24owlsGroup Pte Ltd
← Back to journal
Blog

Who Does AI Search Actually Recommend in Singapore? A GEO Visibility Audit

Only 37.9% of AI Overview citations come from Google's top 10 and 61.7% never name the brand. Inside the audit that shows who AI search recommends in Singapore.

Who Does AI Search Actually Recommend in Singapore? A GEO Visibility Audit

Only 37.9% of AI Overview citations come from Google's top 10 organic results, and 61.7% of AI citations never name the brand behind them. Those two numbers frame everything the research can currently tell you about who AI search actually recommends. Singapore makes the question urgent: it is the world's most AI-saturated search market, holding the highest per-capita Gemini adoption on earth, per Google, The Gemini Report | Southeast Asia 2026 | Singapore - Grow with Google APAC[1], while Google still takes 92.46% of classic search, per Statcounter Global Stats, Search Engine Market Share Singapore | Statcounter Global Stats[3].

Between those two facts sits an unmeasured layer: which brands, publishers and site types ChatGPT, Google AI Overviews, Perplexity and Gemini actually name and cite for Singapore-intent commercial queries. The published evidence says that layer is concentrated, decoupled from the Google top-10, and unstable from run to run. It also says every Singapore-specific number so far is vendor-graded. A disciplined prompt basket closes the gap; this guide walks you through what the research shows first.

Table of Contents

  • How AI-saturated Singapore search is: most Gemini prompts per capita on earth
  • Where AI answers moved in: commercial-intent queries, AI Overviews up 71%
  • Which sources AI engines actually cite: only 37.9% from Google's top 10
  • How concentrated citations are: top 10 domains take 46%
  • Named is not cited: 61.7% of citations never name the brand
  • Do the four engines agree? Only 41.6% of the time on a top pick
  • What Singapore data exists: all of it vendor-graded
  • What an audit cannot tell you: single-run rankings are noise
  • What actually moves visibility: citations, quotes and statistics, up to 40%
  • Frequently asked questions

How AI-saturated Singapore search is: most Gemini prompts per capita on earth

Singapore compresses the whole AI-search transition into one market. Google's own APAC Gemini report states that "Singapore boasts the highest adoption per capita of Gemini globally" (self-reported), with users averaging nearly 10 prompts a day and the Southeast Asia user base doubling in 12 months [1]. A Milieu survey of 1,000 Singapore respondents in June 2026 found 80% use AI tools for personal purposes, ChatGPT leads platform use at 82%, and general information search is the top use case at 73% Milieu Insight, AI in daily life—use, trust, and limits[2]. The IPS national survey, n=3,713, puts chatbot use at 64% of adults, with 81.2% of those users deploying chatbots for information, reviews or recommendations Institute of Policy Studies, NUS, Fraternity and the Social Fabric in the Digital Age[4].

Classic search still owns the layer underneath. Google holds 92.46% of Singapore search as of July 2026, with bing a distant 3.37% [3]. So the audit design in this piece uses the classic Google top-10 as its control group, not its target. Keep in mind that both layers matter at once: the audience prompting Gemini ten times a day is the same audience putting 92% of queries through Google.

Where AI answers moved in: commercial-intent queries, AI Overviews up 71%

The queries this audit targets - "best", "versus" and "recommend" prompts with Singapore intent - are exactly where AI coverage grew fastest. "In our six-month analysis of over 600,000 keywords across 10 industries, AI Overviews grew an average of 71% across search engine results pages (SERPs) with commercial intent," Hannah Pedri, Head of Paid Search & Social at Semrush, wrote of the November 2025 to April 2026 window, with Finance up 231% Semrush, AI Overviews are expanding across commercial intent search [Study][5].

Singapore demand sits ahead of that supply. The Adyen Index 2026 Singapore Retail Report found "more than seven in ten (72%) having used AI assistants to support their shopping in the past year," rising to 85% among Gen Z, as reported by MARKETECH APAC, Singaporean shoppers turn to AI as rising costs drive generational divide in spending: report[6]. The IPS survey points the same direction for information-seeking [4].

One caution. Singapore AI-shopping estimates swing sharply with question wording, so the audit must report its own mention tallies rather than lean on any single adoption figure.

Which sources AI engines actually cite: only 37.9% from Google's top 10

AI citation sets have structurally decoupled from the classic SERP. In an analysis covering what Ahrefs describes as "863K keyword SERPs, and a grand total of 4M AI Overview URLs," only 37.9% of AI Overview-cited URLs sat in the first 10 organic blocks, down from roughly 76% in July 2025; 31.2% ranked 11 to 100 and 31.0% did not rank in the top 100 at all Ahrefs, Update: 38% of AI Overview Citations Pull From The Top 10[7]. On the chat side, a June 2025 Semrush study, now a dated baseline, found that pages ChatGPT Search cites "rank in traditional organic search positions 21+ for related queries almost 90% of the time" Semrush, We Studied the Impact of AI Search on SEO Traffic | Our Findings[8].

Singapore's business press has already branded the local version of the gap. "Ranking on Google and being quoted by an AI answer engine turn out to be two different achievements. The second one runs on rules almost nobody has been told," Nicholas Tung, columnist at The Business Times, wrote on August 7, 2026 The Business Times (Singapore), Singapore's SMEs are winning at Google, losing at AI[9].

Simply put, rank tracking is structurally blind here. A Singapore brand can hold classic share of voice and be absent inside answers, or the reverse. Per-query overlap between AI-cited URLs and the Google top-10 becomes the audit's core measurable.

How concentrated citations are: top 10 domains take 46%

Three independent corpora agree the citation economy is winner-take-most. A 243-country MIT audit of 24,000 queries concluded that "AI search surfaces significantly fewer long tail information sources, lower response variety, and significantly more low credibility and right- and center-leaning information sources, compared to traditional search" arXiv (MIT), The Rise of AI Search: Implications for Information Markets and Human Judgement at Scale[10]. A Gradient Group analysis of 1.2 million ChatGPT responses finds the same head effect at domain level: "The top 10 domains take 46% of all citations in a topic. The top 30 take 67%" The Gradient Group, The Science Of How AI Picks Its Sources[11].

Profound's corpus of roughly 700,000 ChatGPT conversations, data window October to December 2025, shows a wide-but-shallow market with a Gini of 0.8, where "sources travel in packs. ChatGPT doesn't pick one winner" Profound, How ChatGPT sources the web[12].

What does that mean for your Singapore basket? Expect a handful of UGC, reference and comparison platforms plus a few big publishers to mediate most recommendations. Visibility gains will be non-linear, early citation incumbents compound, and brands without third-party mention coverage on those surfaces lose before on-site SEO matters.

Named is not cited: 61.7% of citations never name the brand

Citation dashboards overstate visibility, because most citations never name the brand they inform. The Semrush study with growth advisor Kevin Indig, covering 3,981 domain appearances across 115 prompts and 14 countries, found 61.7% of citations are ghost citations. "The AI knows the information about the brand came from somewhere, but doesn't feel the need to explicitly say so to users. The brand name carries on its own," Indig, who coined the term, explained Semrush, Why 62% of AI citations don't lead to brand mentions [Study][13].

The engine split is stark: Gemini names brands in 83.7% of appearances but cites a source only 21.4% of the time, while ChatGPT cites in 87% of answers but mentions brands in 20.7% [13].

Beneath the brand layer sits a hard inclusion floor. In a 37,000-run audit against a 533-brand catalog, 48 to 52% of specialist and regional brands never surfaced in any run, and even category leaders won only 25 to 41% of the recommendation slots they reached arXiv (Unusual.ai), Prominence-Stratified Failure Modes in Retrieval-Augmented Commercial Recommendation: A 37,000-Run Audit[14]. Independent RankSurf tracking of about 6,500 B2B answers shows the same gap: ChatGPT named a brand in 35.1% of relevant answers but cited that brand's own site in only 15.9% RankSurf, What ChatGPT & Perplexity Actually Cite (6,500 Answers)[15]. Keep the tally sheet honest: separate named from cited for every engine and query, or winners get overstated.

Do the four engines agree? Only 41.6% of the time on a top pick

The engines do not share one recommendation universe. GPT-5.2, Gemini 3 Flash and Perplexity sonar-pro agreed on the same top-recommended brand for only 41.6% of 250 category queries in a June 2026 preprint arXiv (Dmitrij Żatuchin), Who Owns the AI Recommendation? A Multi-Industry Empirical Map of Brand Category Ownership Across Large Language Models[16]. Undated BrightEdge tracking across five engines finds pairwise top-100 overlap in named brands between 36% and 55%, while cited-source overlap spreads from 16% to 59%: brands converge, sources diverge BrightEdge, Why AI Engines Cite Different Sources but Recommend the Same Brands[17].

Even inside Google the surfaces behave differently, with brands in 90% of AI Mode responses against 43% of AI Overviews and far higher week-to-week volatility in Overviews, per July 2025 BrightEdge data Search Engine Land, Brands dominate Google AI Mode, struggle in AI Overviews: Study[18]. "AI Mode provides stable, broad discovery while AI Overviews test new ranking approaches with higher selectivity," Jim Yu, Founder and Executive Chair of BrightEdge, said [18].

The implication is operational. Score each engine separately, because a brand winning ChatGPT can be absent on Gemini or Perplexity for the same Singapore query set without knowing it. Re-audit quarterly; the engine field is moving fast enough to justify the cadence.

What Singapore data exists: all of it vendor-graded

Every Singapore-specific number published to date is vendor-generated. Three comparison sites, SingSaver, MoneySmart and Seedly, drew 17,411 citations against 5,349 for DBS, UOB and OCBC combined, a 3.3-to-1 gap, according to Hashmeta AI, 2026 Singapore GEO Industry Report — Generative Engine Optimisation & AI Visibility in Asia's AI Command Centre[19], which could not be independently verified.

Nick Tung's audit of 46 Google-ranking Singapore SME sites scores them 48/100 on AI-citation readiness, a figure its own author labels directional rather than a census Nick Tung, Singapore AI Search Readiness Report 2026. SME Benchmark[20]. A VantrisAI index of 750 Singapore B2B domains averages 49.6/100 with 95.3% failing its FAQ schema check, likewise vendor-published with undisclosed methodology VantrisAI, Singapore AI Search Visibility Index 2026 | VantrisAI[21].

Together these sketch a market where intermediaries, not principals, own the answers. This is where disciplined measurement earns its keep in practice, and it is the sequence an agency like 24owlsGroup, a Singapore digital marketing and branding agency running SEO, Google Ads, social media and event management for SMEs in Singapore and Malaysia, builds into SEO work: a fixed basket of Singapore-intent commercial queries, scored monthly for citation share per engine, so month-to-month movement stays comparable rather than anecdotal. A transparent prompt-basket audit run that way would be the first non-vendor Singapore data point, but only if it publishes the prompt set, run counts, engine versions and timestamps. Those are exactly the disclosures most commercial trackers fail to make.

What an audit cannot tell you: single-run rankings are noise

The method has hard limits the GEO vendor market downplays. SparkToro's 2,961-prompt experiment found a less-than-1-in-100 chance of the same brand list appearing in two responses, and under 1-in-1,000 for the same list in the same order. "Any tool that gives a 'ranking position in AI' is full of baloney," Rand Fishkin, co-founder of SparkToro, concluded SparkToro, NEW Research: AIs are highly inconsistent when recommending brands or products; marketers should take care when tracking AI visibility[22]. A critical arXiv survey of 45 GEO studies, a preprint pending peer review, states that "no reviewed technique shows a stable, longitudinal, cross-platform causal effect on organic discoverability or downstream behaviour" arXiv (Olivier Martinez), Optimizing Visibility in Generative Engines: A Critical Survey of Generative Engine Optimization (2023–2026)[23]. A separate preprint that re-ran identical prompts within 24 hours measured same-day source-citation Jaccard at just 0.32 to 0.43 arXiv (Schulte, Bleeker, Kaufmann), Don't Measure Once: Measuring Visibility in AI Search (GEO)[29].

So report mention rates across repeated runs per prompt and engine, attach uncertainty to every concentration figure, and disclaim rank-order results. Single-run winner lists are lottery draws.

The honest baseline comes from practitioners. "If a GEO service does not openly tell you that success in AI visibility is 80 percent good fundamental SEO, they are selling you snake oil," Jeremy Moser, co-founder and CEO of uSERP, told Digiday, GEO hype busted: How it differs (and how it doesn't) from SEO[24].

What actually moves visibility: citations, quotes and statistics, up to 40%

The evidence-backed levers are unglamorous. The foundational KDD 2024 study, 10,000 queries across 25 domains, found that citing sources, adding quotations and adding statistics lifted visibility by up to roughly 40% in its experimental setting, conditional on the page already being retrieved; keyword stuffing showed little to no benefit in the reported results arXiv / ACM SIGKDD, GEO: Generative Engine Optimization[25].

Singapore practitioners argue the same from the field. "With GEO, ChatGPT, or Perplexity, a particular wine shop might simply be recommended because its information appears consistent, clear, and credible across multiple sources," Luke Janich, Founder and CEO of RED 2 Digital, wrote in Marketing-Interactive in November 2025 Marketing-Interactive, Generative engine optimisation: The next battleground for brands in Asia[26].

Google's own guidance closes the loop. "To get your content to appear in AI Overview, simply use normal SEO practices. You don't need GEO, LLMO or anything else," Gary Illyes of Google's Search team told an APAC Search Central event in July 2025 Search Engine Roundtable, Google: Normal SEO Works To Get Into AI Overviews[27].

Allocate budget in that order: SEO hygiene first, roughly 80% of the outcome by the practitioner baseline, then the genuinely new residual of entity consistency, third-party mentions on the platforms engines actually cite, recency, and per-engine mention tracking across repeated runs. Fundamentals first, mention coverage second, measurement every month - the sequence an agency like 24owlsGroup should follow when building AI-visibility audits into SEO work for Singapore SME clients.

Frequently asked questions

How is a GEO visibility audit different from classic rank tracking?

Rank tools see the top-10 organic results; AI engines largely source elsewhere. Only 37.9% of AI Overview citations come from the first 10 blocks [7], and ChatGPT cites pages ranking 21 or lower almost 90% of the time [8]. The audit tracks named-and-cited share per engine, with the Google top-10 as control group.

How many times must each prompt run before results mean anything?

More than vendors admit. The same brand list recurs in under 1 in 100 response pairs, and under 1 in 1,000 in the same order [22]; same-day citation overlap in one preprint ran 0.32 to 0.43 [29]. Run repeated samples per engine and report mention rates with uncertainty, never rank order.

Which engine should a Singapore brand prioritize first?

ChatGPT carries the referral weight, but weight is not exposure. Gemini names brands in 83.7% of appearances yet cites a source only 21.4% of the time [13], and three engines agreed on one top pick for only 41.6% of queries [16]. Weight ChatGPT, score all four.

Is GEO just SEO rebranded?

Largely, on the evidence above. Google's Illyes said to "simply use normal SEO practices" for AI Overviews [27], and uSERP's Moser puts the overlap at 80 percent [24]. The genuine residual is entity consistency, third-party mentions, recency and per-engine measurement.

Do Singapore SMEs actually get recommended by AI engines today?

Mostly no, on current local evidence: the Business Times reports SMEs winning at Google while losing at AI [9]; the 46-site audit scores ranking SMEs at 48/100, directionally [20]; and aggregators out-cite the three banks 3.3 to 1 in vendor data [19]. Benchmark with your own basket.

Will AI citations actually drive traffic or revenue?

Barely, today. Users clicked a traditional result on 8% of AI-summary visits versus 15% without, and only 1% clicked a citation, in Pew's July 2025 baseline Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results[28]. Amsive found LLM referral conversion statistically indistinguishable from organic while AI stayed under 1% of sessions Search Engine Land, LLM traffic converts about the same as organic search: Research[30]. Treat share of voice as a brand metric, with branded-search lift as the downstream proxy.

References

Got the next one?

Tell us about your brand, campaign or event. We'll bring the team.

Start a project →