Why No Singapore AI Overview Figure Exists

Somewhere in the last two years, a figure started circulating in Singapore marketing content claiming that up to seventy per cent of queries in certain Singapore verticals now trigger an AI Overview. Trace it back and it does not lead to a Singapore study. It leads to a global industry dataset, relabelled.

We went looking for the real thing. Four independent research passes, across four separate models, briefed to find a primary measurement of how often AI Overviews appear in Singapore search results, or what they do to Singapore traffic. All four came back with the same answer, which is that no credible Singapore-geography measurement was found.

That is an uncomfortable answer to put on a service page, and it is the correct one. This article explains what is actually measured, why the global numbers disagree so violently with each other, why a Singapore figure has not appeared, and what to instrument on your own property instead of waiting for one.

What is genuinely measured, and where

The most methodologically transparent study we found is from the Pew Research Center, published in July 2025. It used browsing data from 900 United States adults covering 68,879 real Google searches conducted in March 2025. Users who encountered an AI summary clicked a traditional search result link in 8% of visits. Users who did not encounter one clicked in 15% of visits. Only about 1% clicked a link inside the summary itself, and roughly 18% of the March 2025 searches produced an AI summary at all.

Note the geography. United States. Nine hundred people. One month. It is a good study, and it is not a Singapore study, and no amount of directional reasoning converts one into the other.

Beyond Pew, the picture fragments, and it is worth naming the individual studies rather than gesturing at a range, because the methods are what explain the disagreement.

Ahrefs published an analysis in April 2025 finding an average click-through drop of 34.5% for the top-ranking page on queries where an AI Overview appeared. A later Ahrefs release, reported in February 2026 and covering more than 863,000 keywords, put the drop nearer 58%. Same publisher, same broad metric, widening result across measurement windows. That is not a contradiction. It is a moving target being measured twice.

Seer Interactive ran the largest sample we found: a study spanning January 2025 to February 2026 across 53 brands, 5.47 million tracked queries and 2.43 billion impressions. Organic click-through on queries where an AI Overview appeared fell from roughly 1.76% at a June 2024 baseline to 0.61% by September 2025, a relative decline of about 61%. Then the interesting part, which almost nobody quotes: the same study showed a partial rebound to around 2.4% by February 2026. Anyone citing the 61% decline without the rebound is quoting a study's worst month as its conclusion.

On the publisher side, Chartbeat data reported via the Reuters Institute in January 2026 found Google search referrals to more than 2,500 publisher websites fell approximately 33% globally in the year to November 2025, with United States publishers down about 38%. That measures total referral traffic, which mixes AI summaries with algorithm updates, seasonality and shifting social referral patterns, so it is directionally useful and causally messy.

Two honesty notes on all three. None of them is Singapore. And we reached the Ahrefs, Seer and Chartbeat figures through secondary write-ups rather than the primary reports, which is a weaker evidential position than the Pew figure and we would rather say so than imply otherwise.

Appearance-rate figures across different global datasets and windows range from about 11% to over 48%. These are not four measurements of one thing that happen to disagree. They are measurements of four different things.

Why a Singapore figure has not appeared

This is not neglect. It is arithmetic and incentive.

A credible appearance-rate study needs a fixed, disclosed query panel, controlled device and login state, repeated observation over time, category segmentation, and a stated denominator. Build that for a market of Singapore's size and the sample gets thin fast in any single vertical, because there simply are not that many high-volume Singapore-specific commercial queries in, say, corporate secretarial services.

The organisations with the infrastructure to run it are the large SEO platforms, and their commercial incentive points at large markets first. A United States cut sells. A Singapore cut is a rounding error in their addressable market. So the country cut does not get published, and the vacuum gets filled by agencies relabelling the global cut, which costs nothing and reads authoritatively.

There is a second, subtler reason. AI Overview behaviour is not stable. Rollout varies by query category, by language, by whether the user is signed in, and by ongoing product changes on Google's side. A figure measured in March would need restating in September. Publishing a country-level number with a long shelf life is genuinely hard, and publishing one with a short shelf life is commercially unattractive.

Five real Singapore AI numbers, none of which answer the question

There is no shortage of Singapore AI statistics. That is precisely what makes this area dangerous, because a page can be dense with genuine Singapore figures and still contain nothing about whether AI systems mention your brand.

IMDA's Singapore Digital Economy Report 2025, released 6 October 2025, recorded AI adoption among Singapore small and medium enterprises tripling in a single year, from 4.2% in 2023 to 14.5% in 2024. Non-SMEs moved from 44% to 62.5% over the same period. Of firms adopting AI, 84% reported using off-the-shelf generative AI tools, 52% adopted AI-enabled solutions for domain-specific tasks such as human resources or accounting, and 44% had implemented customised or proprietary AI tools.

Singapore's Ministry of Manpower measured something adjacent and got a very different number. Its establishment survey, covering 2,560 private-sector establishments with fieldwork in the first quarter of 2026, found 28.5% had started adopting AI. The definition is broader than IMDA's, including preliminary stages such as piloting and planning, and the unit is the establishment rather than the enterprise. Both figures are correct. Presenting them as a range from 14.5% to 28.5% would be wrong, because they are not measuring the same thing.

At population level, the Microsoft AI Economy Institute placed Singapore second globally for generative AI diffusion, with 60.9% of the working-age population aged 15 to 64 having used a generative AI product in the second half of 2025, rising to 63.4% by the first quarter of 2026. That is individual usage, not firm adoption and not search behaviour.

Two consumer studies measure shopping behaviour specifically. The Adyen Index Retail Report 2025, fielded by Censuswide across 41,089 consumers in 28 markets between 26 February and 12 March 2025, found 49% of Singapore consumers had used AI assistants to enhance their shopping experience, rising to 64% among Millennials and 58% among Gen Z, with 59% open to purchasing directly through AI interfaces. Salesforce's sixth Connected Shoppers Report, fielded November to December 2024 including 500 Singapore shoppers, found 36% of Singapore shoppers already using AI for product discovery, rising to 53% among Gen Z.

Five figures. All Singapore. All real. Not one of them tells you whether a generative engine names your brand when a Singaporean asks it for a recommendation, because none of them was designed to. Consumers using AI to shop is a demand-side fact. Brands appearing inside AI answers is a supply-side fact. The gap between those two is exactly where the missing measurement sits.

Singapore AI figure	Unit measured	Source and period	Answers the visibility question?
4.2% to 14.5%	SMEs adopting AI	IMDA, 2023 to 2024	No, firm adoption
28.5%	Establishments that have started adopting AI, including piloting and planning	Ministry of Manpower, 2,560 establishments, first quarter 2026	No, and not comparable to the IMDA figure
60.9% to 63.4%	Working-age population aged 15 to 64 using a generative AI product	Microsoft AI Economy Institute, second half 2025 to first quarter 2026	No, population usage
49%	Consumers who have used AI assistants for shopping	Adyen Index Retail Report 2025, Censuswide, fieldwork February to March 2025	No, consumer behaviour
36%	Shoppers using AI for product discovery	Salesforce Connected Shoppers, 500 Singapore respondents, late 2024	No, consumer behaviour
Appearance rate in Singapore results	Share of Singapore queries displaying an AI Overview	No source located	This is the question, and it is unanswered

One further piece of Singapore context worth having, because it explains why the adoption figures move so fast. Internet penetration here was reported at 98.4% in DataReportal's Digital 2026 report using October 2025 data, up from 95.8% reported in the previous edition using January 2025 data. Those two figures are different measurement windows rather than a contradiction, and the useful reading is that near-universal connectivity means new search interfaces reach saturation quickly. It still says nothing about citation.

Figure	What it measures	Geography	Safe use
8% against 15%	Click rate on a traditional result, with and without an AI summary present	United States	Quote with the geography attached, every time
Roughly 18%	Share of searches in the panel that produced an AI summary	United States, March 2025	Single month, single panel. Not a trend line
34% to 61%	Various click-through reduction measures across different definitions	Global or unstated	Present as a range with the definition problem stated
4.2% to 14.5%	Share of Singapore SMEs adopting AI, 2023 to 2024	Singapore	Adoption context only. Never as a visibility figure
92.46%	Google's share of Singapore search, July 2026	Singapore	Restate with month and year, it is a monthly series
AI Overview rate, Singapore	Appearance rate in Singapore results	Singapore	Not available. Do not substitute a global figure

What to instrument instead

The absence of a market figure is less limiting than it sounds, because a market figure was never going to tell you what happened to your site. Your own data can, and most organisations already have it and are not looking at it correctly.

The core move is to stop reading rankings as a proxy for traffic. If a summary occupies the space above the results, position three can hold steady while clicks fall, and a rank-tracking report will show a flat green line through the entire decline. The measurement that catches it is click-through rate segmented by query, because that is where the divergence appears first.

From there, four things are worth separating. Informational queries behave differently from commercial ones, so segment them and expect the pressure to land on the informational side. Branded and non-branded impressions behave differently, so split them, because branded demand can mask a non-branded collapse for months. Impressions and clicks should be read together rather than either alone, since impressions holding while clicks fall is the signature pattern. And conversions matter more than sessions, because a smaller volume of higher-intent traffic is a different situation from a smaller volume of everything.

Where the ranking question ends and the citation question starts

An AI Overview sits inside Google's own results page, which makes it a shared surface. Whether you are eligible to be summarised there is partly an organic search question, because it depends on the same crawlability, structure and credibility that ranking depends on.

Whether a generative engine names you when someone asks it a question in a chat interface is a different problem. Different surfaces, different retrieval, different instrumentation, and different reporting. Conflating the two produces a dashboard where a rise in one masks a fall in the other and neither number can be acted on.

So we keep them apart deliberately. Organic sessions, non-branded visibility, click-through rate and conversions belong to our SEO service for Singapore. Citation share and inclusion across answer engines belong to our GEO service for Singapore, which exists as a separate service precisely because it needs separate measurement rather than a line item appended to a ranking report.

What would change our position

If one of the major platforms publishes a Singapore geography cut measuring appearance rate and click impact on Singapore-targeted sites, with a disclosed query panel and stated denominator, we will use it and say so. That is a specific, falsifiable condition, and stating it is the difference between honest uncertainty and permanent hedging.

Until then, the answer is that the figure is not available, and the useful work is instrumenting your own property. That is less satisfying than a headline percentage. It is also the only version that survives someone asking where the number came from.

---

Frequently Asked Questions

Is there really no Singapore AI Overview statistic at all?

None that we could find, and we ran four independent research passes across four separate models specifically looking for one. No primary Singapore-geography measurement of appearance rate or traffic impact surfaced. The Singapore percentages circulating in agency content trace back to global industry datasets that were relabelled as local. If someone shows you a Singapore figure, ask for the query panel, the observation window and the denominator before using it.

Why can't I just use the global figures for Singapore?

Because the global figures do not agree with each other, and the disagreement is structural rather than accidental. Appearance rate, top-position click-through, panel click behaviour and publisher referral traffic are four different measurements producing four different numbers. Picking one and applying it to Singapore compounds two errors: the wrong geography and an undefined metric. Quoting a global figure with its geography stated plainly is defensible. Presenting it as local is not.

What is the most reliable global number to reference?

The Pew Research Center study published in July 2025 has the clearest methodology we found: browsing data from 900 United States adults across 68,879 Google searches in March 2025, finding a traditional result was clicked in 8% of visits where an AI summary appeared against 15% where it did not, with roughly 18% of searches producing a summary. Use it with United States attached to it every time, because it is a national panel study and not a global or Singapore figure.

Can I use Singapore AI adoption data instead?

Only for context, and only if you keep the distinction explicit. IMDA's Singapore Digital Economy Report 2025 recorded AI adoption among Singapore SMEs rising from 4.2% in 2023 to 14.5% in 2024, and non-SMEs from 44% to 62.5%. That measures whether firms use AI tools. It says nothing about whether AI systems mention your brand, which is an unrelated question. Presenting adoption data as evidence about visibility is the single most common error in this area.

How do I tell whether AI Overviews are affecting my own traffic?

Watch click-through rate by query rather than average position. The pattern to look for is impressions holding steady while clicks fall on the same query set, which is what a ranking report cannot show you because the rank has not moved. Segment informational from commercial queries, and branded from non-branded, since the pressure lands unevenly. Set the baseline now even if you do nothing else, because retrospective baselines are guesswork.

Should I build my own Singapore measurement panel?

If AI search matters to your category, yes, and it is more achievable than it sounds. Choose thirty to fifty commercial queries relevant to your business, check them from a Singapore location on a fixed schedule with consistent device and login conditions, and log whether a summary appeared and which sources it named. Within two quarters you have category-specific local evidence. It will be more useful to you than a national average would have been, because it is about your queries.

Does this mean SEO no longer works in Singapore?

No. Google held 92.46% of Singapore search in July 2026 according to Statcounter, so organic search remains the dominant discovery surface by a wide margin. What has changed is that ranking position is a weaker proxy for traffic than it used to be, which is a measurement problem rather than a channel problem. Commercial and transactional queries continue to drive clicks, because someone comparing vendors needs to reach the vendor.

How is this different from what your GEO service measures?

An AI Overview sits inside Google's own results page and depends partly on the same signals that drive ranking. Whether a generative engine names your brand in a chat interface is a separate problem with separate retrieval and separate instrumentation. We report organic sessions, non-branded visibility, click-through rate and conversions on the search side, and citation share and inclusion on the generative side, and we keep them in separate reports so a rise in one cannot hide a fall in the other.

---