Google turned on AI Max in your account. Did anyone check the leads?

If your Google Ads account had a Search campaign using the campaign-level broad match setting, or with automatically created assets switched on, there is a good chance it is not the campaign you built anymore. Google spent September converting those campaigns to AI Max, its keyword-optional search product, and the window closed on September 30. Nobody had to click yes.

I don't think AI Max is a scam, and I don't think it's a gift. It is a machine that finds more of whatever you told Google a conversion is. For a lot of B2B and service businesses, what they told Google a conversion is was "someone filled in a form." That is the real problem, and it predates AI Max by years. AI Max just makes it more expensive.

What actually changed on September 1

Google announced this in April. Brandon Ervin, a director of product management at Google Ads, wrote that campaigns using automatically created assets and the campaign-level broad match setting would be auto-upgraded starting in September 2026. The Dynamic Search Ads migration was on the same schedule until a June update pushed it to February 2027. The two settings most small accounts actually use got no delay.

What each campaign got depends on which setting it had. PPC Land's summary of the account notices lays it out: campaigns with automatically created assets were switched to AI Max with search term matching and text customization on, and campaigns with the broad match setting got search term matching only. The only ways to avoid it were to turn the old feature off, or to turn AI Max on yourself first.

Search term matching is the part that matters. Google's own help documentation describes it as "broad match and keywordless technology" that learns from your keywords, your ads, and your landing page copy to decide which searches you show up for. Your keywords stop being the list of searches you pay for. They become a hint.

That changes what a Google Ads account is. For twenty years the keyword list was the strategy. Now your landing page is part of the targeting, and the keyword list is one input among several.

Google's number, and what it measures

When Google launched AI Max in May 2025, the headline claim was 14% more conversions or conversion value at a similar CPA or ROAS, rising to 27% for campaigns running mostly on exact and phrase match. In the April 2026 upgrade announcement, the claim is 7%, for non-retail advertisers, using all three features: search term matching, text customization, and final URL expansion.

Those two numbers come from different datasets a year apart, so I wouldn't read the drop as proof of anything on its own. But notice two things. The claim halved. And the 7% assumes the full feature suite, while a campaign upgraded from the broad match setting got one feature out of three.

The bigger issue is the word "conversions." Google measures AI Max against whatever you've set up as a conversion in your account. If that's a purchase with a revenue value attached, the claim means something. If it's a contact form submission, then 7% more conversions means 7% more form submissions, and nobody at Google knows whether any of them were worth a phone call.

What independent testing found

The largest outside dataset I can find comes from Smarter Ecommerce. Their analysis of more than 250 Search campaigns running AI Max, written up by Search Engine Journal in March, found a median lift in conversion value of 13%, close to Google's original claim. It also found a median CPA increase of 16%. The spread was wide: ROAS ranged from 42% above baseline to 35% below, and only 22% of campaigns landed near their original target.

So on average you get more, and you pay more for each unit of it. In roughly four campaigns out of five, the result was materially different from what the advertiser had before, in one direction or the other. That is what a coin toss looks like with a big sample.

Two caveats on that study, because they matter for most people reading this. It is ecommerce data. And Google has said it offers no performance expectations for retail at all, so those numbers are measured against a promise Google never made for those accounts. They tell you about variance. They don't tell you about lead generation.

The other number people love to quote comes from a test by Ezra Sackett at Monks: across roughly 30,000 search terms, 99% of AI Max impressions produced zero conversions, and fewer than half of the search terms showed a matched keyword. I'd treat the first half of that carefully. Most impressions of any match type don't convert, because an impression is someone seeing an ad, not clicking it. The second half is the telling part. For more than half of those searches, no keyword was involved. The system was deciding on its own.

Brainlabs ran the most careful test I've seen: 23 tests across 16 mature advertisers over nine months. Campaigns using all three features had a 40% higher success rate than those using search term matching alone. At the campaign level, previously untargeted queries added about 7% in conversion value. At the account level, for advertisers already heavy on broad match, the real incremental lift was closer to 3%, because 54% of the "new" queries were already being captured by other campaigns in the same account.

That last finding is the one I'd put on a sticky note. AI Max can look like it's finding new customers when it is partly taking credit for searches your other campaigns were already winning.

Where it hurts lead generation

I couldn't find a large published study of AI Max on B2B or service lead generation. What exists is small tests run by practitioners, and I'm showing you two of them because they point the same way, not because two tests prove anything.

Conner Crowe ran a 14-day split for a lead-generation service business on $5,000 a month: one campaign with AI Max, one identical manual control. AI Max produced 45 form fills against the control's 32, a 41% increase. The CRM told a different story, with 8 qualified leads against 14. Cost per qualified lead went from $178 to $313. The extra volume was early-stage, educational searches from people learning about the problem rather than shopping for a solution.

Dii Pooler tested it on a niche B2B account and got the same result from another angle. Cost per click fell 59% and clicks nearly tripled, which looks wonderful in a weekly report. Conversions fell 38%, and cost per lead went from $493 to $850. After AI Max was switched off, cost per lead stayed above $800, which Pooler calls an "algorithmic hangover." One account isn't a pattern, but it's worth knowing that switching it off may not be an instant reset.

There's a third problem that shows up in B2B specifically, where a lot of buying starts with a competitor's name. In an Optmyzr webinar in April, the panel noted that both Performance Max and AI Max heavily overindex on competitor queries, often bidding on competitor brand terms the advertiser never meant to target. Mery Hayles from Adthena put the cost plainly: "On one side you're paying higher CPCs to bid on a competitor's brand name, and on the other side you're getting your own brand CPCs inflated as well."

Put those together and you get the pattern I'd expect in a B2B account that was upgraded in September without anyone watching. Impressions up, cost per click down, form fills up, and a sales team wondering why the leads got worse.

The part the critics get wrong

It would be easy to write this post as "turn it off." I don't think that's right, and the evidence doesn't say that either.

The same Smarter Ecommerce data that shows the downside also shows campaigns beating their baseline ROAS by 42%. Brainlabs found real, if modest, incremental gains when all three features ran together. Google's own examples, like MyConnect getting 16% more leads at a 13% lower cost per action, come from Google, so weigh them accordingly, but they aren't nothing.

And the two lead-generation tests above share a weakness that says more about the accounts than the tool. In both, the thing Google was optimizing toward was a form fill. Google did exactly what it was told. It found more form fills, more cheaply. The trouble is that form fills were never the goal.

That's where I'd push back on most of the AI Max criticism. A keyword list was quietly doing a job the conversion setup should have been doing, which is filtering out people with no intent to buy. Exact match kept the researchers away, so nobody noticed the account couldn't tell a buyer from a student. Take away the keyword filter and the measurement gap becomes a targeting gap.

What to check this week

If you run Google Ads for a B2B software company, a technical product, or a local service business, this is where I'd start. None of it needs a new tool.

First, find out whether you were upgraded. Open each Search campaign's settings and look for AI Max. If it's on and nobody on your side turned it on, you were upgraded. In the search terms report, AI Max now shows up as its own match type, so you can separate what it brought in from what your keywords did.

Second, compare August to September for each upgraded campaign, and don't stop at the Google Ads columns. Pull the leads from your CRM and count how many your sales team accepted, booked, or closed. If the Google number went up and the CRM number didn't, you have your answer for that campaign.

Third, read the search terms. You're looking for competitor brand names you didn't choose to target, "what is" and "how to" searches that suggest research rather than buying, and anything that has nothing to do with you. Add them as negatives, or use the brand controls to exclude them.

Fourth, if you keep AI Max, test it properly. In August Google added the ability to run AI Max experiments with brand and location controls on, so you can split traffic between the same campaign with and without it. Run it for longer than two weeks, and judge it on qualified leads.

Fifth, and this is the one that matters most: stop telling Google a form fill is a conversion. Google's own enhanced conversions for leads documentation recommends choosing "Qualified lead" or "Converted lead" as the conversion goal. That means sending outcomes from your CRM back to Google, so the bidding learns from the leads your sales team wanted rather than everyone who typed an email address. It's more work than a checkbox. It's also the only way an automated system can learn what you actually want.

One warning if you decide to switch it off. BrandRocket's walkthrough of the upgrade notes that a single campaign-level toggle turns AI Max off entirely, but your original settings don't come back. You're left with a standard Search campaign as it stands now. Screenshot the settings before you change anything.

Where I fit

The line on my paid media page is "You're spending real money on ads. You can't say for certain what any of it is doing." September is a clean example of why. Google changed how a large number of campaigns choose who sees their ads, and for most lead-generation accounts, the only report that would show whether it helped is a CRM export nobody has connected to the ad account.

That's why I rebuild measurement first and touch campaigns second. If Google knows which leads became customers, AI Max has a fair shot at finding more of them, and I'm happy to let it try. If it only knows who filled in a form, it will find more people who fill in forms, and the dashboard will say it's working.

What I can't tell you yet is what AI Max does to B2B lead quality at scale, because nobody has published that study. The tests that exist are small, and they aren't encouraging. When I have enough client data to say more, I'll write it up, including the accounts where it worked.