Google AI Max Took Keyword Control Away. Brand Legibility Is the New Lever.
AI Max moved DTC campaign control from keyword lists to what Gemini infers your brand is. The BattlBox case shows why legibility now beats negatives.
Google's AI Max for Search reached general availability on April 15, 2026, and the complaints landing in agency Slack channels this month share one root cause: brands can no longer tell Google which searches they belong on. AI Max lets Gemini infer relevance from your pages, feed, and audiences — which means the machine now decides your brand's competitive adjacency, not your keyword list.
What actually changed with AI Max
The trade-press framing — "Google killed keywords" — is wrong, and the distinction matters for how you respond. Google's own documentation is explicit that AI Max does not remove keywords; it demotes them from directives to signals. When Google introduced AI Max at Marketing Live in 2025, it described the feature as a suite that layers query expansion, URL expansion, and audience-signal weighting on top of existing Search campaigns. After an eleven-month beta, Google brought AI Max out of beta on April 15, 2026 and put legacy tools on a deprecation calendar — Dynamic Search Ads now upgrade into AI Max, with campaign-level broad match and automatically created assets auto-upgrading in September 2026.
The practical shift is control, not keywords. Under the old model, a keyword list and a wall of negatives were directives: match this, never match that. Under AI Max, keywords are fuel for a matching model that expands into "relevant" queries you never entered and, without explicit exclusions, into brand-intent space you never approved. URL expansion compounds it — Google can route a query to whichever page it predicts will convert, not the one you designated.
The BattlBox case: when the machine picks your competitors
The clearest illustration comes from BattlBox, the roughly $30-million tactical-gear subscription brand. According to reporting in Ecommerce Times, founder Tyler Sullivan paused AI Max two weeks after enabling it because Google began serving BattlBox ads against competitor brand terms.
"We were showing up on searches for brands we have no business appearing next to," Sullivan told the outlet. "It wasn't a bidding issue — Google just decided we were semantically relevant. That's not a signal we want to send to the market."
Read that carefully, because it is the whole story in one sentence. This was not a targeting mistake a negative keyword would fix. The model formed its own theory of what BattlBox is adjacent to, and acted on it. The same report quotes Calla Murphy of DTC agency Belardi Wong describing accounts that had accumulated 4,000 negative keywords over three years — an entire control apparatus that AI Max renders partial overnight.
The pain is not uniform. Commoditized categories with diffuse intent — accessories, apparel basics — report acceptable performance because broad signal is exactly what the model optimizes well. The damage concentrates in specialty and premium brands, where purchase intent is specific and adjacency is identity. When a luxury brand's ad copy is auto-generated from landing-page text, or a specialty brand is placed beside rivals, the cost is not just wasted spend. It is positioning.
Control moved from keywords to legibility
Step back and this looks less like a paid-search update and more like a preview of every discovery surface. In organic AI answers, brands already learned they cannot dictate the queries they surface on — ChatGPT, Perplexity, and Google's AI Overviews infer relevance from how the wider web describes an entity. AI Max brings that same logic into the one channel operators thought they still controlled deterministically: paid search.
This is the shift Jaxon Parrott named Machine Relations — the discipline of making a brand legible to the machines that mediate discovery. The unit of control is no longer the keyword; it is the machine's model of what your brand is and is not. That is a question of entity optimization: the clarity of the signals — landing pages, product feeds, structured data, and the corpus of third-party sources describing you — that let a model place your brand correctly. When those signals are clean, the model's inference works for you. When they are ambiguous, it fills the gap with its own guess, and BattlBox is what that guess looks like.
The levers that actually govern AI Max reflect this. They are not match types. They are brand keyword exclusions that fence off your name and your rivals', first-party audience signals that teach the model who your buyer is, URL inclusion lists that constrain where traffic can land, and text guidelines that bound what the system can say. Every one of them is an act of defining the brand for a machine rather than instructing it.
| Control layer | Keyword-era model | AI Max model |
|---|---|---|
| Query targeting | Explicit keyword lists | Model-inferred from page, feed, audience |
| Blocking | Negative keyword lists | Brand exclusions on the intent space |
| Landing page | Advertiser-designated | URL expansion, model-selected (needs inclusion list) |
| Brand voice | Advertiser-written copy | Auto-generated assets, bounded by text guidelines |
| Primary skill | Campaign structuring | Signal and entity legibility |
What CMOs should do before flipping the switch
Google is aware of the adjacency problem. As of late May 2026, Search Engine Land reported that Google is testing branded search controls in a limited beta — letting advertisers define which brand-related queries a campaign may target. It is a signal that the control gap is real enough for Google to patch. But the timeline is unannounced, and the September auto-upgrade is not waiting for it.
The practical sequence, drawn from how careful operators are staging migrations: set brand exclusions across every non-brand campaign and verify them in the search-terms report before enabling anything; accumulate first-party audiences so the model has real buyer signal rather than raw intent; constrain URL expansion to approved paid landing pages; and only then flip AI Max on, reviewing query expansion weekly for the first month. The brands that survive the migration, in Murphy's framing, are the ones that feed the machine the right signals before they hand it the wheel — not after.
That is the durable takeaway. The keyword playbook is not coming back. The winning capability is AI visibility discipline applied to paid: knowing what your brand is legible as across every machine that now mediates discovery, and closing the gaps before an algorithm fills them for you.
Brands that want to see how machines currently read their entity — across paid and organic AI surfaces — can start with a visibility audit to map the signals a model would infer from today.
FAQ
Does Google AI Max eliminate keywords?
No. Google's documentation describes keywords as signals that fuel AI Max's matching model, not targets it obeys. Exact and phrase match campaigns still function. What changes is that the model expands beyond your keywords into queries it deems relevant, so keywords stop being hard directives.
Why did BattlBox pause AI Max?
Per Ecommerce Times, BattlBox paused two weeks in because AI Max served its ads against competitor brand terms. Founder Tyler Sullivan framed it as a positioning risk, not a bidding problem — Google's model decided the brand was semantically relevant to rivals it had never targeted.
How do you control brand adjacency under AI Max?
Through brand keyword exclusions, first-party audience signals, URL inclusion lists, and text guidelines — plus the branded search controls Google is testing in limited beta. These define what the brand is and is not for the model, rather than listing individual queries to match or block.