AI Max shipped a report that changes what a wasted impression can tell you. The Search terms and ad combinations report for AI Max for Search campaigns puts the search term, the headline and description that were served, and the landing page the user reached into a single row. For years the search terms report answered one question — which queries triggered my ads — and a wasted row had one fix, a negative keyword. This report answers three questions at once, and in doing so it reveals that a bad row no longer has one fix. The query might be wrong, the creative might be wrong, or the destination might be wrong, and those are three different problems with three different controls.
That is the whole argument of this post: before you negate anything in an AI Max campaign, read the combination, not just the query. A negative keyword is the correct tool for a wrong query and the wrong tool for a right query that was mishandled after the click. The combinations report is what finally lets you tell those cases apart, because it stops hiding the two variables — creative and landing page — that a query-only report left you to guess at. Get the diagnosis right and you stop throwing away relevant traffic to fix problems a negative was never going to solve.
What the report actually joins
The report's value is in the join. Each row carries the search term, the headline and description shown against it, and the landing page the user was sent to, so you see the full combination AI Max assembled rather than the query in isolation. In the Google Ads API the same data surfaces as the ai_max_search_term_ad_combination_view, which, as one breakdown put it, “joins the search term, the landing page and the headline for each combination, which is the level at which AI Max actually makes decisions.” That last clause is the point. AI Max decides at the combination level, so to audit it you have to see at the combination level.
Two more columns carry the diagnosis. A match-source column tells you whether the row came from AI Max expanded matching or from a landing-page and URL inclusion, and an asset-source column tells you whether the headline and description were ones you supplied or ones Google generated. Together with the query and the landing page, those four facts are enough to route almost any bad row to the right control. This is a richer surface than the standard AI Max match-type and source columns covered in the AI Max source column post — that one tells you how a term was matched; this one tells you what was served once it did.
A wasted impression now has three possible fixes
The reason this matters for negatives is that the three columns correspond to three separate controls, and only one of them is a negative keyword. A wrong query — a search that should never have triggered your ad regardless of what you showed it — is a negative keyword, exactly as it always was. A right query sent to the wrong landing page is a URL exclusion: the problem is the destination, and Google provides URL exclusions precisely so you can stop a page from serving as a landing page without touching the query that reached it. A right query served the wrong creative is an asset problem, fixed by tightening or pinning your own headlines so the automation has less room to assemble something misleading.
Collapsing all three into “add a negative” is the mistake the old report encouraged because it was the only lever it showed you. When your only visible control is a negative keyword, every bad row looks like a query problem, and you end up negating relevant searches to fix landing-page and creative issues the negative cannot address. The combinations report breaks that reflex by making the other two problems visible in the same row. Google's own guidance in the AI Max reporting documentation is to review the search terms alongside their headlines, descriptions and landing pages and then refine negative keywords, ad assets and campaign settings — three outputs, not one.
Read the query first, then the destination, then the creative
Work each bad row in a fixed order so you never reach for the wrong lever. Start with the query in isolation: ask whether this search should ever trigger your ad, ignoring everything else in the row. If the answer is no — wrong product, informational intent, a free-version hunter, a job seeker — it is a negative keyword, and you are done; the creative and landing page are irrelevant because you are removing the query entirely. This first pass catches the genuinely irrelevant traffic, and for that traffic nothing about the new report changes your decision.
Only when the query survives that test do the other two columns come into play. If the query is relevant but converted badly, look at the landing page: was the user sent somewhere that does not match what they searched? If so, that is a URL exclusion or a fix to your URL inclusions, not a negative — negating it would remove a query you want. If the landing page was right too, look at the asset source: a relevant query sent to the right page but served a Google-generated headline that misstates the offer is a creative mismatch, fixed by supplying or pinning better assets. The order matters because it stops you from blaming the creative for a query that was wrong all along, or negating a query for a problem that lived in the headline.
Let combinations settle before you act
AI Max assembles these combinations dynamically, which means any single row is a snapshot of one pairing the system tried, not a fixed fact about the query. A relevant query that converted badly against one generated headline may do fine against the next one the automation tests, with no action from you. So the significance discipline you already apply to search-term negatives applies here too, and arguably harder: wait until a combination has accumulated enough spend or impressions to be a signal, and still underperforms, before you intervene. The reasoning is the same as how many clicks before negating a search term, just with an extra variable in play.
There is a specific trap the settling rule guards against. Because the report now shows creative and landing page, it is tempting to treat a bad combination as proof the query is bad and negate it on the spot. But the combination's performance is a joint result of three moving parts, and attributing a poor result to the query when the headline was the problem means you lose the query and never learn that a better asset would have fixed it. Give the automation room to try other combinations for a relevant query before you conclude the query itself is the fault, and reserve the immediate negative for queries that are wrong on their face regardless of what was served.
Fit it into the AI Max audit routine
This report does not replace your negative-keyword work; it sharpens where that work applies. Most of the AI Max hygiene routine is unchanged — you still reconcile what the automation reports against what it actually did, which is the subject of AI Max attribution discrepancies, and you still run the broader migration checklist when a campaign moves onto AI Max. What the combinations report adds is a triage step inside the search-terms pass: for every row you would previously have negated on sight, check whether the query was actually the problem or whether a URL exclusion or asset fix is the real lever. That one habit keeps relevant traffic from being collateral damage in a negatives pass.
Treat the report as a periodic diagnostic rather than a daily chore. Pull it when you are investigating high-spend queries or a conversion drop, sort by cost, and work the expensive combinations top-down with the query-first, destination-second, creative-third order above. For the end-to-end context this sits inside, see the search hygiene audit playbook, and for the workflow of getting AI Max and Performance Max search-term data out of the interface in the first place, see the Editor 2.11 search-terms workflow.
Related reading
For how AI Max labels the way a term was matched — the column that tells you broad-match expansion from keywordless matching — see the AI Max source column. For reconciling AI Max's reported numbers with what it actually served, see AI Max attribution discrepancies, and for the full routine this triage step belongs to, see the search hygiene audit playbook.