Consider the following search query: “I need a hotel in Soho with connecting rooms, featuring a Tesla charger, dog-friendly, and located within walking distance of a specific wedding venue.”
This is not a search. It is a brief. The user is not looking for a list of options to evaluate. They have already mapped their constraints and are directing an AI system to solve a specific logistical problem. They will not scroll through results. They will act on the first relevant answer the algorithm delivers.
This is the guest who booked your competitor last weekend.
Traditional keyword strategies cannot reach this person. A campaign targeting “Hotel Soho” or even “luxury hotel Soho dog friendly” does not penetrate the semantic layers of a query this specific. Google's answer to this gap is AI Max for Search: a campaign type that reads the meaning of a prompt, not its surface keywords, and matches ads to intent rather than terminology.
The promise is real. The risks are equally real.
What AI Max Changes Fundamentally
The shift is architectural. Legacy search campaigns operated on deterministic logic: if a user types this keyword, show this ad. AI Max operates on probabilistic logic: if a user demonstrates a particular intent, the algorithm autonomously chooses which ad to show, which creative to use, and which landing page to send them to.
Every element of the campaign, including the headline, the landing page, and the audience match, becomes a system output rather than a human input. The marketer no longer controls the execution. They control the conditions under which the execution happens.
This distinction is more important than most hotel marketing teams have yet absorbed.
Three Risks That No One Warns You About
Without guardrails, the mechanisms designed to expand AI Max’s reach become the source of its most damaging failures. Three deserve specific attention.
Broad Match Bleed
AI Max scans your website to interpret the semantic meaning of a query and determine whether your property is a relevant match. The problem is that semantic proximity is not the same as commercial intent. For a luxury hotel, the algorithm may register a short semantic distance between your property and a query for “cheap hostel near Soho” because your website mentions Soho, mentions rooms, and references promotional offers in a page from 2021. The ad appears. The click costs the same. The conversion probability is near zero. This budget erosion is gradual, invisible in aggregate metrics, and entirely preventable.
The Zombie Page
AI Max’s URL Expansion feature allows the algorithm to bypass your designated landing page and send traffic to any page on your website it deems more relevant to the query. In practice, a guest searching for a suite for a wedding anniversary may land on a 2019 blog post about your Christmas dinner menu, because that page contains a high density of terms the AI associated with the query. The booking engine is never reached. The campaign records the click. The attribution chain is broken.
Brand Hallucinations
AI Max generates headlines in real time based on the signals it detects in the query. If it registers price sensitivity in the user’s prompt, it may generate a headline emphasizing value or affordability, regardless of where your property sits in the market. A five-star hotel can find itself serving headlines that read like a budget brand, not because a human made that decision, but because the algorithm optimized for click probability rather than brand alignment. The guest registers an impression of the property that has nothing to do with what the Brand actually represents.
The Swimlane Architecture
At Influence Society, we reject the binary framing of “full AI” or “no AI.” AI Max belongs in specific contexts where its strengths operate without threatening the assets that are hardest to rebuild: brand equity and high-intent traffic. The framework we have developed for our clients is what we call the Swimlane Architecture. It assigns each campaign type to a defined lane, with explicit boundaries that prevent the lanes from bleeding into each other.
Lane 1 — The Brand Fortress
AI Max is excluded entirely. We use Standard Search campaigns with Exact and Phrase Match for branded queries: the property name, booking-intent variations, any term where the user has already identified the hotel. These guests have already made their choice. Handing them to an AI system that may generate off-brand messaging or redirect them to a non-commercial page is an unnecessary risk on the highest-value traffic in the account.
Lane 2 — The Net
This is where AI Max operates. Its mandate is discovery: capturing the complex, conversational, keyword-less queries from users who have not yet identified a specific property. The Soho Tesla charger query lives here. The “romantic coastal retreat for two non-swimmers with a spa and a Michelin-starred restaurant within walking distance” query lives here. AI Max is given latitude to match, create, and expand, within the exclusions established in the operational defense layer.
Lane 3 — The Conversion Anchor
Performance Max for Travel Goals, running against the Google Hotel Center feed with real-time rates and availability, handles retargeting and final conversion. It acts as the closing mechanism for users who entered the funnel through Lane 2 but have not yet booked.
The three lanes work in sequence. Lane 1 protects what is already yours. Lane 2 expands the pool of potential guests. Lane 3 converts them.
Operational Defense
The Swimlane Architecture only holds if the exclusion mechanisms are maintained with the same discipline as the campaigns themselves.
Negative keywords are not a refinement in the AI Max context. They are a structural requirement. An adults-only property that does not aggressively exclude “family,” “children,” “baby,” and equivalent terms will find its Lane 2 budget redirected toward entirely irrelevant traffic by an algorithm that perceives semantic proximity where there is no commercial alignment.
URL exclusion protocols are equally non-negotiable. Every non-commercial subdirectory, /blog, /press, /careers, /events, must be excluded from the AI’s index before the campaign launches. The Zombie Page risk is not hypothetical. It is a default behavior of URL Expansion that must be actively prevented.
Signal Quality
The Swimlane Architecture controls where AI Max operates. Signal management controls how well it performs within those lanes.
Standard AI signals, browsing history and demographic patterns, optimize for volume. For luxury hospitality, volume is the wrong objective. Left to its default signals, AI Max will learn to find guests likely to click and book a single night at the lowest available rate, because that behavior is statistically common. It will ignore the guest profile that actually drives revenue.
The correction is to upload hashed first-party data lists directly into the campaign: Suite Bookers, Repeat Guests, High-Spend Ancillary customers. This teaches the algorithm to weight its probabilistic matching toward the behavioral patterns that correlate with genuine value, not just conversion volume.
Value-Based Bidding completes this correction. By connecting the Property Management System to Google Ads and feeding back actual stay data, the algorithm learns to differentiate between a booking that cancels three days later and a stay that includes spa spend, restaurant covers, and a return the following season. A canceled booking is worth zero. A high-ancillary stay is worth multiples of the room rate. When the algorithm understands this distinction, it stops being a volume engine and starts operating as a value engine.
The Agentic Horizon
The Swimlane Architecture is relevant today. It becomes more relevant as the booking landscape shifts toward Agentic Travel, where systems like Google Canvas generate itineraries rather than search results and the AI acts as the travel agent rather than the retrieval mechanism.
In that environment, the properties that are visible are those whose data is structured for machine readability: amenities, pricing, and availability transmitted through clean Schema.org markup that AI agents can read and trust. Signal quality and structured data will determine visibility before any human guest makes a decision.
The guardrails built for AI Max today are the foundation for the data architecture that agentic systems will require tomorrow.
A question worth sitting with: does your current Google Ads structure have explicit protections around your brand terms and your highest-intent traffic, or is AI Max operating across your entire account without lane boundaries?