The property is a full-service hotel operating under a global flag, in a central Singapore location. Its most distinctive commercial assets are specific and named (anonymised in this version): several independently positioned food-and-beverage outlets, a signature event venue, and a rooftop leisure facility. The hotel operates its own standalone microsite and worked with another SEO provider before OmniHyper. OmniHyper began working with the property in August 2024, activating Helium (organic search) to build a foundation across organic discovery.

In late 2025 and early 2026 the hotel reached a genuine crossroads. The team could see discovery shifting into AI: the question was no longer only “which page ranks on Google” but “which property does AI recommend when someone asks ChatGPT, Google’s AI Mode, Perplexity or Gemini.” The budget to be a first mover in AI search did not exist alongside the existing programme, so the hotel chose to be dynamic with the budget it had: Helium was suspended and AI Source Dominance was activated on 1 April 2026. This was a reallocation, not an addition. AI Source had to replace the value of traditional search optimisation, not merely supplement it. This case study measures its first three months.

It did so into a falling market. Singapore’s international visitor arrivals declined year on year in both April and May 2026: May arrivals fell 9.7% to 1.24 million (the lowest month of the year), overnight visitors fell 11.8%, and cumulative January–May arrivals were 1.2% below 2025 (June data had not been published at the time of writing).2 Traditional search demand fell harder still. Across an identical set of 1,413 tracked Singapore hotel keywords, Google’s own Keyword Planner records total average monthly search volume down 13.2% year on year for the quarter, with 253 keywords declining against 40 growing; “hotel singapore” itself dropped a full volume tier, from the 50,000 bucket to 5,000.3 Google Trends shows the same pattern, with most top related queries for “hotel singapore” declining, by as much as 30%.⁴

The search that remained was also resolving differently. Over the measurement period the property’s Google Search Console impressions rose 9.3% (to 533,400) while clicks fell 18% (to 18,226). The hotel was seen more and clicked less, consistent with AI Overviews, AI Mode and other answer surfaces resolving traveller questions before the click.5 This is precisely the environment AI Source exists for: to make the hotel the answer inside AI-generated recommendations, and to convert that visibility into direct demand.

AI Source replaced the suspended Helium programme with a structured answer-engine (AEO/GEO) programme:

• Named, specific, extractable content. Building the hotel’s story around distinctive named assets (the signature event venue, the independent dining outlets, the central location and the property’s design story), structured with clear answers and FAQs so AI systems can find, understand and cite it. The working pattern: distinctive, named, specific content earns citations; generic hotel copy is invisible.⁷
• Named cultural moments. Supporting culturally distinctive, named collaborations, most notably a seasonal food-and-beverage collaboration with a much-loved local heritage brand, that create their own search categories and AI citation classes.⁶
• Full-platform AI tracking. Prompt-level monitoring of how the hotel appears across ChatGPT, Google Gemini (including AI Overviews and AI Mode), Perplexity, Claude and Microsoft Copilot: visibility, sentiment, position and citation sources, updated continuously in the OmniHyper dashboard.⁸
• Competitive share-of-voice tracking. AI share of voice measured against a named compset of five directly competitive hotels (anonymised here as Competitors A–E), so progress is judged against the hotels the property actually competes with.⁶
• Monthly reporting by Mae, OmniHyper’s AI digital marketing agent, to hotel and brand stakeholders, drawn from Google Analytics 4, Google Search Console, the hotel’s commercial workbook and AI Source brand-performance data, including direct written responses to stakeholder questions on methodology and attribution.⁹

1 April 2026 to 30 June 2026, vs 1 April 2025 to 30 June 2025 unless labelled otherwise.

Search ranking

Alongside visibility and engagement, the hotel’s own commercial workbook, populated by the hotel team independently of OmniHyper and of Google Analytics, shows a property that held its ground in a falling market and shifted its mix toward direct:¹³

Read in isolation, Google Analytics on the microsite tells a misleadingly negative story for this property over this period: recorded “book” events down 46.9% and tracked purchases down 25.8% year on year, even as sessions rose 18.4%, and even as the hotel’s own systems recorded 600 additional direct web room nights^{.11, 13} Three well-understood, documented factors explain the divergence:

• GA4 is a last-click model watching a subset of the funnel. It tracks only bookings completed on the microsite in the same tracked session: tracked revenue equivalent to roughly 7% of actual direct web revenue in the hotel’s systems. Guests increasingly discover the hotel through AI answers and branded search, then complete on the brand’s central booking site, by phone, or in a later direct session, none of which credit the microsite^{.11, 13}
• The 2026 traffic mix diluted per-session intent. Paid Social sessions grew 655% (1,135 → 8,573) at an 8.5% engagement rate, with zero tracked purchases and $0 tracked revenue, and Direct sessions grew 88.9% (9,976 → 18,848), the latter inflated by untagged campaign links and a single-day 28 June anomaly that Google’s own anomaly detection flagged: roughly 3,100 direct visits against a forecast of 130.14 With thousands of additional low-intent sessions in the mix, event rates fell mechanically; Organic Search, the highest-intent channel, declined 18.6% in sessions in line with the market-wide click decline. Event-count comparisons against the 2025 baseline therefore overstate any real change in booking behaviour. The hotel’s own workbook, with direct room nights up 9.1%, is the ground truth.^{11, 13}
• Google changed the channel taxonomy mid-period. GA4’s “AI Assistant” channel began absorbing AI traffic during June 2026. As a result, the legacy chatgpt.com referral line shows −19.4% year on year (62 → 50 sessions) even though total AI-origin traffic grew strongly: AI-referred users +146% year on year, and 104 ChatGPT sessions in June alone. Comparing any single AI line item across the taxonomy change understates AI traffic; the channel must be read in aggregate ^{.8, 9, 11}

A mixed-currency tracking issue in January 2026 (resolved in Q2) further limits Q1 revenue comparability.10 There is also a measurement gap in the other direction: meetings and events demand, where the hotel’s AI visibility has grown despite not being actively targeted, is captured by neither GA4 nor the current workbook, so any value arising there goes unrecorded until MICE and C&E revenue is added to the workbook.13 For all of these reasons, this case study follows a simple discipline: commercial outcomes are drawn from the hotel’s own internal reporting, the system of record for what guests actually booked, while Google Analytics and Search Console are used for what they measure reliably: discovery, engagement and intent.

This is a hotel that responded to a structural change in guest discovery by making a deliberate, budget-constrained bet, suspending traditional SEO to fund AI search, in the same quarter its market turned down. Three months in, the evidence runs in one direction. AI mentions tripled to 48, with top-3 positions in 15 non-branded categories and two #1s in the hotel’s target territory. AI-referred users grew 146% year on year and arrived at engagement rates well above site average. Total sessions grew 18.4% while the traditional click pool shrank. And the hotel’s own reporting recorded identical occupancy, a 2.1% ADR gain, 2.0% revenue growth, and a 600-room-night direct gain against a near-flat OTA channel, in a city where visitor arrivals were falling.2

The trajectory matters more than the quarter. April and May were transition months, with direct revenue slightly behind prior year while the programme built its citation base; June, the month AI visibility broke out, delivered direct web revenue up 19.6% year on year while indirect declined. That sequencing, visibility first and bookings following, is exactly the shape an AI-discovery programme should produce, and it is corroborated across independently owned systems. Whatever the programme’s meetings-and-events rankings convert into will be additional to these figures, because MICE revenue is not yet tracked in the hotel’s workbook.

One alternative explanation deserves to be addressed directly, because the hotel invested meaningfully in paid social over the same period: could the direct-channel gains be paid social rather than AI Source? The recorded data says no. Paid Social generated 8,573 sessions in the quarter (up 655% year on year), but those sessions engaged at an 8.5% rate, stayed an average of three seconds, and produced zero tracked purchases and $0 of tracked revenue; recorded booking-intent (“book”) events on the channel actually fell 92% year on year (26 → 2).11 AI-referred visitors show the opposite profile: 55–57% engagement rates and roughly two-minute sessions, with tracked bookings present.9, 10 Timing points the same way. Paid social ran at many times its prior-year level throughout the quarter, yet direct revenue lagged prior year in April and May; the inflection arrived in June, the month AI mentions tripled and the AI Assistant channel went live, when paid social’s own month-on-month growth was at its smallest (+10%).9, 13 Paid social may contribute upper-funnel awareness that the data cannot capture, but on every signal that is measured (engagement, intent events, purchases, revenue and timing), the direct-channel shift tracks AI Source, not paid social.

Attribution deserves honesty. Three months is a short window; absolute AI-referral volumes are still small; the AI Source programme inherited organic foundations built by Helium over twenty months; the hotel and its brand run their own marketing; and non-branded Google rankings declined once traditional optimisation was suspended, a real, visible cost recorded in this document. No single model can isolate one programme’s contribution, and this case study does not claim one. Instead it triangulates five independently owned data sources, none of which rely on OmniHyper to verify:

• Market demand: Singapore Tourism Board arrivals data, Google Keyword Planner and Google Trends (all public).2, 3, 4
• Search presence: Google Search Console (Google’s own reporting, brand-accessible).5
• Website engagement and intent: Google Analytics 4 / Analytics 360 (brand-accessible property).11
• AI visibility: AI referral traffic in GA4, plus prompt-level AI Source tracking and compset share of voice.6, 8
• Commercial outcomes: the hotel’s own commercial workbook, populated by the hotel team.13

All five move coherently over the same window, and the improvement concentrates precisely where AI Source was engaged to act: AI citations and referrals, named-asset discovery, and the direct channel. When the market fell, tracked search demand fell, and traditional rankings were deliberately de-funded, and the hotel still held occupancy, raised rate and grew direct share, with the commercial inflection landing in the month AI visibility broke out, the most economical explanation is that AI-assisted discovery is beginning to do the work traditional search used to do. Early days, honestly labelled, but the first ninety days point the right way.

This case study was intentionally designed to minimise reliance on agency-reported metrics. Wherever possible, outcomes are drawn from systems owned by Google, the hotel’s brand, the hotel itself, or public statistical agencies, allowing every claim to be independently verified.