AI Max for Search: Why It Paid for Our Lead Gen Client When It Fails for Most

Your search campaigns probably got moved to AI Max in September whether you asked for it or not, and almost everything published about it says lead generation should switch it off. HBT Digital ran it next to manual keywords for 14 months on a home services account. The manual keywords produced 36 conversions over that period. AI Max matched 993 search terms, and 964 of them never earned a single click. AI Max produced half a conversion. Monks found 99% of roughly 30,000 AI Max search terms never converted.

We run AI Max for a US financial advisor matching platform, where a qualified lead is someone with $250,000 or more to invest. Our first week looked a lot like HBT's. By the end of the first month it had broken even. Four months in, we are doubling it.

What AI Max actually does to a Search campaign

Google announced AI Max for Search on May 6, 2025. It is a layer over a normal Search campaign rather than a new campaign type, and it does three things:

• Search term matching reaches queries your keywords would never have matched, using broad match and

keywordless matching off your ads and pages.

• Text customization writes headline and description variants from your landing page and assets.
• Final URL expansion sends a visitor to whichever page on your site Google thinks fits the query better

than the one you set.

The first one is easiest to understand from a single search. Someone looking to hire a certified financial planner typed a query that matched none of our keywords, so under the old setup we would never have seen them; AI Max read one of our pages on choosing an advisor, decided the two fit, and showed the ad. That visitor became an approved lead with more than $1 million to invest.

Google's launch numbers were 14% more conversions at a similar cost per acquisition, and 27% for campaigns that still lived mostly on exact and phrase match. On April 15, 2026, it said accounts using all three features see 7% more conversions than those using search term matching alone, and that campaigns on automatically created assets or campaign-level broad match would be upgraded automatically from September 2026. Dynamic Search Ads were on the same September schedule until a June 11 update pushed their move back to February 2027. Every one of those figures comes from Google's own internal data, and none has been reproduced independently.

We tried it the obvious way first

This client's Search account had run for years and was not broken: leads converted at 4.5% of clicks, and the client reached about 45% of the people who filled the form. We wanted to reach the people searching around it, the ones whose question did not look like any keyword we bid on.

Performance Max was the first attempt. Over six weeks it returned about 55 cents for every dollar, and much of the spend went to YouTube and mobile placements rather than to people searching for an advisor.

Then we launched AI Max on Google's defaults. The first week earned nothing. More than half of its clicks came from people trying to log into a retirement account they already had. The search terms report read like a payroll help desk: "401k login", "my plan john hancock", "walmart 401k", "fedex retirement", "find my old 401k", "lost and found dol gov". Google had read "401(k)" on our pages and taken it literally.

The leads that did come through were no better. Every person who said they had $250,000 or more was rejected. Some gave contact details that did not work, some did not match the buyer's requirements, and one wanted help with crypto and a trust. The few that were approved had small accounts and paid next to nothing. My note to the team was "Something is wrong." When the rejections came back, it was two words: "basically rejected."

My first suspicion was the conversion tracking. GA4 showed it was working perfectly. The problem was who AI Max had decided to talk to. The fix was teaching it who not to talk to. We built a dedicated negative list, and within days the login searches had fallen to almost nothing, while conversions per day rose about sevenfold. That was the first sign the campaign might be worth more money, not less.

About two weeks after launch, our PPC lead tripled the daily budget, before the campaign had earned any real revenue. The reason was quality. The leads already coming in were getting approved, and that was worth buying more of. The dashboard showed there was room to buy them: the campaign was losing between 76% and 88% of its eligible impressions to budget every day, so the demand was there and the old budget could not reach it, however good the leads were. The next day brought a run of approved leads, each with $250,000 or more. By the end of the first month, revenue per approved request covered what each one cost to buy.

Phase 1: 197 pages in seven days

Underneath the campaign sit 197 landing pages, built in seven days on six templates. Three of those days went on design and on the parts of each page that help someone who has not decided anything yet: checklists, slider calculators, and cited sources. The pages were built and researched by AI agents, which also tested every source for credibility, and then people reviewed every page before launch, because a page about someone's retirement money is not a place to trust a first draft.

Each page is written for one context a person might be searching from: a kind of advisor, a credential, a life event, an account they already hold. Every one is an advertorial, written so that a person who arrives with a narrow question about their own money gets a useful answer whether or not they ever fill in the form.

We drew the first intent clusters ourselves and mapped one ad group to each. Google has taken it much further. Final URL expansion is on, and of the 162 pages that have received clicks since launch, Google chose 148 of them itself, mostly state-specific and life-event versions of the core pages. We also let it rewrite headlines for those pages, so the ad matches whichever page it picks.

The conversion rate went down, and that was the good news:

Fewer people asked. Those who did were far more often the investors the client wanted, people with $250,000 or more to place who were also more likely to pick up when the client's team called them back.

What a week looks like now

We group the landing pages by what they are about, because that turned out to matter more than any single page. The first two columns come from one full week, Monday to Sunday. The last two cover every lead since launch, with each person counted once. Revenue ÷ spend is what the client earned for each dollar of media.

None of the losing pages lost because their visitors were bored. The product pages sent more of their visitors to the form than the account average, and those visitors answered every question honestly. Most of them reported between $50,000 and $100,000, which is real money and not what an advisor buys. Many had been searching for an employer plan they wanted to get into, not for someone to manage it.

The life-event page drew people who said they had the money, and almost none were approved. We have not confirmed what the buyers meant, but the search terms suggest it: these were people who needed their assets split once, not an advisor for the next twenty years.

That page was paused for a different reason, though. One request came in with a message that had nothing to do with investing. It described a personal situation that needed a different kind of help than a financial advisor gives. The client read it and decided this was a conversation they were not set up to have and did not want their advisors walking into. We paused the ad group that week. No dashboard would have flagged it. We think the client was right.

It was the same campaign. Budget, bidding and targeting matched too. The page was the only thing that differed. AI Max decides what to match by reading the page, so a page about a retirement account draws people who already have one, and a page about hiring an advisor draws people already shopping for one. The samples are small, and one advisor page lagged well behind the others, so we treat this as the strongest pattern we have seen rather than a rule.

One page earns without any ads at all. The parent page the intent pages sit under has no ad group pointing at it, and it still earned about 8% of the week's revenue. A campaign reporting view never shows you this, because that page has no cost against it.

Phase 2: 400 pages and a bigger campaign

Four months in, we have moved from proving it to scaling it: Phase 2 takes the site past 400 landing pages and widens the campaign to every context we drew up at the start, plus the ones Google's matching has turned up since. Most of the new pages are about choosing an advisor, for the reason in the table above.

We think AI Max marks the end of the keyword as the unit of paid search work. For twenty years, paid search meant choosing which words to pay for and arguing about match types; Google now does that part itself, does it well, and will keep taking on more of it whether or not advertisers ask it to. What is left for the people running the account is deciding who the campaign should not talk to, which situations deserve a page at all, and when a lead that looks good on a dashboard is one the client should never have received.

Most of what is published about AI Max tells you to steer clear of it. We read those same results differently. The accounts that failed switched it on and left it alone, and AI Max does not run itself: someone has to tell it who not to talk to, build the pages it chooses from, and judge it on revenue instead of clicks. That is a job for people who do it every week. If you want that kind of help with your Google Ads, we can help. Talk to our paid search team.

We do this for clients as part of our paid search management work, with the page builds handled by our landing page optimization team. The part most accounts are missing is the tracking that carries a qualified lead back into bidding, which our conversion tracking setup covers. For the wider picture of where automated buying is heading, see our piece on AI in digital advertising.