Google AI Overviews Are Changing SEO: How Singapore Brands Should Respond

Canlah AI finds, in this review of five click-through studies and its own September 7, 2026 audit, that Google AI Overviews are changing SEO by separating two outcomes that used to travel together: ranking on the results page and being chosen as a source for the answer above it. Studies from Pew Research Center, Ahrefs, Seer Interactive and a randomized experiment by Carnegie Mellon University and Indian School of Business researchers agree on the direction. When an AI Overview appears, fewer searches end in a click to a website. They disagree on the size, and the disagreement is informative.

Google AI Mode, available in English in Singapore since August 2025, extends the same shift by answering multi-step questions in a conversational pane. For Singapore brands, whose buyers ask in English and Chinese across several engines, a page can rank and still be absent from the answer. Ranking is one input to AI visibility, not AI visibility itself.

Key findings

• Clicks fall when an Overview appears: Ahrefs measured a 58% lower click-through rate for the top-ranking page in December 2025, and a randomized study of 1,065 US Chrome users found a 39.8% drop in outbound organic clicks.
• Citation changes the size of the loss, not its direction: Seer Interactive reports that brands cited in an AI Overview earned 120% more organic clicks per impression than uncited brands, but still 38% fewer than on queries without an Overview.
• Canlah AI was absent from its own category’s Overviews: in a browser-rendered Singapore sample on September 7, 2026, AI Overviews appeared on 62 of 65 completed renders and never mentioned Canlah AI on the 51 non-brand renders.
• A four-layer response model: eligibility, citation, mention and click should be recorded separately, because each layer fails for a different reason and needs a different owner.

What the AI Overviews click-through rate studies measured

The five studies below use different samples, windows and definitions, so their figures should not be averaged into one number. Only the study by Saharsh Agarwal and Ananya Sen randomly hid Overviews from real users, which makes it the one causal estimate in the set.

Five AI Overviews click-through studies compared, reviewed September 23, 2026.

Study	Sample and window	Reported result	Operational meaning
Pew Research Center, Jul. 22, 2025	900 US adults, 68,879 Google searches, March 2025	A traditional result was clicked on 8% of visits with an AI summary and 15% without; summary links were clicked on 1%	Sources inside the summary receive very few direct clicks.
Ahrefs, Feb. 4, 2026	300,000 keywords, Search Console CTR, December 2023 against December 2025	Position 1 CTR 58% lower where an Overview appeared; position 10 19.4% lower	The top result loses most because the Overview sits directly above it.
Seer Interactive, Apr. 24, 2026	53 brands, 5.47 million queries, January 2025 to February 2026	Organic CTR on Overview queries fell to 1.3% in December 2025, then rose to 2.4% in February 2026	The decline may be levelling, not returning to pre-Overview rates.
Agarwal and Sen, SSRN, revised Jun. 17, 2026	Randomized field experiment, 1,065 US desktop Chrome users	Hiding the Overview raised outbound organic clicks from 0.37 to 0.62 per search, reported as a 39.8% loss	The first causal estimate; earlier figures are associations.
Semrush, Jul. 30, 2025	About 69 million US desktop Google sessions, May to July 2025	92% to 94% of AI Mode sessions ended without an external visit	AI Mode sends even fewer clicks than a results page with an Overview.

Source: study pages and publisher summaries reviewed September 23, 2026. Figures are reported as published and were not averaged; only the Agarwal and Sen experiment supports a causal reading.

Sources: Pew Research Center; Ahrefs; Seer Interactive; PPC Land summary of Agarwal and Sen; Semrush

Ahrefs, Seer Interactive and Semrush sell SEO software or services, so their studies are not neutral market consensus. All five samples are US-based. None measures Singapore directly, and none measures whether the lost clicks would have converted.

Why AI Overviews click-loss studies disagree on size but not direction

The spread between 39.8% and 58% is mostly a difference in what was counted. Ahrefs isolates the top-ranking page, where the loss is largest, while the randomized study counts every outbound organic click, including lower positions that lose less.

Seer’s intent data explains where the effect concentrates. Informational queries showed an Overview 36% of the time, commercial queries 8% and transactional queries 5%. Comparison queries triggered an Overview 95.4% of the time and question-format queries 85.9%. Those are the query shapes that open a B2B shortlist.

Seer also reported a measurement trap. Cited brands saw lower click-through rates through 2025, but impressions more than doubled while clicks stayed flat. Google’s guidance calls clicks from pages with AI Overviews “higher quality”, a first-party claim that was not independently tested here.

Sources: Seer Interactive; Google Search Central, AI features and your website

Why ranking stops predicting visibility

The answer sits above position one

In the randomized study, more than 87% of Overview appearances placed the summary in the top position, above every organic listing. Removing it from that slot produced an 88% relative increase in outbound clicks, with no measurable effect when the Overview appeared lower on the page.

Query fan-out widens the source pool

Google states that AI Overviews and AI Mode may use a “query fan-out” technique, issuing multiple related searches across subtopics before composing an answer. A page that answers one subquestion well can be cited without ranking for the head query.

Search Console blends the evidence

Google reports AI Overviews and AI Mode traffic inside the overall Web search type in Search Console. A lower click-through rate and a new citation can therefore sit in the same report line, and the report alone cannot show which happened.

A first-party view from Canlah AI

On September 7, 2026, Canlah AI ran its own audit pipeline on canlah.ai for 39 English buyer questions in Singapore. Google AI Overviews were sampled through browser-rendered requests, two runs per question. Of 70 requests, 65 completed after transport failures and a budget cutoff, and an Overview appeared on 62. On the 51 non-brand renders with an Overview, Canlah AI was never mentioned and canlah.ai was never cited. On the 11 branded renders, it was mentioned every time and cited 10 times.

The organic result page showed the same structure. A scoreboard recorded on September 3, 2026 placed canlah.ai on one of 10 non-brand commercial queries, and on “geo agency singapore” it was absent from the top nine results. Across the audit’s API engines, the most-cited domains were aistudio.com.sg with 91 citations, mediaplus.com.sg with 85 and stridec.com with 78. This is a visibility gap in Canlah AI’s own marketing, not a customer success claim. The site itself was technically clean, with 80 of 80 crawled pages indexable.

The evidence shows that technical eligibility and Overview inclusion measured different things on this question set. It does not prove that a higher organic ranking would have produced a citation, or which change would alter the result.

Sources: Canlah AI, The Own-Brand GEO Audit Experiment

Citation, mention, recommendation and click are separate outcomes

An AI Overview can give a brand four separate outcomes, and a report that merges them hides which one failed. A citation means a page supplied information to the answer. The brand may never appear in the prose. A mention means the brand was named, but the framing may be neutral, outdated or wrong. A recommendation means the answer presents the brand as a fit for the buyer’s need. A click means the buyer left Google for the site, the rarest of the four outcomes in the Pew data, where summary links were clicked on 1% of visits.

The Canlah AI audit of September 7, 2026 shows how far the outcomes can diverge. The 11 branded renders produced 11 mentions and 10 citations, while the 51 non-brand renders produced no mention and no citation. A single blended visibility score across both groups would have reported a partial result that described neither.

What AI Mode changes for SEO

AI Mode became available in English in Singapore in August 2025, as HardwareZone reported on August 25, 2025. Google describes it as suited to complex comparisons, and states that AI Mode and AI Overviews may use different models and techniques, so the responses and links they show will vary.

That point matters most for AI Mode SEO. A brand cited in an Overview for a question has no guarantee of appearing in the AI Mode answer to the same question, because the two surfaces can select different sources. Semrush’s clickstream figure of 92% to 94% zero-click sessions suggests that visibility inside the AI Mode answer, not the click after it, is the main outcome available. Canlah AI’s September 7 audit did not sample AI Mode, so no first-party AI Mode figure is reported here.

Sources: HardwareZone Singapore

A four-layer response model for Singapore brands

The response to AI Overviews is a measurement structure, not a tactic list: each layer answers one question and routes the fix to one owner.

Four-layer AI Overviews response model for Singapore brands.

Layer	What to record	Decision it supports
1 Eligibility	Indexing status, snippet eligibility, Googlebot access in robots.txt and structured data that matches visible text	Whether a page can be used at all; a technical fix, not a content fix.
2 Citation	For a frozen set of buyer questions, whether the Overview or AI Mode answer links the domain, which URL and which competing domains	Which owned page or third-party source needs to change.
3 Mention and framing	Whether the brand is named, its position in any list and whether the stated facts are correct	Whether the gap is retrieval or persuasion.
4 Click and conversion	Search Console clicks and impressions by query group, AI referral sessions and key events in analytics	Whether visibility produces business value.

Source: Canlah AI editorial framework drafted September 23, 2026 from Google Search Central guidance and the audit described above. It is a measurement structure, not a tested benchmark.

Google states that no special schema.org markup, AI text files or new machine-readable files are required to appear in AI Overviews or AI Mode. Structured data therefore belongs in the eligibility layer as a readability requirement, not a citation lever. A higher citation count without a click and conversion record is not the same as a working program.

What changes in Singapore

Singapore buyers do not ask one engine in one language. English questions reach Google’s AI surfaces, ChatGPT, Perplexity and Gemini, while Chinese-speaking buyers may also use DeepSeek, Qwen or Doubao, which draw on different sources.

The engines in Canlah AI’s own audit also disagreed sharply on the same questions. ChatGPT mentioned Canlah AI in 7 of 84 grounded answers to open buyer questions (8.3%), while Gemini mentioned it in 0 of 93 (0.0%). These figures apply only to that audit and category. The September 3 search audit found eight “best GEO agency in Singapore” listicles on the category’s result page, three of them written by competitors about themselves. When an Overview summarises that page, it is largely summarising those listicles. The response therefore reaches beyond the owned website.

What buyers should test before paying for AI Overviews optimization

• Freeze 20 to 40 non-branded buyer questions and keep branded questions in a separate group.
• Render Overviews from a Singapore location and record appearance rate separately from citation rate.
• Sample AI Mode separately from AI Overviews, because the two surfaces can cite different sources.
• Read Search Console clicks and impressions separately before calling a falling click-through rate a loss.
• Run a positive control before accepting a zero, because a blind tool and an absent feature report the same zero.
• Retest on the same questions and definitions, and label before-and-after results as association unless a control design supports causation.

Otterly.AI’s public page lists AI Overviews and AI Mode among the surfaces it monitors, and Ahrefs publishes an AI Overviews tracker. A tool can collect the record; it cannot decide which layer failed or who owns the fix.

Where Canlah AI fits

Canlah AI is a Singapore-based SEO + GEO agency that measures and improves brand visibility inside ChatGPT, Gemini, Google AI Overviews and Google AI Mode, reporting AI visibility as re-verifiable ranges with timestamped evidence. Engagements start from a locked buyer-question pool with raw answer archives, and its GEO service starts at $1,500 a month per AI platform.

It is most relevant to brands that need Overview citation measured and acted on alongside conventional SEO, not to every company seeking a low-cost rank tracker. The company’s own data also shows a genuine limitation: as of September 7, 2026, Canlah AI was absent from every non-brand Overview in its own category.

Methodology and limitations

Third-party figures come from public pages reviewed on September 23, 2026 and describe US samples, not a Singapore benchmark. The studies differ in unit, window and definition and were not combined. Canlah AI figures come from its September 3 and September 7, 2026 audit of canlah.ai, one property in one category with two Overview runs per question. That data has no control group and identifies visibility gaps, not revenue lost to AI Overviews.

Frequently asked questions

How are AI Overviews changing SEO?

AI Overviews are changing SEO by separating ranking on the results page from being chosen as a source for the answer above it. Third-party studies estimate click losses of 39.8% to 58% when an Overview appears, with Ahrefs reporting a 58% lower click-through rate for the top-ranking page. The practical consequence is that eligibility, citation, mention and click have to be measured as separate outcomes.

Do AI Overviews replace traditional SEO?

No. Google states that a page must be indexed and eligible for a snippet to appear in AI Overviews or AI Mode, so SEO remains the entry requirement. AI Overviews add a citation layer on top, where ranking no longer guarantees inclusion.

Is AI Mode SEO different from optimizing for AI Overviews?

Partly. Both use the same eligibility rules and may use query fan-out, but Google states they can use different models and show different links. A brand should measure citation in each surface separately rather than assume one result predicts the other.

Related articles

• SEO vs AEO vs GEO
• The Own-Brand GEO Audit Experiment
• How AI engines choose sources
• How to check if ChatGPT recommends your brand