Google says advertisers activating AI Max see 14% more conversions or conversion value at a similar CPA or ROAS, rising to 27% for campaigns built primarily on exact and phrase keywords. Those are the numbers that got the feature adopted. After running it across accounts for a full quarter, my view is that the headline understates one part of AI Max and badly oversells another, and the difference between those two is worth more than the average.
AI Max is not one feature. It is three switches bundled under one label: search term matching, text customisation, and final URL expansion. Google’s own framing makes the bundling explicit, reporting an average 7% more conversions at similar CPA or ROAS when the full suite runs compared with search term matching alone. Read that number carefully, because it is the most useful one in the whole announcement. It means the majority of the reported gain comes from the matching change, and the other two switches split a much smaller remainder.
Search term matching is the part that earns its place
The 27% figure applies to campaigns where more than 70% of conversions come from exact or phrase match keywords. That scoping is not a caveat, it is the whole finding. A tightly-keyworded account is, by construction, an account that has been systematically refusing traffic for years. Every negative list and every exact-match decision was a bet about which queries were worth buying, made with the information available at the time.
Google reports MyConnect seeing a 30% increase in conversions from net-new queries.
That is the mechanism, stated plainly. The lift is not the machine bidding more cleverly on queries an account already had. It is the machine buying queries the account had never seen. For a mature account, most of that inventory was excluded by decisions nobody has revisited since 2022.
Text customisation is a brand problem, not a performance one
The middle switch rewrites headlines and descriptions to fit the query. In performance terms it is the least interesting of the three, which the 7% full-suite figure implies. In review terms it is the one that generates the most work. Auto-written copy will eventually produce a claim the legal team has spent years keeping off the site, in a regulated vertical it will produce one that is genuinely non-compliant, and it will do it in an ad that ran for a week before anyone noticed. The mitigation is unglamorous: pin the headlines that carry claims, leave the rest open, and read the served combinations weekly for the first month rather than trusting the aggregate.
Final URL expansion is where I part company
The third switch lets Google choose a different landing page than the one specified. In the named case studies the results look excellent: Google reports L’Oréal at a 2X higher conversion rate and 31% lower cost-per-conversion, and MyConnect at 16% more leads with a 13% lower cost-per-action.
On one lead-gen account I run, the problem was not that expansion performed badly. It performed fine on the platform’s own numbers. The problem was that it started routing paid traffic to pages that were never built to convert it, including a careers page and an old campaign landing page that still existed only because nobody had deleted it. Conversions were recorded because the form on those pages fires the same event. The leads were unusable.
That is the failure mode nobody screenshots. Expansion optimises toward the conversion signal it is given, and if that signal is defined loosely, it will find the cheapest path to firing it. The fix is not to distrust the feature. It is to make sure the conversion action means what the business thinks it means before handing over page selection, which for most accounts is a bigger project than switching AI Max on.
The migration is the actual story
AI Max is out of beta with more than 500,000 advertisers on it, and Google has been steadily moving the deadline for automatic migration of Dynamic Search Ads, pushing it from September 2026 to February 2027. A delay of that size, on an automatic upgrade, tends to mean the upgrade was not going smoothly for a meaningful minority of accounts.
The practical read is that this is not optional in the long run. The choice is whether to migrate deliberately, with the conversion signal cleaned up first, or to be migrated on Google’s schedule with whatever configuration happens to be live that week.
What I would turn on, in order
- Search term matching first, on the campaigns that are keyword-heavy and have been running long enough to have accumulated a large negative list. That is where the reported 27% lives.
- Text customisation second, once there is enough creative variety for it to have something to work with.
- Final URL expansion last, and only after auditing which pages can fire the conversion action, with the page exclusions configured before it goes live rather than after.
Where this could be wrong
Every figure above is Google’s, published to encourage adoption of a Google product, and the case studies are named advertisers selected by the vendor. No independent measurement of AI Max exists at any scale, and the honest position is that nobody outside Google can currently run the counterfactual.
There is also a selection problem in my own experience. The accounts I saw were ones where the conversion signal was already suspect, which is exactly the population where expansion misbehaves. An account with a clean, deduplicated, business-meaningful conversion action might get the case-study result and nothing else. If that is the common case, my caution is overfitted to a sample of broken measurement, and I would rather be told that than keep repeating it.
What I am confident about is narrower: the gain is concentrated in the matching change, the scoping on the 27% is doing real work, and a feature that selects landing pages should never be enabled before somebody has checked what those pages count as a conversion.