyan's piece also shows where the reliable numbers sit. Google splits AI Max into two reportable buckets, and both roll up next to your own keywords at the bottom of each campaign's Keywords tab.
| Summary row | What it counts | The question to ask it |
|---|---|---|
| Total: Your keywords | Results from the keywords you picked | Your baseline. What does the campaign earn on your terms? |
| Total: AI Max expanded matches | Searches Google matched beyond your keywords | Are these people who buy what you sell? |
| Total: AI Max landing page matches | Searches Google found from your landing pages and assets | Is Google reading your pages the way a customer would? |
| Total: Campaign | Everything above, combined | None. It is the denominator for your AI Max share. |
AI Max buys you volume, and that volume costs more than your keywords
SMEC's median result is where the price shows up. Across the campaigns Ryan studied, AI Max delivered a median 13% uplift in conversion value and a median 16% increase in CPA. If a campaign hit both medians, one that paid $100 per conversion before the switch would now pay about $116 to bring in about 13% more value. The two medians come from separate campaigns. Treat that as an illustration of the trade, not a forecast. Whether that trade makes you money comes down to what a customer is worth to your business, and no study can answer that for you.
Search Engine Journal's takeaway from SMEC's data: AI Max behaves more like a layer that adds volume than a tool that makes your account more efficient. The incremental conversions also tend to cost more than the traffic your keywords already bring in. Volume at a higher price is a perfectly good deal for some businesses. For others it is a slow leak.
Of one million AI Max impressions SMEC analyzed, 80.11% of the expansion started from exact match keywords, 19.52% from phrase match, and just 0.38% from broad match. Exact match is the setting you use when you want your ad shown only for a specific search. AI Max most often broadens the very keywords you locked down on purpose. Your tightest keywords are its launch pad.
| What is measured | Google's claim | SMEC's findings |
|---|---|---|
| Lift | 14% more conversions or conversion value; 27% for campaigns with more than 70% of conversions from exact or phrase match | Median +13% conversion value |
| Cost per conversion | At a "similar CPA/ROAS," per its announcement | Median CPA +16% |
| Hitting the target | Not reported | 22% close to ROAS target; 78% significantly over or under |
| ROAS spread | Not reported | From 42% above baseline to 35% below |
| Who was measured | Non-retail advertisers, Google internal data, 2025 | 250+ ecommerce Search campaigns |
The two datasets do not measure the same businesses. Google's figure covers non-retail advertisers, while SMEC's covers ecommerce. A straight comparison is not apples to apples. They are still the two best public numbers available. They agree on the part that matters to you: the lift exists in aggregate, and its cost swings widely from one account to the next. A dental practice, a law firm or a roofer in Tampa running lead-gen is closer to Google's sample than SMEC's. Even then, test your own slice instead of borrowing either average.
Sargsyan calls it broad match with extra reach, and the reach comes from your own landing pages.
His own worked example shows how uneven it gets inside a single account. Across five campaigns, AI Max produced 36 of 221 total conversions, or 16.29%. Campaign by campaign, the share ran from 8.33% to 29.17%. Same advertiser, and one campaign leaned on AI Max three and a half times harder than another. A single account-wide AI Max figure blurs all of that. Every judgment in the rest of this post happens one campaign at a time.
Pull the AI Max share from the Keywords tab, then price it against a real customer
You can do this yourself inside Google Ads, or hand the exact method to whoever runs your account. It uses only the summary rows from the table above.
The percentage tells you how much of the campaign AI Max now drives. It does not tell you whether that is good. For that you need the cost side, and the same summary rows carry the cost column. Divide the combined cost of the two AI Max rows by their combined conversions and you have AI Max's own CPA, separated from the keywords you chose.
Google will never calculate the next number for you. It comes from your business, and your ad account cannot see it. Your break-even cost per customer is the most you can spend to win one customer before that customer loses you money. Picture your own physio clinic. Say a new patient is worth $600 in gross profit over a course of treatment, and one in three enquiries turns into a booked patient. Your break-even cost per customer is $600. Your break-even cost per enquiry is a third of that, $200, because you pay for three enquiries to win one patient. These are illustration numbers; replace them with yours before you trust anything that follows.
Owners then assume AI Max enquiries turn into patients at the same rate as enquiries from your own keywords. Loosely related searches bring loosely related people. Stay with the illustration: if your keywords and AI Max both report $150 per enquiry, your keywords look safe at one booking in three, or $450 per patient. If only one in five AI Max enquiries books, that identical $150 becomes $750 per patient, well past your $600 break-even. Same CPA on the dashboard. One slice makes money and the other loses it on every patient.
The cleanest fix is making the conversion Google counts the one that means money: a booked appointment or a qualified call, sent back into Google Ads. A raw form fill is not enough. In practice you import booked appointments from your booking system or CRM into Google Ads as the conversion it optimizes toward. It is a one-time setup job for whoever manages the account, and fixing what your Google Ads account counts as a conversion walks through it. With that in place, the AI Max rows report customers directly and the math above does itself. Until then, open the campaign's Search Terms report. It is still the right place to read individual queries, even though it cannot compute the share. Look at the AI Max searches it does itemize and put one blunt question to each: would a person typing this buy what we sell? If you sell online or bid to a revenue target, run the same logic on revenue instead of patients.
If an agency runs your account, send them one question. "For each Search campaign, what share of conversions came from AI Max last month, what did those conversions cost, and how many became paying customers compared with our own keywords?" An agency that answers with Google's 14% has not opened your Keywords tab. If they cannot answer the paying-customers part, your conversion tracking is the first thing to fix.
Watch the slice weekly, and make the keep, cap or kill call monthly
Sargsyan recommends monitoring AI Max % of total weekly at the campaign level, which is the right rhythm for spotting trouble early. When a campaign's AI Max share jumps, he asks four follow-up questions. Is the traffic relevant? Is it converting at an acceptable CPA? Is Google finding useful new query themes, or is the expansion getting too aggressive? Only one of those words needs translating for your business. Acceptable means at or below your break-even per customer, not "roughly what we paid last month."
Weekly is for watching, not deciding. In a small account, a single week holds a handful of conversions, and a handful swings on one good or bad day. Make the call on a full month of data, per campaign, against your break-even. Capping means adding negative keywords, which tell Google never to show your ad on searches containing a given word or phrase. You trim the themes that never turn into customers and keep the ones that do. Killing means switching AI Max off in that campaign's settings. Before you do, read how to test AI Max inside a contained box and what leaving it costs, because walking away carries consequences of its own.
Some numbers will look like progress and prove very little. Total conversions rise because AI Max adds volume, which is exactly what SMEC's data says it does. The AI Max percentage is neither good nor bad on its own: a 29% share that wins customers under break-even beats an 8% share that loses money on each one. The blended campaign CPA is the most dangerous of them, because a set of strong keywords can hide an expansion that burns cash long before the total moves.
Frequently Asked Questions
How do I see how many of my conversions come from AI Max?
Open a Search campaign, go to the Keywords tab and scroll to the summary rows at the bottom. Add the conversions in Total: AI Max expanded matches and Total: AI Max landing page matches, then divide by the conversions in Total: Campaign. That is the method Armen Sargsyan of 405 Ads laid out in Search Engine Land in September 2026. Skip the Search Terms report for this, because its Other search terms line can hide 40% to 50% of searches in some accounts.
What is a good AI Max percentage for my campaign?
No percentage is good or bad on its own. In Sargsyan's worked example, AI Max drove 16.29% of conversions across five campaigns, with individual campaigns ranging from 8.33% to 29.17%. A high share is fine when those conversions cost less than your break-even cost per customer, and a low share is still a problem when every one of them loses money. Judge the cost of the slice, not its size.
Is AI Max worth keeping if I have a small Google Ads budget?
Yes, in any campaign where its conversions turn into customers at or below what a customer is worth to you. A small budget makes that check more urgent, not less. SMEC's study of 250+ ecommerce campaigns found a median CPA increase of 16%. It also found 78% of campaigns significantly over- or under-shooting their ROAS target. Pull AI Max's own cost per conversion from the Keywords tab and compare it with your break-even. Cap it with negative keywords or switch it off in the campaigns where it keeps losing.
Google reports AI Max in its own currency: conversions, lift, and averages drawn from other people's accounts. Your business runs on a different one, the cost of a customer against what that customer is worth. For years, the keywords you chose were the one part of your account you fully controlled. Since September a slice of your budget answers to Google's guess about what your customers search for, starting from your best exact-match terms. If you want a second pair of eyes on that slice against your own break-even, book a call with me and bring last month's Keywords tab.
