AI Max for Search changes the deal on negative keywords, and the change is easy to miss because the controls look the same. AI Max takes your existing keywords and expands them — generating query variations, widening match, and reaching new searches through URL expansion off your landing pages. The setup guides describe a system that matches on real-time intent rather than your selected keywords. The problem for anyone who manages waste for a living: your negative keywords were written against your keywords, and AI Max is deliberately reaching past them.
This is the AI Max sibling to the Performance Max negatives problem, but on a different surface with different mechanics. It matters now because the deadline pressure is real, if not quite what the headlines say. Below is what actually happens to negatives when AI Max expands a query, what changes in September 2026 versus February 2027, and the practitioner routine that keeps spend contained through the migration.
The deadline, stated accurately
Get the dates right first, because a lot of the commentary is stale. Google delayed the full DSA-to-AI-Max automigration from September 2026 to February 2027, confirmed by the Google Ads Liaison on 11 June 2026, citing advertiser feedback and a desire not to disrupt Q4. So the "DSA dies in September" framing is wrong. What genuinely happens in September 2026 is narrower: you can no longer create new DSA campaigns through the UI, Editor or API, and automatically created assets plus campaign-level broad match settings begin auto-upgrading.
The distinction matters for how you spend the runway. You are not being forced onto AI Max in September; existing DSA campaigns keep running until the February 2027 migration. That gives you roughly half a year to build and test the negative controls AI Max needs before it becomes the only option, rather than discovering the expansion behaviour under fire. Treat September as the moment new DSA creation closes and the auto-upgrade of assets begins, and February 2027 as the hard migration — and use the gap to get your negative lists AI-Max-ready on a campaign you migrate deliberately.
How a negative gets overridden in practice
The override is not AI Max deleting your negatives — it is expansion reaching a query your negatives did not anticipate. The mechanism is URL expansion: AI Max reads your landing pages and serves on queries it judges relevant to that page content, independent of your keyword list. When an expanded page looks like a strong match for a query, the system can serve there even if a keyword-level negative would have blocked the same query in a classic ad group. The Groas analysis frames this as the conflict between negatives and what AI Max is trying to target through URL expansion.
The important nuance is what you keep: campaign-level negative keywords still apply, and you retain full search terms report visibility. So this is not the old Performance Max situation where you negate partly blind. You can see the queries AI Max reached, and a campaign-level negative on a term does hold. The failure mode is one of coverage, not authority — expansion surfaces queries your list never contemplated, and they spend until you notice them in the report and add them. That is a manageable problem, but only if you review the report often enough to catch expansion early rather than after a month of drift.
Lean on campaign-level lists, not ad-group structure
In a classic Search campaign, tight ad-group structure did a lot of the containment: narrow keywords in narrow ad groups meant intent stayed roughly where you put it, and keyword-level negatives handled the edges. AI Max breaks that model on purpose, so structure can no longer be your primary containment tool. The lever that still works is the campaign-level and account-level negative list, which applies across the expansion the same way it applies to your keywords.
That means the shared negative lists you maintain become more important, not less, under AI Max. Build robust thematic lists — the job and salary words, the free and DIY words, the wrong-product and wrong-geography words, the competitor and research-intent words that never convert for you — and apply them at the campaign level so they cover expanded queries too. This is the same thematic discipline the wasted-spend work already teaches; AI Max raises the stakes because the surface those lists have to cover is wider than your literal keywords. A thin negative list that was survivable with tight structure becomes expensive when expansion is doing the matching.
Use the search terms report AI Max still gives you
The single biggest advantage AI Max has over Performance Max for waste control is that it keeps full search terms report visibility. Every query expansion reaches is auditable, which means the containment loop is the familiar one: read the report, find the wasteful themes, add campaign-level negatives, repeat. You are not fighting an opaque black box; you are managing a wider matching surface with the same instrument you always used. The mistake is assuming AI Max is as blind as PMax and under-investing in the report as a result.
Read that report against outcomes, not just wording — a caution that matters doubly now that Google may show an AI-interpreted version of the query rather than the literal text on some surfaces. So AI Max hands you a wider surface to police and a report that is simultaneously less literal. The response is the same either way: judge themes by spend and conversions, negate the concept rather than the exact string, and confirm against landing-page and first-party signals before blocking at scale. The visibility is real and worth using; it just is not a promise that every row is a verbatim query.
A containment routine for the migration window
Turn all of this into a schedule you can run before AI Max is mandatory. When you enable it on a test campaign, check the search terms report daily for the first week — this is when expansion reaches its widest and least-filtered, and catching irrelevant queries early stops them compounding. Feed the waste into campaign-level negatives as you find it. After the two- to four-week learning period, drop to a weekly thematic pass: n-gram the report, target high-cost zero-conversion words with phrase negatives, and keep the shared lists current.
Keep the written log every good negative workflow uses — the term, the level, the date, and why — because AI Max expansion and interpreted reporting together make after-the-fact reconstruction hard, and a campaign-level negative that goes too far can quietly cut volume across the whole campaign. Use the migration window deliberately: migrate one campaign, build the negative list that contains it, prove the routine, and only then move the rest before February 2027. The account that comes through the DSA sunset in good shape is the one that treated AI Max not as a black box to fear but as a wider matching surface to police with thematic negatives and a report it actually reads. For the underlying method, start with the search terms report walkthrough.