State of AI Search in Singapore 2026

The five numbers

Headline figures cover 14 industries and 280 businesses. Comparisons of who is and is not cited use the 275 whose service the question panel covers; five preschool enrichment providers are left out of those because the panel holds no enrichment questions. Managed IT and cybersecurity is reported separately because the Ahrefs Brand Radar filters we used returned no Singapore buying questions for it (see the scoreboard).

Key findings in two minutes

• Google’s leader is rarely AI’s leader in this sample.
  Read the finding and its scope

  In 11 of the 13 industries with an eligible business result for the selected Google term, a different business has the highest composite citation score. The two exceptions are vet clinics (RehabVet) and accounting and corporate secretarial (Sleek). Of the 12 Google leaders with an AI rank, seven sit clearly below AI’s top ten and one is tied across it.
• Whether AI cites businesses or aggregators depends on the industry.
  Read the finding and its scope

  Pooled across 14 industries, the businesses’ own sites take 33% of 4,544 listed citations, and aggregators take 42% under our classification rules. That average hides a split: businesses take 59% of 1,273 citations in the five health industries and 7% of 1,577 in the three lifestyle industries, where food guides, booking platforms and Google Maps links take most of the citations.
• Ranking in Google’s results does not mean AI cites you.
  Read the finding and its scope

  55 of the 121 businesses that entered the study from Google’s results for their industry’s main search terms, and whose service the question panel covers, had no citation from ChatGPT, Gemini, Perplexity or Google AI Overviews.
• No sampled site’s robots.txt bars an AI search crawler from its homepage.
  Read the finding and its scope

  None of the 279 sites we could assess does so, though most fence off some paths, such as admin pages. 33 of 278 put some barrier up. 27 of 276 refuse a request that identifies itself as GPTBot while serving browsers, a pattern typical of bot-protection settings. 122 of 276 publish llms.txt.
• Domain Rating says little about who gets cited.
  Read the finding and its scope

  The correlation between Domain Rating and AI composite score is r = 0.15 across the 275 businesses whose service the panel covers. The median Domain Rating of the 16 businesses with the highest score in their industry is 24; for the 10 Google leaders that do not share that top score, it is 28.

Google #1 is not AI #1

For each industry we took the highest-ranked business in Google’s recorded organic results for the first main search term and looked up its AI rank, where equal scores share a rank.

Organic rank is the position among organic results for the first search term in Ahrefs’ Singapore SERP snapshot, on the date shown. Where it is not 1, the results above it are government, public, social, aggregator or product-brand sites. ”=” marks a rank shared by businesses with equal scores. NTUC Health is outside the 20 because it is a multi-service co-operative; the highest-ranked physiotherapy business in the 20, Physio & Sole (organic 5), shares AI rank =3 with two other clinics. For the second search term, “chiropractor singapore”, Livewell Chiropractic is first on Google and shares the highest composite score.

Counts of one or two businesses are shown as “fewer than 3” so that no named business can be identified.

Which websites AI answers link to

Across the four scored engines and 14 industries, the top-15 cited-domain lists hold 4,544 citations. Under our classification rules, aggregators (directories, listicles, review, comparison and booking sites, and publishers’ guides) take 42% (1,906). The businesses’ own websites take 33% (1,496), a share that does not depend on those rules. Forums and social links take 10% (454), government 9% (430) and news media 2% (113).

Share of citations in the four scored engines’ top-15 lists, pooled within each segment. Source: e-Alchemists analysis of Ahrefs Brand Radar data, Singapore, November 2026.

Citations are counted only where the domain appears in an engine’s top-15 list. Full lists by engine are in each chapter.

The engines do not agree

100 of the 219 cited businesses (46%) were cited by only one or two of the four scored engines. The source mix differs by engine.

Share of citations in each engine’s top-15 lists, pooled over 14 industries. The last column counts the 275 businesses whose service the panel covers. Source: e-Alchemists analysis of Ahrefs Brand Radar data, Singapore, November 2026.

Consumer vs B2B

Only two B2B industries have usable panels (accounting and corporate secretarial, and recruitment and HR), so this comparison is indicative. The last column leaves out the five preschool enrichment providers the panel does not cover.

In the two B2B industries, AI cites the firms and the regulators (ACRA, MOM) and rarely a directory. The bigger B2B finding is a gap in this Ahrefs IT panel. With the filters we tested, Ahrefs Brand Radar’s Singapore prompt database returned no managed-IT or cybersecurity question with a Singapore or buying-intent word in it. The closest are 7 general questions such as “What are managed IT services?”, and the engines answer them with global vendors. 11 of the 18 Singapore IT firms we tracked had no citation, and seven tie at the top on 8.3 points from a single citation each. Because this extract contains no matching local buying questions, a firm in this sector that wants to assess its AI visibility needs a separate, consistently tracked set of buyer questions.

Are Singapore websites ready for AI crawlers?

Robots.txt rules that bar an AI crawler from the homepage are rare. 10 of 279 businesses have one, and all 10 bar training crawlers (GPTBot, ClaudeBot, Google-Extended or CCBot); none bars a search crawler. Path rules are common: 179 of 256 assessed rechecks, including two sites with no robots.txt, keep at least one AI search crawler out of some paths while leaving the homepage open, which is usual for admin and login pages. The bigger number is the 27 of 276 sites that return an error to a request identifying itself as GPTBot while serving the same page to a browser. That pattern is typical of firewall and bot-protection settings, and we did not ask the owners whether it was intended.

Checks from a Singapore IP on 14 November 2026. Robots.txt rows use a second fetch that day, saved and read with full rule matching; for 23 sites that fetch failed and the first reading stands, and for one site neither fetch succeeded. “Bars from the homepage” means the rules disallow ”/” for that crawler. Path rules that leave the homepage open, usually for admin, login or search pages, are counted separately, only where the second fetch succeeded, and are not treated as a barrier. Five sites could not be fully assessed on the other checks: one was unreachable, one has no website of its own, one refused every automated request, one served a challenge page, and one llms.txt request failed. Unassessed checks are left out of the denominators rather than counted as “no”. A site can treat real crawler traffic differently from a request that only names GPTBot. Including managed IT: 37 of 296 have a barrier and 134 of 294 publish llms.txt.

The 15-industry scoreboard

Score is out of 100 across ChatGPT, Gemini, Perplexity and Google AI Overviews. “Stored answers” counts the answers the four scored engines actually hold; vet, tuition, family law and restaurants each have one question with no stored answer on one engine. ”=” marks a rank shared by businesses with equal scores. “One citation ahead” means the lead rests on a single citation and is not a meaningful difference. Shares are of the listed citations in those four engines’ top-15 lists. “Not cited” means no citation on any of the four scored engines; in preschools and childcare, five enrichment providers are not assessed because the panel holds no enrichment questions. llms.txt counts are out of the sites we could assess. Managed IT is excluded from all headline figures. In tuition, tutor agencies are classed as aggregators. In fitness, one question names Fitness First, which adds to its count. Source: e-Alchemists analysis of Ahrefs Brand Radar data, Singapore, November 2026.

The five-move playbook

See where your business ranks

We measured 298 businesses. Request the measured scorecard for your business if it is included in this study: your citations on each engine, your rank in your industry panel, and your technical readiness checks. If it is not included, we will confirm that and explain what a separate measurement would cover.

See where your business ranks →

Cite this report e-Alchemists. State of AI Search in Singapore 2026. Data measured November 2026 from Ahrefs Brand Radar. https://e-alchemists.com/state-of-ai-search-singapore-2026/ Charts and tables may be reused with a link to this page and the line “Source: e-Alchemists analysis of Ahrefs Brand Radar data, Singapore, November 2026.”

How we measured this

The e-Alchemists team studied 298 Singapore businesses across 15 industries in November 2026. We used industry-specific filters on Ahrefs Brand Radar’s stored Singapore questions. Most panels required local or buying intent (for example “Where is the cheapest dentist in Singapore?”); the exceptions are listed below. We counted the citations to each business’s website in the stored answers from ChatGPT, Gemini, Perplexity and Google AI Overviews. Google AI Mode and Microsoft Copilot were recorded as well, but not scored. Across the 14 headline industries that is 368 questions and 1,468 stored answers from the four scored engines. Four of the 1,472 question-and-engine slots have no stored answer.

Which businesses

In each industry we tracked 20 businesses: every eligible business in Google’s recorded organic results for the industry’s two main search terms, then the businesses AI engines cited most, until we reached 20. Managed IT and cybersecurity has 18, because no further eligible firm was cited. Restaurants and cafés has no Google-side businesses: in Ahrefs’ snapshots for “best restaurants singapore” (28 August 2026) and “cafes singapore” (27 August 2026), none of the recorded organic results was a restaurant or café website, judging by the result titles. Directories, government sites, media and product brands were tracked separately, not ranked.

The score

For each engine, we set a reference equal to the number of questions divided by three, rounded up. The engine score is 25 × citations ÷ reference, capped at 25. We add the four engine scores, for a maximum of 100. The reference uses the number of questions in the panel, so a missing stored answer counts as no citation.

Because of the cap, the composite score is not the same as raw citations. We say “highest composite score” for the score leader and “most-cited” only for the business with the most raw citations across the four engines. The two usually match; in physiotherapy and chiropractic they do not. The score is not a quality rating, and it is not the order in which an engine names or recommends businesses.

Businesses with equal scores share a rank, shown as “=7”. Where a lead rests on one citation or less, we describe a leading group with its counts, not a winner.

Google comparison. Organic rankings for each industry’s main search terms come from Ahrefs’ stored Singapore SERP snapshots. For the first search term, which drives the Google comparison, the snapshots are dated between 28 August and 11 November 2026. Two second-term snapshots are older (recruitment and managed IT, both March 2023).

Technical readiness. We fetched each business’s homepage, robots.txt and llms.txt once on 14 November 2026, checked whether a request identifying itself as GPTBot was refused, and recorded the structured data on the homepage. The same day we fetched every robots.txt again, saved it and read the rules for seven AI crawlers with full rule matching (crawler-specific groups, wildcards and path rules). We use that second reading; the first reading stands only for the 23 headline sites whose file could not be fetched again.

What this is not

These are measured citations from stored AI answers, not modelled or predicted results, and we did not question the AI engines ourselves. Measured, not modelled.

• Small panels. Panels run from 9 to 72 questions per engine in the headline industries. Five of them have fewer than 15 questions (eye clinics 10, physiotherapy and chiropractic 9, vet clinics 12, family law 14, recruitment 10) and are marked indicative. One or two citations either way is not a meaningful difference.
• Ahrefs’ questions, not ours. The questions come from Ahrefs’ stored Singapore database, which is built from questions people search. It leans to price: 178 of the 368 questions (48%) ask about cost, fees or “how much”.
• Host counts, added together. Ahrefs reports citations per web address (host). Where a site appears with and without “www”, or on a subdomain, we add the counts. Ahrefs does not show whether one answer linked to both, so a summed count can include the same answer twice. We checked this answer by answer only for some gyms and salons. In the six industries where we saved Ahrefs’ host lists, summing could overstate the 4,544 total by at most 18; elsewhere it could not be checked. Every citation figure here is a count of listed citations, summed across host variants, not a deduplicated count of answers.
• Cited, not recommended. We count links to a business’s website. Brand Radar does not record the order in which an answer names businesses, and mentions without a link are not counted. A link is not a click or a referral.
• A snapshot. Brand Radar keeps the latest stored answer for each question. The answers in this study were last updated between 19 August and 14 November 2026. AI answers change from day to day.
• Not a random sample. Selection favours businesses that are visible on Google or in AI answers, so this sample cannot estimate the share of all Singapore businesses that receive no citations.
• One domain, one business. Businesses that trade on two domains are counted per domain (for example Kumon in tuition, and one aesthetic doctor’s two sites in skin and aesthetics). Group brands such as the NTUC First Campus preschools are listed separately.
• Classification calls. Aggregators include directories, listicles, review, comparison and booking sites, and publishers’ guides. Two of our rules matter for the aggregator share. Food, lifestyle, beauty and parenting publishers (31 in these lists, from Honeycombers and Eatbook to Sassy Mama) count as aggregators; count them as media instead and the pooled aggregator share falls from 42% to 25.4%, below the businesses’ 33%. Listicles of providers published by sites outside the industry, such as a web-design agency’s “best of” posts, also count as aggregators; count them as “other” and it is 40.5%. The businesses’ 33% does not move under either rule. In tuition, tutor-matching agencies are aggregators, which is why aggregators take 72% of that industry’s 445 citations. Leave tuition out and the pooled aggregator share is 39%. Every domain under each rule is listed in the study data.

Spotted an error about your business? Email admin@e-alchemists.com and we will check it against the data and correct it within five working days.

Appendix

Prompt list by industry

Data tables

Glossary

Corrections and how to request one

Change log