Is Hong Kong behind Singapore online? SEO and AI search compared
Slightly, and not in the way the headline suggests. Across 135 published sites in five trades, 34% of our checks did not pass in Hong Kong against 30% in Singapore. Four points. The useful part sits under that average: the difference is almost entirely in the machine-readable basics.
Mr Cheung asked for a Cantonese-speaking lawyer
Mr Cheung's landlord has stopped answering, so on a Tuesday night in Central he stops scrolling through old messages and asks Google AI Mode instead:
My landlord is keeping my deposit, I need a lawyer who works in Cantonese, can I see someone this week, and roughly what does it cost.
Mr Cheung is angry and wants a name he can trust in his own language before bed. The AI weighs firms by what their sites state outright. Somewhere in Central a firm fits that sentence exactly, yet his call never comes, because its pages say nothing about Cantonese or tenancy. Is that your firm? Are other lawyers already hearing from clients like him?
Once, ranking for "tenancy lawyer Central" was enough. In September 2026 we scanned 147 business websites across five trades in both cities, then read each again as plain text, as a reader that cannot run JavaScript sees it. 135 scored well enough to publish. The results are public in the dataset repository; no business is named.
The headline is dull. Of all checks run, 34% did not pass in Hong Kong and 30% in Singapore: four points, on 46 Hong Kong sites and 89 Singapore ones. The interest is in the average taken apart.
Law firms and clinics carry the whole gap; salons are level
Hong Kong law firms sit 19 points below Singapore on Search SEO and 22 below on AI Visibility; medical clinics are 13 and 21 behind. In the other three trades the difference is a few points or none.
That rules out the easy story. It is not a condition of the city, since Hong Kong salon sites match Singapore's. It is a trade norm.
One build decision, measured six times
Take a diner who writes: "A table for six on Sunday, somewhere halal, and I would rather book on WhatsApp." To answer that, the menu, the hours and a way to book must sit in the page the AI fetched. These are the checks where the markets differ most, as the share of sites not passing.
Every row is the same failure wearing a different name. A site that assembles its content with JavaScript has no text or headings before the script runs, and rarely carries structured data, so nothing says which business it is about. One build decision, not six problems. The AI arrives in an empty room with a very confident title.
None of this is writing for machines: Google's own guidance is that its generative AI features call for no special style of writing, and that structured data is not required 1. These checks ask something more boring: did any of it arrive?
Singapore looks worse on identity, and that number is a mirage
The gap does not run one way. A customer who asks whether this is the firm on the professional register relies on a site that points at a record somebody else maintains. On that check Singapore looks much worse: 72% fall short against 39% in Hong Kong. It also trails on sections written to be lifted whole, 20% against 2%, and on pages carrying any statistic, 38% against 24%.
The identity number is a mirage. The check runs on the entity a page declares, and where none is declared (59% of the Hong Kong sample) there is nothing to test. A check that stands down has not passed. Hong Kong looks better because it is worse at the step before.
The other two are real writing habits. What a model actually reads covers the first.
What an owner in either city can do this afternoon
If you are in one of the two trades carrying the gap, the finding is not that Hong Kong is behind. Your trade has a build norm, your competitors share it, and it costs all of you the same thing. These fixes take an afternoon. As a customer, I would want the services, the price range and a way to reach you in the first thing the AI reads.
- Check what arrives before JavaScript. Turn scripts off and look at your homepage. If the services and address are gone, that is what many machine readers receive.
- Say who you are, in markup. One block of structured data naming the business, its address and its contact route closes the widest gap in the table above.
- One H1, and a contact route. Both were missing on more than half the Hong Kong sample.
- Point at a record of yourself. A register entry or professional profile.
- Put a number in your copy. Hours, prices, years in practice. The AI quotes specifics and skips adjectives.
Each trade in detail: law firms, clinics, restaurants, dentists and salons.
The old way still works, a little less every year
A firm that ranks in Google will still hear from some tenants. The share shrinks, though: across the United States, 68% of Google searches ended with no click through to any site in early 2026, up from 49% in 2019 2.
Our work is to make a firm's site say, in the words a worried tenant uses, what it does and for whom, so that Mr Cheung's question lands on a page that replies. The free scan reads your homepage as the AI does and tells you where you stand.
Questions people ask
Is 147 sites enough to say anything about a city?
No. This is a purposive sample of trades we can scan consistently, roughly 10 sites per trade in Hong Kong and 20 in Singapore, not a market survey. Treat any figure here as a prompt to check your own site.
Why publish the numbers where anyone can check them?
Because a claim nobody can re-derive is marketing. The repository carries the per-trade summary, the per-check results, the plain-text read and the checks we withheld, with reasons. No business is named.
Why are some checks withheld from the dataset?
They measured something narrower than their name, and we corrected them after these scans ran. The batches are other people’s websites on one day and cannot be re-run, so a rate from a rule we no longer stand behind would be a false statistic.
Why are law firms so far apart between the two cities?
We can say what differs, not why. The Hong Kong law-firm sites here lean much harder on JavaScript-rendered content and carry structured data far less often, which moves every dependent check at once. Whether that is procurement, agency habit or site age, this data cannot say.
What would it take to close the gap on my own site?
Usually a build change: serve the text in the HTML, name the business in structured data, keep one H1, publish a contact route. The free scan reads one site the way the AI does and returns both scores and the three findings that matter most.
Next: Can ChatGPT read your website? Why AI search is a lazy guest
Sources
- Google Search Central, Optimizing your website for generative AI features on Google Search, 10 July 2026. Accessed 18 September 2026.
- SparkToro, In 2026, less than one third of Google searches still send a click, 9 June 2026. Accessed 13 September 2026.
Our own counts come from our Hong Kong and Singapore sector studies of September 2026, 147 sites, one scan and one text read each, scored with the published rubric. Not a market survey.