AI Search Readiness Singapore: 1 in 4 Pages Pass

You search "plumber Tampines" or "dentist Sengkang", and a page answers. Ask an AI assistant the same thing and it names a few businesses, maybe not yours. A Singapore restaurant owner asked on Reddit why AI doesn't recommend their Korean restaurant. So in 2026 we ran a study of AI search readiness Singapore-wide: 109 local pages across 5 regions.

The gap is on the page, not at the door. On 13 September 2026 we checked 109 provider pages that appeared for local searches in all five regions, according to our metasearch sample. Just 28 (26%) had an answer-first opening, question headings and business schema together. Not one of the 69 local-service websites blocked the five AI crawlers we tested.

The crawler number surprised us. Blocked crawlers aren't what's holding these pages back. The misses we'd fix first are words on the page, and one of them hides in the very first paragraph. We show exactly what it looks like in the section on direct answers below.

Key takeaways on AI search readiness Singapore-wide

Every figure comes from a metasearch sample taken with SingRank's self-hosted SearXNG on 13 September 2026 (blended Bing/Yandex results, not Google's exact ranking). Percentages are rounded and always shown with their count.

#	Finding	The number
1	Provider pages with an answer-first opening paragraph	65 of 109 (60%)
2	Pages with question headings or an FAQ	57 of 109 (52%)
3	Pages with business-entity schema	84 of 109 (77%)
4	Pages passing all three on-page checks	28 of 109 (26%)
5	Websites with an llms.txt file	24 of 69 (35%)
6	Websites whose robots.txt blocks GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot or ClaudeBot	0 of 69
7	Sampled on-topic slots held by directories, listicles or social pages	58 of 181 (32%)
8	"Local SEO agency" pages that mention GEO, AEO or AI search	11 of 23 (48%)

If you only read one row, read row 4.

What we checked, and why these five towns

To measure AI search readiness Singapore-wide, we ran 25 searches: the same five phrases in one large residential town per region. The regions are the ones set out in URA's Master Plan, Singapore's statutory land use plan. Bukit Merah stood in for Central, Jurong West for West, Woodlands for the North, Sengkang for the North-East and Tampines for the East.

The phrases were "aircon servicing", "plumber", "dentist" and "tuition centre" plus the town. Google Autocomplete returns real town-level versions, such as "plumber sengkang price list" and "aircon repair tampines". A fifth phrase, "local seo agency in" plus the town, checked the firms that sell AI-search work.

From each search we kept the first 10 results about the service in Singapore. Then we fetched each provider page without running JavaScript, because, as our technology notes put it, "AI crawlers do not run JavaScript."

Our earlier Singapore Search Visibility Study 2026 looked inward at our clients' Search Console data. This one looks outward, at other businesses' public pages, so the two give different Singapore search insights.

Are AI crawlers locked out of local websites?

No. Across 69 provider websites, robots.txt blocked none of GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot or ClaudeBot. For the local services we sampled, the door is open.

That matters, because the assistants explain what these crawlers are for. OpenAI's crawler documentation describes OAI-SearchBot as the bot that surfaces websites in ChatGPT's search features. Perplexity's documentation describes PerplexityBot as the bot that surfaces and links sites in its answers, not a training crawler.

Don't read "robots.txt is open" as "crawlers get in", though. Google also tells site owners to make sure nothing between the crawler and the page, such as a CDN or the host, stops the crawl.

We've seen access problems on our own site. Our method page records a 12 August 2026 review that found AI crawler requests to singrank.com not getting through, and comparing them with our server's responses pointed to a specific access problem. So the 0 of 69 is a floor, not a clean bill of health. We checked robots.txt only; an SEO audit is where access gets tested properly.

How many pages open with a direct answer?

According to our sample, 65 of 109 provider pages (60%) opened with a paragraph that names the service within 120 words. The other 44 didn't, and 40 of those 44 missed for one reason: the first real paragraph never says what the business does.

Our check is literal on purpose. It reads the first paragraph of 15 words or more, outside the header, menu and footer. An opening that lists servicing and chemical wash for a town misses if the word aircon sits only in the heading.

A pattern kept repeating when we read the misses by hand. 85 of the 109 pages were town pages, with the town in the web address.

Many opened by describing the town: its region and the areas bordering it. Fine for geography. No help to someone asking who services aircon in Jurong West.

Here's the stake. If your town page spends its first paragraph on the town, the paragraph a machine is most likely to quote says nothing about you. That one habit cost 40 of the 44 pages their pass.

How many pages write their headings the way buyers ask?

Per the same sample, 57 of 109 pages (52%) had at least one H2 or H3 phrased as a question, or an FAQ block. That makes it the weakest of the three on-page checks.

Question headings do two jobs at once. A buyer scanning the page finds the question they arrived with. A system looking for a passage finds a clean question-and-answer pair it can quote without guessing where the answer starts.

You don't need schema for this part. Google's guidance on AI features says you don't have to create special machine-readable files or extra markup to appear in them. Visible questions with real answers underneath are simply good page structure, and a tuition centre or a dental clinic can add them this afternoon.

Is business schema the part most sites already have?

Yes. According to our sample, 84 of 109 pages (77%) carried business-entity structured data, the most common of the three signals.

68 declared a LocalBusiness-type entity, the type schema.org reserves for a physical business or a branch of one. Another 16 declared only a generic Organization. The remaining 25 declared no business entity at all.

Having schema is the easy part. The catch is accuracy. Google lists structured data that matches the visible text of the page among its fundamentals. In our view, schema that gives one address while the footer shows another is worse than a gap, because it hands a machine two answers.

Our guide to schema markup for AI search covers the types in detail. For this sample the lesson is simpler: schema is the signal Singapore's local pages already have most often, so it's the smallest of the three on-page gaps.

Just 28 of 109 pages pass all three checks

Put the three on-page checks together and the number drops to 28 of 109 (26%), the figure cited at the top of this page. Only 6 pages (6%) missed all three. Most pages sit in the middle, doing one or two of the three things right.

Add the two website-level signals, llms.txt and an open robots.txt, and every page gets a score from 0 to 5:

Score (0–5)	Provider pages	Share
5	17	16%
4	33	30%
3	29	27%
2	24	22%
1	6	6%
0	0	0%

Read that table as good news if you own a local business. 81 of the 109 pages fall short of the full on-page set, and for 56 of those 81 the missing piece is text: an opening paragraph or a question heading. That's an edit, not a rebuild, and nobody needs a developer for it.

How Singapore's five regions compare

Here are the same counts split by region. The samples are small, so treat the differences as a description of 13 September 2026, not a league table.

Region (town searched)	On-topic slots: provider / third-party	Pages passing all three	Sites with llms.txt
Central (Bukit Merah)	17 / 21	7 of 15	7 of 15
West (Jurong West)	27 / 13	8 of 24	11 of 24
North (Woodlands)	18 / 6	4 of 16	9 of 15
North-East (Sengkang)	29 / 10	7 of 29	10 of 28
East (Tampines)	32 / 8	4 of 31	8 of 30

One difference was large enough to stand out. In Bukit Merah, directories, listicles and review sites held 21 of 38 on-topic slots (55%). In Tampines they held 8 of 40 (20%), per the table above. For "dentist bukit merah", a provider's own site held just 1 of the 9 on-topic slots.

That's a real stake for a Central business. When a directory holds the slot, its description of you may be the first page a buyer, or a machine, reads. Across those 21 Bukit Merah slots, you risk being judged on a listing you didn't write, so check what those listings say.

The pass rates by region (47%, 33%, 25%, 24% and 13% from Central to East, source: the table above) look like a ranking. They aren't one. With 15 to 31 pages per region, a single page more or less shifts a region's figure by 3 to 7 points.

Which services are furthest behind?

Split by service, dentists did best and aircon servicing did worst on the full three-check set, though each service had fewer than 40 pages:

Service searched	Pages passing all three
Dentist	6 of 16 (38%)
Plumber	10 of 30 (33%)
Tuition centre	6 of 28 (21%)
Aircon servicing	6 of 35 (17%)

Aircon servicing had the most provider pages in the sample, and 9 of the 22 aircon websites had separate pages for 3 or more of our 5 towns. When a shared template opens with a town description, every town page inherits the same miss, and a site with 4 town pages risks losing the same check 4 times over. If that's your setup, fix the template once and every town page improves together, instead of editing 4 pages one by one. Our guides for aircon servicing companies, plumbers, dentists and tuition centres go deeper on each trade.

What does llms.txt tell you about a site?

24 of 69 provider websites (35%) had an llms.txt file. The llms.txt proposal suggests a markdown file on your site that offers LLM-friendly content: a short summary and links to key pages.

Two facts keep that number in proportion. Across both groups we sampled, 31 websites had the file, and 6 of those 31 opened with a "Generated by" line from an SEO plugin. On some sites the file exists because a plugin produced it, not because anyone decided what it should say.

The second fact is Google's position. Its guidance says AI Overviews and AI Mode have no extra requirements and need no special optimisation, a point our Google AI Overviews guide builds on. Our GEO page makes the same point in plain terms: an llms.txt file can summarise key pages, but Google's AI search features don't require one. Our llms.txt guide for Singapore businesses explains when it's still worth adding.

Are the agencies selling AI search ready themselves?

We ran "local seo agency in" plus each town as a fifth search. It produced 35 on-topic slots, 30 of them agency pages, and 23 we could measure.

According to the sample, 11 of those 23 pages (48%) mentioned GEO, AEO or AI search. Fewer, 9 of 23 (39%), passed all three on-page checks. 7 of the 14 agency websites (50%) had llms.txt. And 2 of the 14 (14%) blocked at least one of the five AI crawlers in robots.txt, the only blocks anywhere in the sample.

4 of the 17 agency websites had separate pages for 3 or more of our 5 towns, often near-identical apart from the town name. A templated town page tells you little about how an agency's client work reads to a machine.

So don't hire on the words "AI search" in a page. Open the page, run the five checks below and see whether it follows its own advice.

Why Woodlands businesses should write "Woodlands, Singapore"

Of the 120 results returned for our five Woodlands searches, 31 (26%) were about somewhere else called Woodlands, mostly The Woodlands in Texas (source: our hand-tagged sample). For "plumber woodlands", just 2 results were Singapore plumbers.

A place name shared with a town on another continent gives every search system one more thing to untangle. The fix costs nothing: write "Woodlands, Singapore" in your opening paragraph, your page title and your address block. Our local SEO guide for Singapore covers the Maps side.

What the numbers mean for a local business

Our verdict first: fix the first paragraph before you touch llms.txt. In this sample the open door was universal, schema was common, and the biggest on-page gaps were words: question headings (absent on 52 of 109 pages) and a plain opening that says what you do and where (absent on 44).

Some readiness checklists put llms.txt and FAQ schema near the top of the job list. We see why: both are quick to add and easy to tick.

But Google says its AI answers need no AI text files or special schema, and 40 of the 44 answer-first misses in our sample were about words, not files. A file can't make a vague page specific. Neither file hurts, but neither rescues a page whose first paragraph talks about something else. We explain the citation side in why AI search ignores some brands.

What AI SEO for local business owners should fix first

Three calls we'd make for a Singapore service business:

Rewrite the opening of every town page. Name the service and the town in the first sentence, then the answer a buyer needs. Skip the paragraph about the town.
Choose questions for your headings from real enquiries. Put the answer in the first two sentences underneath. Add FAQ markup later, if at all.
Never block OAI-SearchBot or PerplexityBot if you want to appear in those assistants' search features.

And if a directory holds your town's search, as it did for most of Bukit Merah's slots, check your listing before you rebuild your site.

A five-step check you can run on your own page

Pick the one page that should win your town's search. Then work through these steps, in order:

Read the first paragraph. Does it name your service and your town in under 120 words? If not, rewrite it.
Scan the headings. Is at least one H2 or H3 a question a customer actually asks, with the answer right under it?
View the page source. Search for "LocalBusiness" and check its name, address and phone match the page.
Open /robots.txt on your site. Look for GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot and ClaudeBot. A "Disallow: /" under any of them blocks it.
Open /llms.txt on your site. If a plugin wrote it, check it lists your real service pages. A missing file isn't the first fix.

Then ask the assistants themselves. Our AI search optimisation Singapore visibility test shows how to run it without fooling yourself, and how to compare results after you change the page.

Method and limits — read these before quoting

We'd rather you know exactly what this sample can and can't carry.

Metasearch is not Google. Results came from SingRank's self-hosted SearXNG on 13 September 2026. Most of its engines returned errors that day and Bing returned unrelated pages, so we kept the first 10 on-topic results from up to 30, a change logged before scoring. Of 557 results read, 340 (61%) were off-topic, according to our method file.
One day, one phrase per service per town. A snapshot, not a trend.
Small numbers. 109 provider pages in total, 15 to 31 per region.
Checks are signals, not proof. Passing them doesn't mean an assistant will cite a page.
Raw HTML and robots.txt only. JavaScript content, firewalls and CDN blocking weren't measured.
One exclusion. A page from a business SingRank has worked for was replaced with the next result.

A second fetch of all 132 measurable pages later the same morning matched on all three on-page checks for 132 of 132. Our method page sets out how we separate findings from assumptions.

Want the same five checks on your page?

Send us the one page that should win your town's search. We'll run these five checks on it and tell you what to fix first. It starts as a free visibility check, and our SEO, AEO and GEO programme starts from S$500/month if you want us to do the search marketing work in Singapore for you.

After a change, we measure instead of guessing. SingRank Pixel can identify an AI referral visit when the visit carries a recognised AI referrer or campaign marker, so you see whether assistants send people to the page you fixed.

What we won't do is promise an outcome. As our GEO page puts it: "No. AI platforms choose their own answers."

These findings describe pages sampled on 13 September 2026 through metasearch. This page names no business and is not a rating of any company.

FAQ

What is AI search readiness?

AI search readiness is how easily AI search tools can find, read and quote a web page. Here it means five checks: an answer-first opening, question headings or an FAQ, business schema, an llms.txt file, and a robots.txt that doesn't block the main AI crawlers.

Does a website need llms.txt to appear in Google's AI Overviews?

No. Google's guidance says AI Overviews and AI Mode have no extra requirements, and that you don't need AI text files or special markup to appear. In our Singapore sample, 24 of 69 provider websites had llms.txt anyway, and 6 of the 31 files we found carried an SEO plugin's signature.

Should a Singapore business block GPTBot or other AI crawlers?

Only with a clear reason. OpenAI describes OAI-SearchBot as the bot that surfaces websites in ChatGPT's search features, and Perplexity describes PerplexityBot the same way. None of the 69 local-service websites in our sample blocked any of the five crawlers we tested.

Why use SearXNG instead of Google results?

SearXNG is the metasearch engine SingRank runs itself, so every result could be saved and re-checked. The trade-off is real: its blended Bing and Yandex ordering isn't Google's ranking or its Maps pack, and we say so in the method section.

Who offers GEO marketing for Singapore SMEs?

Several agencies do, SingRank among them. In our sample, 11 of the 23 agency pages that appeared for "local seo agency in" searches mentioned GEO, AEO or AI search. Anyone selling AIO or AEO implementation should pass the five checks on their own pages first.

Can SingRank guarantee my business will be recommended by AI?

No. AI platforms choose their own answers, and no agency controls them. An agency can make your pages clear, accurate and easy to read, then measure what changes. SingRank records identifiable AI referral visits with SingRank Pixel and reports what those visits show.

Can I cite these figures?

Yes. Please cite "SingRank, AI Search Readiness Singapore 2026" with a link to this page and the collection date, 13 September 2026. Quote each percentage with its count, such as 28 of 109 pages, plus the metasearch caveat. Challenges to a number are welcome through our contact page.