Original research · 1,100 answers · 5 and 6 October 2026
Who ChatGPT and Perplexity recommend in Singapore
We asked both assistants 55 questions about 11 local trades, three times each, then seven more times: 1,100 answers. Across the ten asks, ChatGPT’s first pick changed at least once for 50 of 54 questions. Neither assistant named 12 of the 33 businesses in Google’s top three map results in any of the 1,100 answers.
- questions where ChatGPT’s first pick changed at least once in ten asks (in the other 1, fewer than five of the ten answers named a business)
- 50 of 54
- businesses in Google’s top three map results, as captured by Ahrefs, that neither assistant named in any of the 1,100 answers
- 12 of 33
- of the businesses named for a question that both assistants named, first three runs (140 of 808)
- 17%
- of Perplexity’s citations in the first three runs were “best X in Singapore” lists (1,311 of 3,109)
- 42%
Ask again and the answer moves.
In the first round of three runs, we compared the three answers each assistant gave to the same question. ChatGPT named the same business first in all three for 11 of 51 questions, so its first pick changed for the other 40. Perplexity was steadier, and still changed its first pick for 22 of 53.
Same first pick in all three answers
Businesses named again in another answer to the same question
Top pick: questions where all three answers named at least one business. Named again: for every pair of answers to the same question, the share of one answer’s businesses that the other also named.
Asked ten times.
From 01:20 on 6 October, under two hours after the first three runs ended, we sent all 55 questions to both assistants seven more times with the same code and models: 770 more answers, 1,100 in all. Over ten answers to each question, ChatGPT's first pick changed at least once for 50 of 54 questions and Perplexity's for 35 of 55.
One question, asked ten times in ChatGPT
“Can you recommend a good aircon servicing company in Singapore?”
- Ask 1LM Brothers Aircond
- Ask 2338 Aircon
- Ask 3SoCool
- Ask 4LM Brothers Aircond
- Ask 5Natural Cool
- Ask 6Cool Aircon
- Ask 7Lion City Aircon
- Ask 8Lion City Aircon
- Ask 9SoCool
- Ask 10338 Aircon
The business named first in each answer, in the order the answers came. In ten answers, 6 different businesses came first, none more than twice. Across ChatGPT’s questions, the business that came first most often did so in a median of 5 of the 10 answers.
Ten answers to every question, by assistant
| Measure | ChatGPT | Perplexity |
|---|---|---|
| Questions where the first pick changed at least once | 93%50 of 54 | 64%35 of 55 |
| Answers where the most common first pick came first, median across questions | 5 of 10 | 9 of 10 |
| Different businesses named first, median across questions | 4 | 2 |
| Business and question pairs named in 7 or more of the 10 answers | 11%109 of 1,030 pairs | 49%274 of 561 pairs |
First pick: the first business an answer names. A question counts when at least five of its ten answers named a business, which leaves out ChatGPT’s “maid agency singapore”, where 3 of the 10 did. Answers that named no business are left out of the comparison. The last row counts each business once for each question it was named for; the rest were named in 4 to 6 of the ten answers (ChatGPT 132, Perplexity 86) or in 1 to 3 (ChatGPT 789, Perplexity 201). Every question’s own figures are in results-wave2.json.
How often the most common first pick came first, by trade
| Trade | ChatGPT | Perplexity |
|---|---|---|
| Tuition centres | 70% | 100% |
| Lawyers | 63% | 100% |
| Movers | 60% | 90% |
| Accounting | 60% | 90% |
| Car workshops | 50% | 70% |
| Physiotherapy | 50% | 80% |
| Aesthetic clinics | 40% | 60% |
| Dental clinics | 30% | 60% |
| Interior design | 30% | 70% |
| Aircon servicing | 30% | 100% |
| Maid agencies | 30% | 80% |
| All questions | 50% | 90% |
Each figure is the median, over the trade’s five questions, of the share of a question’s answers in which its most common first pick came first. The last row takes the median over all 54 ChatGPT and 55 Perplexity questions: the 5 and 9 of 10 in the table above. ChatGPT’s maid agencies figure covers 4 questions, as “maid agency singapore” is left out.
Named in the most answers in each trade, out of 100
| Trade and its most-named business | Of 100 | ChatGPT | Perplexity |
|---|---|---|---|
| MoversMove Move Movers | 78 | 38 | 40 |
| Tuition centresThe Learning Lab | 76 | 36 | 40 |
| PhysiotherapySingapore Physio | 71 | 31 | 40 |
| Aesthetic clinicsSL Aesthetic Clinic | 56 | 17 | 39 |
| Car workshops1 Car Service | 52 | 30 | 22 |
| Dental clinicsNational Dental Centre Singapore | 51 | 31 | 20 |
| Maid agenciesNation Maid Agency | 48 | 29 | 19 |
| AccountingSleek | 45 | 8 | 37 |
| Aircon servicingSoCool | 42 | 22 | 20 |
| LawyersAllen & Gledhill | 37 | 18 | 19 |
| Interior designSpace Atelier | 29 | 1 | 28 |
Each trade had 100 answers over the ten runs: five questions, ten runs, two assistants. The ChatGPT and Perplexity columns are out of 50 answers each. No trade had a tie for first place. The three most named in each trade, split the same way, are in results-wave2.json.
Mentions taken by each trade’s three most-named businesses
Businesses named in a single answer
A mention is one business named in one answer. The first three runs named 541 different businesses, and all ten named 911.
Google’s map results over ten runs. A check added after the wave 2 frame was written. With the same name matching as the first three runs, 14 of the 33 businesses in Google’s top three map results were not matched to a name in any of the 1,100 answers, against 15 of 33 in the first 330. The one that changed is M-Aesthetic Clinic, which Perplexity named in 7 answers in runs 4 to 10. A search of the text of all 100 answers in each trade then found 2 of the 14 under other names: ChatGPT named Ashford Dental Centre Thomson as “Emergency Dental by Ashford Dental” in 2 answers, and Perplexity named Koh Mgmt as “Koh Management” in 1 answer. That leaves 12 of 33 that neither assistant named in 1,100 answers. A search of the text of all 1,100 answers, in every trade, found none of those 12. Each trade’s matches are in results-wave2.json, and notes.md records the search of every trade and the reasoning.
The two assistants rarely name the same businesses.
For each question we pooled each assistant's three answers and compared the two lists. Added up over the 55 questions, the lists held 808 businesses and 140 were on both. Comparing single answers instead (same question, same run), the overlap was 16%.
Businesses named by both, out of all named, by trade
They cite different things.
Perplexity's answers carried a median of 19 citations and ChatGPT's a median of 5. The bigger difference was what those citations pointed to.
| Measure | ChatGPT | Perplexity | Google page 1 |
|---|---|---|---|
| Business websites | 57%460 | 23%713 | 53%50 |
| “Best X in Singapore” lists | 9%73 | 42%1,311 | 14%13 |
| Directories and review sites | 10%83 | 19%606 | 11%10 |
| News and magazines | 2%20 | 10%326 | 3%3 |
| Government | 19%150 | 2%57 | 3%3 |
| Forums and social media | <0.5%2 | 1%42 | 13%12 |
| Other | 2%15 | 2%54 | 3%3 |
Share of each one’s citations, by type, counting every citation. 3,912 citations in 330 answers. Google page 1: the organic top 10 for each trade’s T3 keyword, labelled with the same rules (94 results). Counting each page once instead, lists were still Perplexity’s largest category (228 of 660 pages, 35%) and business websites still ChatGPT’s (316 of 512, 62%).
Perplexity’s answers cite lists. 1,311 of its 3,109 citations were “best X in Singapore” lists. Magazines and news sites published 768 of them, directories and review sites 369, businesses in the same trade 105 and other sites 69.
ChatGPT’s answers cite business websites, and the government. 460 of its 803 citations were the websites of businesses in the trade, and 150 were government sites, most often the Ministry of Manpower and the Ministry of Health.
Ahrefs recorded no AI Overview in its captures of Google Singapore for any of the 11 searches, and the three we checked live on 5 October 2026 showed sponsored results and a business map with no AI Overview.
Google's top three didn't carry over.
Neither assistant named 12 of the 33 businesses in Google's top three map results in any of the 1,100 answers. In the first round of three runs, 15 of the 33 were not named in their trade's 30 answers, as the table shows. Counting each trade's cited websites separately in those three runs, 52 of 681 (8%) were on Google's first page for that trade's T3 keyword.
| Trade and search | Google’s top three on the map | Named by |
|---|---|---|
| Aircon servicing “aircon servicing singapore” | SoCool Pte Ltd - Aircon Servicing Singapore | ChatGPTPerplexity |
| Affordable Aircon Services Pte Ltd | ChatGPTPerplexity | |
| Newway Air Conditioning | Neither | |
| Dental clinics “dental clinic singapore” | Ashford Dental Centre Thomson | Neither |
| B9 Dental Centre (Toa Payoh) | Perplexity | |
| Denticare - Thomson | Perplexity | |
| Interior design “interior design singapore” | Homevera Interior Design Singapore | Neither |
| MyDesign Interiors Pte Ltd | Neither | |
| Flo Design Pte. Ltd. | Neither | |
| Movers “movers singapore” | Soon Seng Transport & Movers Pte Ltd | ChatGPTPerplexity |
| LS House Movers Singapore | Neither | |
| Cheap Movers Singapore - Office, House, Piano & Furniture | Perplexity | |
| Aesthetic clinics “aesthetic clinic singapore” | Become Aesthetics Clinic (Novena) | ChatGPTPerplexity |
| Dr D Aesthetics Medical Clinic | ChatGPT | |
| M-Aesthetic Clinic | Neither | |
| Physiotherapy “physiotherapy singapore” | The Rehab Lab Plus | ChatGPT |
| Specialist Physio Jonathan, Central Singapore | Neither | |
| City Osteopathy and Physiotherapy @ Bukit Timah | ChatGPT | |
| Tuition centres “tuition centre singapore” | New Dawn Learning Studio | Neither |
| EduFirst Learning Centre (Ang Mo Kio AMK) | ChatGPTPerplexity | |
| Kiasu Maths and Science Tuition Centre (Ang Mo Kio AMK) | Neither | |
| Maid agencies “maid agency singapore” | Ministry of Helpers | ChatGPT |
| Universal Employment Agency Pte Ltd | ChatGPTPerplexity | |
| Best Home Employment Agency (Thomson) | ChatGPTPerplexity | |
| Accounting “accounting services singapore” | AG Singapore | Perplexity |
| Koh Mgmt | Neither | |
| One Accounting Services Singapore | Neither | |
| Lawyers “lawyer singapore” | Tembusu Law LLC | Neither |
| PKWA Law Practice LLC | ChatGPTPerplexity | |
| Regal Law LLC | Neither | |
| Car workshops “car servicing singapore” | 1 Car Service | ChatGPTPerplexity |
| Carro Care (Sin Ming) - Workshop | Neither | |
| Garage R Performance Centre Pte Ltd | ChatGPTPerplexity |
Map results: Google Singapore’s local pack for each keyword, from Ahrefs, captured between 11 August and 5 October 2026 (dates in google-serps.json). “Neither” was checked by searching the text of all 30 answers in that trade for the business’s name.
Linked websites on Google’s first page, by trade, first three runs
| Trade | ChatGPT | Perplexity |
|---|---|---|
| Physiotherapy | 19%6 of 31 | 13%4 of 31 |
| Dental clinics | 17%5 of 30 | 9%4 of 43 |
| Tuition centres | 15%4 of 26 | 13%5 of 40 |
| Car workshops | 10%4 of 40 | 10%4 of 39 |
| Lawyers | 10%2 of 21 | 12%4 of 33 |
| Movers | 8%2 of 25 | 7%3 of 46 |
| Maid agencies | 6%1 of 16 | 13%6 of 48 |
| Interior design | 5%2 of 42 | 11%4 of 38 |
| Accounting | 5%2 of 43 | 2%1 of 52 |
| Aircon servicing | 4%1 of 27 | 15%6 of 39 |
| Aesthetic clinics | 0%0 of 37 | 6%2 of 36 |
| All 11 trades | 9%29 of 338 | 10%43 of 445 |
These counts come from the first three runs, 15 answers per trade from each assistant. Each website counts once per trade, and as on Google’s first page when it was in the top 10 for that trade’s T3 keyword. A website both assistants linked to counts in both columns, so together they come to more than the 52 of 681 above.
Over all 10 runs (550 answers from each assistant), ChatGPT linked to 601 websites counted once per trade (535 different websites), and 39 were on Google’s first page (6%). Perplexity: 46 of 459 (10%). For aircon servicing, ChatGPT linked to 49 websites and 1 was on the first page. We counted this on 16 October 2026, after both run plans were written and in neither of them, with the same rule and the same Google results; every answer it uses is in the downloads below.
Who got named most.
In the first three runs, each trade had 30 answers: five questions, three runs, two assistants. A few businesses came up again and again: the three most named in each trade took 443 of the 1,753 mentions (25%), while 233 of the 541 businesses named (43%) appeared in a single answer. A mention is one business in one answer. Ties for third place are all listed.
| Trade | Named in the most answers, out of 30 | Different businesses named |
|---|---|---|
| Aircon servicing32 different businesses named | Cool Aircon (12), Gain City (12), SoCool (12) | 32 |
| Dental clinics45 different businesses named | National Dental Centre Singapore (17), Nuffield Dental (11), Advanced Dental (7), TSD Dental Group Tampines (7) | 45 |
| Interior design95 different businesses named | Space Atelier (10), Dyel Design (9), Swiss Interior (8) | 95 |
| Movers29 different businesses named | Move Move Movers (24), Cube Movers (22), Shalom Movers (15) | 29 |
| Aesthetic clinics49 different businesses named | SL Aesthetic Clinic (16), The Clifford Clinic (16), Edwin Lim Medical Aesthetic Clinic (7), V Medical Aesthetics & Laser Clinic (7) | 49 |
| Physiotherapy38 different businesses named | PhysioActive (21), Core Concepts (18), Singapore General Hospital (11) | 38 |
| Tuition centres44 different businesses named | The Learning Lab (24), AGrader Learning Centre (15), Mavis Tutorial Centre (15) | 44 |
| Maid agencies33 different businesses named | Island Maids (15), Universal Employment Agency (13), JForce Employment Service (11), Nation Maid Agency (11) | 33 |
| Accounting63 different businesses named | 3E Accounting (13), Sleek (13), PwC Singapore (11) | 63 |
| Lawyers55 different businesses named | Allen & Gledhill (10), Legal Aid Bureau (10), Pro Bono SG (8), Rajah & Tann (8), WongPartnership (8) | 55 |
| Car workshops58 different businesses named | 1 Car Service (16), Precise Auto Service (12), Revol Carz Garage (9) | 58 |
What this means if you run a Singapore business.
Being named once is luck. Being named in most answers is the goal. In every trade a few businesses came up again and again: Move Move Movers and The Learning Lab were each named in 24 of their trade’s 30 answers, while 43% of the businesses named appeared just once. The way to know where you stand is to ask the same questions many times, in more than one assistant, and count.
Perplexity’s answers cite the lists for your trade. 42% of its citations were best-of lists, and 87% of those came from magazines, directories and review sites. A business that is not on those lists is missing from the pages Perplexity cited most.
ChatGPT’s answers cite business websites. 57% of its citations were the sites of businesses in the trade: price pages, branch pages, service pages. What your own site says in plain words is part of what it cites.
Google rank is a different race. 15 of the 33 businesses in Google’s top three map results were not named in the first three runs, and 12 were never named in all 1,100 answers. In those first three runs, 92% of the sites the assistants cited were not on Google’s first page for their trade’s T3 keyword.
Check your own business in ten minutes.
A small version of the study, run on your own business in the ChatGPT and Perplexity apps your customers use.
- 1Ask what your customers ask: the question they would type, such as “best aircon servicing near Tampines”.
- 2Start a fresh or temporary chat for every ask, because your own chat history can change the answer.
- 3Ask the same question 10 times and count how many of the answers name you.
- 4Read the sources linked under each answer, the pages the assistant read before it picked.
- 5Get onto those pages: ask the writer of a “best of” list to add you, get listed on review sites and directories and ask customers for reviews, and give your own website a page that answers the question.
- 6Repeat the check in Perplexity, which leans on “best of” lists: they were 42% of its citations.
How the study was done.
We wrote the design down first: frame.md, saved at 23:18:09, 34 seconds before the first question was sent at 23:18:43. The time typed in its header (23:20) was an estimate; notes.md records that and everything decided afterwards, with the reason.
The questions
The 11 trades are the ones in our July and August website scan below, kept so the two studies sit side by side. Each trade got the same five question templates, 55 questions in all. One template is a real search typed as is: of the 44 candidate keywords we checked in Ahrefs, the highest-volume one containing “singapore” for that trade. The location is in the words because the assistants were asked with no location set.
| Template | Question |
|---|---|
| T1 | Can you recommend a good {provider} in Singapore? |
| T2 | What are the best {providers} in Singapore? |
| T3 | {keyword}: the highest-volume of the candidate keywords containing “singapore”, typed as is |
| T4 | I need an affordable {provider} in Singapore. Who should I go to? |
| T5 | I live in Tampines. Which {provider} near me would you recommend? |
| Trade | Keyword (T3) |
|---|---|
| Aircon servicing | aircon servicing singapore |
| Dental clinics | dental clinic singapore |
| Interior design | interior design singapore |
| Movers | movers singapore |
| Aesthetic clinics | aesthetic clinic singapore |
| Physiotherapy | physiotherapy singapore |
| Tuition centres | tuition centre singapore |
| Maid agencies | maid agency singapore |
| Accounting | accounting services singapore |
| Lawyers | lawyer singapore |
| Car workshops | car servicing singapore |
The assistants
ChatGPT here means OpenAI's gpt-5.6-luna with OpenAI's own web search, through the API. Perplexity means Perplexity's sonar, which searches Perplexity's own index, through the API. Both were called through OpenRouter, every question alone, with no system prompt, no history and no location. The first 330 answers arrived on 5 October 2026 between 23:18 and 23:28 Singapore time. This is not the ChatGPT or Perplexity app: the apps add their own instructions, memory and your location, and may use another model. The API is the version anyone can rerun.
How answers were coded
A small model (openai/gpt-5-nano, no web access) listed the businesses each answer names. A name was kept only if that exact text appears in the answer, so the model could miss a name but never add one. We then read 20 answers in full against their lists: 99 names, none missed, none invented. Names were matched across answers after lowercasing and dropping “Pte Ltd”, “Singapore” and punctuation, counting a branch name (“TP Dental Surgeons, Orchard”) as the same business. The looser matching rule in the frame moves the headline figures by two points at most: ChatGPT’s same-first-pick rate is 23.5% under it against 21.6% here.
Every cited link got one label by fixed rules: forums and social media, government, best-of lists (a “best” or “top” list in the title or address), then the website’s own label. Website labels were drafted by a model and every one was read by hand; the full table is in domain-labels.json.
What this can still get wrong
One evening’s answers: they change, which is partly the finding. Per-trade figures rest on 30 answers each and are descriptive. A business called two different names in two answers counts as two, which pushes agreement down. One answer listed a building (“Carros Centre”) as a workshop. Google’s captures are from Ahrefs and not all from the same day. English only.
The run cost US$5.02 in API calls. We sell SEO and AI search work to Singapore businesses, which is why we ran it; nothing in the method depends on who we are, and the raw answers are below for anyone who wants to check.
Seven more runs
We wrote down what the extra runs would add before sending them: frame-wave2.md, saved at 01:19:47 on 6 October, 17 seconds before the first of them was sent at 01:20:04. Runs 4 to 10 used the same code, the same models and the same request for every question, in rounds as before, and the names were coded the same way. All 770 answers arrived between 01:20 and 01:41 Singapore time, and the API calls cost US$12.02.
What the extra runs can still get wrong: all ten runs fell on one night, so they show how answers move from one ask to the next and say nothing about change over days. Ten answers per question is still a small sample, and the figures for any one question are descriptive. The hand check of 20 answers covers runs 1 to 3. The limits above still apply: the assistants were asked through their APIs, in English, and the Tampines question names one town. The map-pack check and its text search were added after frame-wave2.md was saved, and notes.md records both. Websites first cited in runs 4 to 10 have not been labelled yet (320 of 9,142 citations); no figure on this page uses them, and the transcript marks them Not labelled.
The data
- answers-full.jsonl Runs 1 to 3: every answer in full, with the names and labelled sources coded from it
- answers.csv Runs 1 to 3: one row per answer, with question, engine, run, time and businesses named
- citations.csv Runs 1 to 3: one row per cited source, with its label
- questions.csv The 55 questions, exactly as sent
- google-serps.json Google's map results and top 10 for each trade, with capture dates
- domain-labels.json The label given to every cited website
- frame.md The study design, saved before the first question was sent
- notes.md What changed after that, and why
- results.json Every figure from runs 1 to 3, with its numerator and denominator
- answers-full-wave2.jsonl Runs 4 to 10: every answer in full, with its names and sources (sites first cited in these runs are marked Not labelled)
- answers-wave2.csv Runs 4 to 10: one row per answer
- frame-wave2.md The plan for runs 4 to 10, saved before they were sent
- results-wave2.json Every ten-run figure, each question's first picks, and the map-pack text search
- results-page-one-all-runs.json Google's first page over all ten runs, by trade and assistant (counted 16 Oct 2026, after both plans)
Cite as: rankr (2026). Who ChatGPT and Perplexity recommend in Singapore, 1,100 answers collected 5 and 6 October 2026. https://rankrhq.com/singapore-sme-seo-benchmark. Free to reuse under CC BY 4.0. The charts on this page are free to reuse on the same terms, credited to rankr.
SEO performance benchmarks: 71 Singapore websites on page one.
Our earlier study set SEO performance benchmarks for Singapore SMEs. On 19 July and 30 August 2026 we ran 74 Singapore business websites, all on Google's first page for an ordinary buyer search, through our free checker. 71 returned a result. The checker scores 15 on-page signals.
- median technical score
- 92
- mean technical score
- 88
- scored 90 or above
- 59%
- lowest and highest score
- 45 to 100
| Industry | Sites | Median | No question headings or FAQ markup | Scanned |
|---|---|---|---|---|
| Aircon & home servicesScanned 19 Jul 2026 | 7 | 99 | 0 of 7 | 19 Jul 2026 |
| MoversScanned 30 Aug 2026 | 5 | 99 | 0 of 5 | 30 Aug 2026 |
| Accounting & corporate servicesScanned 19 Jul 2026 | 6 | 92 | 3 of 6 | 19 Jul 2026 |
| Tuition centresScanned 30 Aug 2026 | 6 | 93 | 1 of 6 | 30 Aug 2026 |
| Aesthetic clinicsScanned 30 Aug 2026 | 5 | 90 | 3 of 5 | 30 Aug 2026 |
| Renovation & interior designScanned 19 Jul 2026 | 7 | 91 | 1 of 7 | 19 Jul 2026 |
| Law firmsScanned 19 Jul 2026 | 4 | 87 | 3 of 4 | 19 Jul 2026 |
| PhysiotherapyScanned 19 Jul 2026 | 8 | 93 | 3 of 8 | 19 Jul 2026 |
| Dental clinicsScanned 19 Jul 2026 | 10 | 88 | 5 of 10 | 19 Jul 2026 |
| Car workshopsScanned 30 Aug 2026 | 5 | 78 | 5 of 5 | 30 Aug 2026 |
| Maid agenciesScanned 30 Aug 2026 | 8 | 90 | 3 of 8 | 30 Aug 2026 |
| Check | Not passing |
|---|---|
| llms.txt (emerging) | 63% |
| Question headings or FAQ markup | 38% |
| Share image (og:image) | 34% |
| Single clear H1 | 27% |
| Title tag | 20% |
| Meta description | 20% |
| Business schema | 17% |
| robots.txt lets crawlers in | 15% |
| XML sitemap | 10% |
| Structured data (JSON-LD) | 8% |
| Open Graph tags | 7% |
| Name, address, phone | 7% |
| Canonical URL | 3% |
| Content depth | 3% |
| Served over HTTPS | 0% |
Every site came from page one of Google for an ordinary buyer search, so this is a sample of your visible competitors, not of the average Singapore SME. The two scan dates covered different industries, so the two medians are not a trend. After llms.txt, the most common miss was “Question headings or FAQ markup”, on 38% of sites.
Ignore the llms.txt row: no major AI search system has been shown to read third-party llms.txt files (the evidence). The robots.txt row uses the rule the checker applied at scan time, which also failed a missing robots.txt file; rechecked on 3 October 2026, none of the 11 sites it failed blocks Googlebot, Bingbot or an AI search crawler from its homepage. Run the same checker on your site.
Questions about the data.
Does ChatGPT recommend the same Singapore business every time you ask?
No. We asked the same 55 questions three times each on 5 October 2026. ChatGPT's first-named business was the same in all three answers for 11 of 51 questions (22%), so its top pick changed for the other 40 (78%). Perplexity kept its top pick for 31 of 53 (58%). Over ten asks of each question, ChatGPT's first pick changed at least once for 50 of 54 questions, and Perplexity's for 35 of 55.
Do ChatGPT and Perplexity recommend the same businesses?
Rarely. For each question we pooled each assistant's three answers and compared the two lists. Added up over the 55 questions, the lists held 808 businesses, and 140 of them (17%) were on both. Interior design had the least overlap and movers the most.
What do ChatGPT and Perplexity cite when they recommend Singapore businesses?
Different things. 42% of Perplexity's 3,109 citations were "best X in Singapore" lists. 57% of ChatGPT's 803 citations were the websites of businesses in the trade, and 19% were government sites such as the Ministry of Manpower and Ministry of Health.
Are the businesses in Google's top map results the ones AI names?
Often not. Of the 33 businesses in Google's top three map results for the 11 trades, as captured by Ahrefs, 15 were not named by either assistant in their trade's 30 answers. Over all 1,100 answers, 12 of the 33 were never named.
Does Google show AI Overviews for local service searches in Singapore?
Not in what we saw for the 11 searches in this study. Ahrefs recorded no AI Overview in its captures of Google Singapore for any of them, and the three we checked live on 5 October 2026 showed sponsored results and a business map with no AI Overview.
What is a good technical SEO score for a Singapore business website?
In our July and August scan of 71 Singapore businesses already on Google's first page for buyer searches, the median was 92 out of 100 and 59% scored 90 or above. A score below 70 put a site in the bottom 10% of this already-visible sample.
Can I use this data?
Yes. Both studies are free to cite and reuse under CC BY 4.0. Link to this page so readers can check the method; the raw files for the AI study are in its data section.