Skip to main content
Perspectives 2026-09-23 Updated 2026-09-23 · 8 min read

Google Brings Preferred and Highly Cited Sources Into AI Search: What Singapore Brands Should Do Next

Google Preferred Sources reached AI Overviews and AI Mode on May 27, 2026, with wider Highly Cited labels. What Singapore SMEs, clinics and schools should do next.

A single dark card with a narrow indigo strip across it

Google Preferred Sources entered AI Search on May 27, 2026, when Google extended the setting from Top Stories into AI Overviews and AI Mode. For Singapore SMEs, clinics and schools, the practical opportunity is larger than earning a badge: it is building evidence that AI systems can retrieve, attribute and corroborate. Canlah AI’s own audit, run on one site on September 7, 2026, shows why those three verbs must be measured separately. On that date Google AI Overviews appeared on 62 of 65 rendered Singapore searches, including 11 that named the brand, yet named Canlah AI in none of the 51 open buyer answers sampled. The lesson is not that a label fixes visibility. A brand has to become a source before it can be preferred or cited.

Key findings

  • Adoption so far: Google says people had selected more than 345,000 unique Preferred Sources and were twice as likely to click through to a source they had selected.
  • What the two labels are: Preferred Sources is a personal setting and Highly Cited marks articles that many other stories cite; neither is described as a general ranking factor in Google’s announcement or its Search Central guide.
  • Clicks are thin either way: Pew Research Center recorded traditional-result clicks in 8% of visits with an AI summary against 15% of visits without one, and clicks on a cited source in 1% of visits.
  • Recognition is not recommendation: in the Canlah AI self-audit of September 7, 2026, ChatGPT named the brand in 7 of 84 open buyer answers, Gemini in 0 of 93 and Google AI Overviews in 0 of 51.
  • What that audit cannot show: it covers one young agency site, one market and one day of sampling, so it describes a baseline rather than a Singapore benchmark.

What Google changed

On May 27, 2026, Google announced that Preferred Sources, already used in Top Stories, would appear in AI Overviews and AI Mode. Users add domains through the source preferences tool in Search personalisation settings, and links from those domains carry a “Preferred” label inside AI responses. Google said any website that publishes fresh content is eligible. It also added a prominent link carousel for some developing topics and a carousel of first-hand perspectives from forums and social media.

The same announcement extended the Highly Cited badge to more web article links. The label marks articles that many other stories have cited, and Google also indicates when an article explicitly references a Highly Cited source. Google’s Search Central guide states that only domain-level and subdomain-level sites are eligible for Preferred Sources; a subdirectory such as example.com/blog is not. Sites must also be included in Search generative AI features in Search Console. Preferred Sources is available in AI Mode and AI Overviews in all languages and locales where those features run.

Sources: Google product announcement; Google Search Central Preferred Sources guide

The update should be read as a discovery and trust signal, not a shortcut to AI inclusion. Preferred Sources depends on a reader’s own selection, so it reaches an audience the brand already has. Highly Cited has no application process, and the badge sits on the link rather than on the sentence above it.

Why Google Preferred Sources matter for Singapore brands even when clicks decline

Pew Research Center analysed 68,879 Google searches by 900 US adults in March 2025, including 12,593 searches with an AI summary. Users clicked a traditional result in 8% of visits with a summary and in 15% of visits without one. They clicked a source cited inside the summary in 1% of visits. Citation can therefore shape the answer a buyer reads while producing very little referral traffic.

Sources: Pew Research Center click study

For a Singapore business, that changes what a label is worth. A parent comparing tuition centres in Bishan or a patient looking for a GP clinic in Tampines may never click a link at all. What matters is whether the AI answer names the business and describes it accurately. A Preferred badge helps only readers who already chose the site, and Highly Cited mainly rewards original reporting that other publishers reference.

The labels still matter, but they sit downstream of the citation pool. A domain that no engine cites is not a candidate for either label, because there is no link to attach the badge to. The work therefore starts on the pages that engines already read for Singapore category questions.

Canlah AI evidence: presence, mention and citation are different

Canlah AI ran its own audit pipeline on canlah.ai on September 7, 2026, for Singapore in English. The question set held 39 buyer questions grouped into funnel layers. Each question was sent three times to ChatGPT through the OpenAI API with web search and three times to Gemini. Google AI Overviews were sampled separately as browser renders with Google’s Singapore country setting. Only grounded answers were counted, and branded questions were never mixed with open ones.

Canlah AI self-audit of three AI surfaces, Singapore, September 7, 2026

Surface Open-question answers Canlah AI named Mention rate Branded answers naming Canlah AI What the row implies
ChatGPT (OpenAI API, web search) 84 7 8.3% 18 of 18 Recognition is complete on branded questions, so the open-question gap is an evidence problem rather than a name-recognition problem.
Gemini (API, grounded) 93 0 0.0% 21 of 21 Grounded Gemini answers drew on a citation pool that did not include canlah.ai on any open question sampled on September 7, 2026.
Google AI Overviews (rendered, Singapore) 51 0 0.0% 11 of 11 An AI Overview appeared on 62 of 65 rendered Singapore searches, including 11 that named the brand, so the zero reflects source selection rather than a missing feature.

Source: Canlah AI evidence archive, September 7, 2026. Open questions do not name Canlah; branded questions do, are disclosed and are never scored.

“Not named in a sampled answer” means the engine did not write the brand name in that answer on September 7, 2026. It is not proof that the engine will never surface the brand.

Every surface recognised the brand when the question named it. On open questions only ChatGPT named it, in 7 of 84 answers; Gemini and AI Overviews never did. The AI Overview existed on almost every search, so for Google the surface was present, the brand was known and the recommendation was still absent.

The cited sources explain the gap. Across the 216 grounded ChatGPT and Gemini API answers to all 39 questions, 382 domains were cited (381 third-party plus canlah.ai). The most-cited were aistudio.com.sg with 91 citations, mediaplus.com.sg with 85 and stridec.com with 78, mostly agency service pages and “best agency in Singapore” listicles. canlah.ai drew six citations on open questions and ranked 46th by citation count. No badge applies to a domain that is rarely in the evidence set to begin with.

What Singapore SMEs, clinics and schools should do next

  1. Confirm eligibility first: check that the site is a domain or subdomain, is included in Search generative AI features in Search Console and appears in the source preferences tool. Content kept in a subdirectory of a parent brand site cannot be a separate Preferred Source.
  2. Ask existing readers, not strangers: place the “Add to preferred sources” button or deeplink on newsletters, parent updates and patient resource pages. It will mostly reach people who already know the brand.
  3. Publish something others can cite: original, dated material such as fee surveys, waiting-time data, enrolment calendars or local price comparisons, with sample size and method stated. Highly Cited follows references from other publishers, not self-description.
  4. Get onto the pages the engines already read: agency websites, including their service pages and self-published “best agency in Singapore” listicles, supplied most citations in the audit above, led by aistudio.com.sg, mediaplus.com.sg and stridec.com; third-party listicles and directories are the pages a new brand can realistically get onto. Inclusion on those pages matters more than a badge on the brand’s own domain.
  5. Keep regulated claims clean: clinics must stay within the Healthcare Services (Advertisement) Regulations 2021 and the Singapore Medical Council Ethical Code, and private schools within the Private Education Act and the requirements of the Committee for Private Education (CPE) under SkillsFuture Singapore. A widely cited page that overstates outcomes is a liability.
  6. Measure mention, citation and recommendation separately: track the three outcomes by engine, question and date on a fixed question set, keep the raw answers so any figure can be re-checked later and label before-and-after changes as observational.

Where Canlah AI fits

Canlah AI occupies a narrower layer than Google’s own tools. 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. Its role is to fix the buyer question set, record which sources each engine cites and coordinate the on-site and off-site work that puts a brand into that evidence pool. It cannot make a reader select a Preferred Source, and in its September 7, 2026 audit its own site sat outside most Singapore category answers.

Methodology and limitations

Google feature details come from the May 27, 2026 announcement and the Search Central guide, both reviewed on September 23, 2026. Canlah AI figures come from a first-party audit of canlah.ai run on September 7, 2026: 234 planned API answers, 216 of them grounded and 65 completed AI Overview renders. It covers one site, one market, English only, three runs per question on the API engines and two planned browser renders per question for AI Overviews. Google AI Mode, Perplexity, DeepSeek, Qwen and Doubao were not sampled. The data supports measurement recommendations, not claims about Google’s ranking systems.

Frequently asked questions

Is Google Preferred Sources a ranking factor in AI Search?

No. Google describes Preferred Sources as a personalisation setting that labels links from sites a user has already selected. It can make those links more visible to that user inside AI Overviews and AI Mode. Neither the May 27, 2026 announcement nor the Search Central guide describes it as a general ranking factor.

Can a Singapore clinic or school apply for the Google Highly Cited label?

No. Google awards the Highly Cited label to articles that many other stories cite, and the announcement and Search Central guide describe no application process. A clinic or school earns the label only by publishing original material that other publishers reference.

Does being cited in Google AI Overviews mean the brand is recommended?

Not necessarily. In the Canlah AI self-audit of September 7, 2026, every engine recognised the brand on branded questions, yet on open buyer questions ChatGPT named it in 7 of 84 answers, and Gemini (0 of 93) and Google AI Overviews (0 of 51) named it in none. Citation, mention and recommendation should be tracked as three separate outcomes, by engine and by date.

References

Related articles

← All articles Run a free website audit →

FREE WHITEPAPER

Marketing in the Agent Era

All 13 chapters public — no email wall. Includes an original dataset on the agent-readiness of 50 cross-border DTC brands.

Read it free →

KEEP READING