The search terms report you have read for years now has a section that is not really yours. Since AI Max reached general availability on 15 April 2026, Google Ads matches queries using broad match, your assets, and your landing page content — not just the keywords you added — and the report has been updated to tag those AI-Max-generated rows and add a source column that tells you why the ad matched. The queries are real, the spend is real, and a growing share of them were chosen by Google rather than by you. This post is about reading that new segment: how to isolate the AI Max rows, judge them, and cut the waste without throttling the expansion you turned the feature on to get.
This matters now because the query surface is shifting under most accounts at once. Google is upgrading Dynamic Search Ads and legacy broad-match settings into AI Max through the September 2026 window, so advertisers who never opted in are about to see AI Max rows appear in a report they thought they understood (practicalecommerce on the search terms report update). If you treat those rows the same way you treat keyword-matched traffic, you will misjudge both the waste and the wins. The segmentation exists precisely so you do not have to.
What the AI Max segment actually is
AI Max is intent-based matching that expands beyond your keyword list using three signals: broad match, asset-based matching, and landing-page-based matching. When any of those catch a query, the resulting search term is attributed to AI Max rather than to one of your keywords. In the report that shows up as a row whose match is credited to AI Max’s technology, and the new source column names which of the three signals did the catching. That is the whole idea of the segment: it separates the traffic your keywords earned from the traffic Google’s inference generated, so you can hold each to a different standard.
The distinction is not cosmetic. A query matched by a keyword you chose reflects a decision you made; a query matched by your landing page content reflects a decision Google made about what your page is about. Those deserve different scrutiny. Google has also added an AI Max search term combination report that shows how specific ad combinations and landing pages performed against AI-Max-matched searches, so the reporting is moving toward showing not just the query but the machinery that served it. For a practitioner, the practical takeaway is simple: the AI Max rows are where the account is being run on autopilot, and autopilot is exactly what you audit first.
The source column: why your ad matched
The single most useful addition is the source column, which tells you the signal behind each match rather than leaving you to guess. A query caught by broad match, a query caught by one of your assets, and a query caught by your landing page are three different kinds of expansion, and they fail in different ways. Broad-match expansion tends to drift on theme; asset-based matching can pull in queries that echo your ad copy more than your offer; landing-page matching can surface queries about anything your page happens to mention, including boilerplate. Reading the source tells you which failure mode you are looking at before you decide what to do.
In practice you sort or filter the report on that column to get a clean list of AI Max expansion queries, then read down it the way you would read any search terms list — except that here the relevance bar is yours to enforce, because nothing on this list was a keyword you vetted. This is the same skill as reading the classic report, covered in the pillar on how to read the search terms report and turn it into negative keywords, applied to a surface where Google, not you, decided the query was a fit. The source column is what makes that judgement tractable at scale.
Spotting the wasted rows
The waste in AI Max rows clusters into two shapes: off-topic theme drift and research-only intent. Theme drift is a query that shares a word or a concept with your offer but not the intent — the expansion reached sideways into an adjacent topic that does not buy. Research-only intent is a query like a comparison, a definition, or a “how does it work” that signals someone learning, not someone buying; it can be perfectly on-topic and still convert at nothing. Both are easy to spot once the rows are isolated, because they stand out against the keyword-matched traffic that has a proven relationship to your conversions.
Sort the AI Max rows by cost descending and read the top of the list first, exactly as you would for the account-wide report — the expensive zero-conversion expansion is where the money is actually leaking. Watch for the same trap the top-cost view hides everywhere: a swarm of small, cheap, individually-ignorable queries on the same off-topic theme can outweigh any single expensive row, which is why n-gram thinking matters here too. The mechanics of finding those themes are in the wasted spend the top-cost view hides and n-gram search-term analysis; the AI Max segment just gives you a pre-filtered place to point them.
Negating without throttling reach
Here is where the AI Max segment demands a different reflex than the classic report. With keyword-matched traffic, an off-topic query is almost always safe to negate. With AI Max expansion, the whole value proposition is the incremental query your keyword list would never have found — so a heavy negative habit does not just trim waste, it undoes the feature. Google’s own guidance is to use negatives sparingly with intent-based matching and add them only when a term consistently underperforms, because over-blocking starves the expansion of the ambiguous queries that sometimes turn out to convert.
The discipline that fits: negate the clearly-off-topic and clearly-research-only rows now, and let the genuinely ambiguous expansion accumulate data before you judge it. Then verify. Because AI Max matches more expansively and can reinterpret intent, a query you blocked can be re-reached through a different signal, and a negative can conflict with the matching in ways that are not obvious — the failure modes are covered in when AI Max overrides your negative keywords. Add the negative, then re-check the report a few days later to confirm the query actually stopped, rather than assuming the block held the way it would in a plain Search campaign.
A weekly routine for the AI Max rows
Turn this into a standing pass rather than a one-off. Once a week, filter the search terms report to the AI Max source rows, sort by cost, and read the top of the list plus any new theme that has appeared since last week. Negate the off-topic and research-only queries, note the ambiguous ones to revisit, and leave the converting expansion alone. During the September 2026 auto-upgrade window run this more often — twice a week or even daily on high-spend accounts — because that is when the newest expansion rows arrive fastest and the biggest one-time cleanup is available.
The mindset that keeps this from becoming either negligence or over-control is to treat the AI Max segment as a probation list, not a blocklist. Every row on it is traffic Google chose on your behalf; your job is to confirm or overrule that choice with evidence, quickly for the obvious cases and patiently for the ambiguous ones. If you are handling the broader migration, this pass slots into the wider AI Max migration search-terms checklist; the segment reading here is the part you keep doing every week after the migration itself is done.