AEO agency · Tsuen Wan, Hong Kong

Hong Kong AEO agency: get your business recommended by ChatGPT, Gemini and Perplexity

AEO — answer engine optimisation, also called AI search optimisation or GEO — is the work of getting AI assistants to name, describe and cite your business correctly when someone asks them for a recommendation. Webka is a Hong Kong agency that does it: we measure what ChatGPT, Gemini, Claude, DeepSeek and Perplexity currently say about you, fix the technical and content foundations underneath, and re-test a fixed query set by hand every month.

Not to be confused with the customs scheme: in Hong Kong “AEO” also stands for Authorized Economic Operator, a Customs and Excise trade-facilitation programme with no connection to marketing.

Run the free 5-engine check

Free · No sign-up · 1–2 minutes · Report in Traditional Chinese

Based in
Tsuen Wan, HK
Engines tracked
5 engines
Retainer from
HK$4,800/mo
Ranking guarantees
None, ever

Who this is for

We are built around the Hong Kong market and the Cantonese / Traditional Chinese language. That is a real constraint, so here is who it fits — and who it does not.

Hong Kong SMEs selling to a Cantonese-speaking market

Our probe prompts are written in Cantonese and the market frame defaults to Hong Kong, so the engines are asked the way your actual buyers ask. If your customers type questions in Traditional Chinese, that is the language the measurement has to happen in — an English-only audit would miss most of what matters to you.

Service businesses where the buyer asks before they search

Driving schools, clinics, accountants, yacht brokers, agencies, contractors — categories where a shortlist gets built by asking an assistant "who's good at X in Hong Kong". If the model does not name you, you are not on the shortlist, and your Google ranking never gets a chance to matter.

Companies with an existing site and no idea what AI says about it

You do not need a rebuild to start. Most of what moves an AEO score is structural: crawlability for AI user agents, consistent entity data across canonical, og:url, JSON-LD and sitemap, llms.txt, and answer-first content. All of that lands on the site you already have.

Not a fit: anyone who needs a guaranteed ranking

Nobody controls what a third-party model outputs, and any agency that guarantees a ChatGPT recommendation is selling you something they cannot deliver. If a contractual guarantee of AI placement is your requirement, we are the wrong firm and we would rather say so on this page than in month three.

How we work: four steps, reviewed by a person each month

This is a monthly human review, not an always-on automated platform. A live trend dashboard and real-time alerts are on the roadmap and are deliberately not in the price.

  1. 1

    Baseline

    We run the free 5-engine check on your URL and then re-run it, because AI replies vary by roughly ±10 points between runs. What you get is a dated starting score with the raw answers attached, not an adjective.

  2. 2

    Query set

    We agree the queries that actually matter to your business — how a Hong Kong buyer would phrase them in Cantonese, not keyword-tool phrasing. Watch tracks 20, Grow tracks 50. The list is yours and stays fixed, so month-to-month numbers compare.

  3. 3

    Fix the foundation

    File-level work: AI-crawler robots rules, canonical / og:url / JSON-LD @id / sitemap consistency, llms.txt and llms-full.txt, FAQPage and LocalBusiness schema, and answer-first (BLUF) rewrites. On Grow we implement it; on Watch we hand your developer the list.

  4. 4

    Re-measure by hand, monthly

    A person re-runs the query set against each engine and writes the report — this is not an always-on dashboard, and we do not pretend otherwise. A live trend dashboard and automated alerts are on our roadmap and are deliberately not in the price.

What you get before you pay anything

Every engagement starts with the same free diagnosis anyone can run. No sales call required to see your own number.

5 engines, asked live

ChatGPT, Gemini, Claude, DeepSeek and Perplexity are queried in real time, not looked up in a cached database. Any engine whose API is unreachable that day is marked unreachable and excluded from the score rather than guessed at.

A score out of 100, in five parts

Recommendation rate 35, description accuracy 25, brand awareness 15, sentiment 15, citation sources 10. Five sub-scores rather than one opaque number, so you can see exactly where the points leak.

The competitors AI names instead of you

When the engines are asked who is best in your category, the brands they volunteer are listed. That is the shortlist you are being left off, today.

Honest limits, stated up front

It takes 1–2 minutes, you get 3 runs a day, there is no sign-up — and the report is written in Traditional Chinese. There is no English report. We would rather you knew that before you spent the 1–2 minutes.

Pricing, published

A one-off onboarding fee of HK$3,800, then one of two monthly retainers. All figures in Hong Kong dollars, billed monthly.

Watch

HK$4,800 / month

Know where you stand in AI answers, every month, without guessing.

  • Full monthly AEO visibility audit across 5 AI engines, run by our team
  • 0–100 visibility score, with last month's score printed beside it for comparison
  • 20 target queries re-tested by hand every month
  • Up to 3 competitor baselines, measured once at onboarding
  • A written monthly report, emailed to you by a person
  • Manual monthly review — if something drops sharply we email you about it
  • Unlimited use of the free self-service check (3 runs a day) between reports

Not included

  • A historical trend dashboard and automated real-time alerts (planned, not built, not priced in)
  • Content production and Schema implementation
  • Citation outreach
  • A monthly strategy call

Grow

Most chosen

HK$12,800 / month

We do the work on your site, not just the measuring.

  • Everything in Watch
  • 50 target queries re-tested by hand monthly, with no cap on competitor comparisons
  • Two pieces of AEO content produced each month (FAQ pages or comparison articles)
  • Structured-data (Schema.org) implementation on your site
  • Citation outreach
  • A 45-minute strategy call each month
  • Priority Slack / WhatsApp support, with same-day notice on any sharp drop

Not included

  • A historical trend dashboard and automated real-time alerts (planned, not built, not priced in)

Both plans are a monthly human review. Neither buys a guaranteed placement in any AI answer, because nobody can sell one — see criterion 05 below.

Five criteria for vetting any AEO agency in Hong Kong

Written as a framework you can take away and use on anyone, including us. Each criterion comes with how to verify it and the red flag that fails it.

Criterion 01

Can they show you public, checkable before-and-after data?

AEO results are measurable: the same rubric and the same query list, measured once before and once after. An agency that can do it will publish it. One that cannot will give you adjectives.

How to verify

  • Ask for a dated before/after, and check that both runs used the same methodology and re-ran the same queries one by one
  • Ask for a report link you can open yourself, or the raw JSON — not a cropped screenshot
  • Ask: how many days between the two measurements, and did the scoring change in between? A changed ruler invalidates the comparison

Red flag: "Visibility up 300%" with no baseline, no date and no query list attached.

Criterion 02

Does their own website pass the test they are selling?

A firm that teaches AI search optimisation has the cheapest possible demonstration available: its own site. You do not have to trust anyone on this one — you can check it in thirty seconds.

How to verify

  • Open their-domain.com/llms.txt in a browser and see whether anything is there, and whether it actually describes their site
  • View source and search for <script type="application/ld+json"> — look for Organization and FAQPage nodes
  • Ask what their own AEO score is. Willingness to quote an unflattering number is usually a better signal than a claimed perfect one

Red flag: They sell Schema markup and llms.txt, and their own site has neither.

Criterion 03

Can they name the files they will change in month one?

"AI optimisation" can be wrapped around anything. Real AEO delivery is a list of dull, specific jobs: AI-crawler robots rules, domain consistency across canonical, og:url, JSON-LD @id and sitemap, llms.txt and llms-full.txt, BLUF content structure, FAQ and LocalBusiness schema.

How to verify

  • Ask: if we sign, which files on my site do you actually edit in the first month?
  • A competent answer is at file and schema-type level — FAQPage on each service page, llms.txt at root, canonical consistency repair
  • Then ask how they verify it went live. curl, view-source or the Rich Results Test are real answers

Red flag: Every answer stays at the level of "improving AI visibility" and "building semantic authority", with no filename ever mentioned.

Criterion 04

Did an AI actually answer, or did a human estimate the score?

The easiest part of an AEO report to fake is the "what AI says about you" section. The real method runs a fixed query set and records each engine's raw reply on the day. The fake method renames last quarter's SEO report.

How to verify

  • Ask to see the query list: how many, how they cluster, and whether any are phrased in colloquial Cantonese
  • Ask to see the raw engine reply per query, with test dates. Engines change fast, so a year-old record tells you little
  • Read the metrics: keyword rankings and organic sessions are SEO. Share of queries where the brand was mentioned or cited is AEO

Red flag: The report is a traditional SEO monthly with a new cover page.

Criterion 05

Will they tell you what they cannot do?

Several AEO levers cannot be pulled with code: Google Business Profile, genuine reviews, third-party backlinks, editorial inclusion, and when an engine chooses to re-crawl you at all. An honest provider separates those out on day one. A dishonest one guarantees them along with everything else.

How to verify

  • Ask what is outside their control. An answer naming GBP, reviews, backlinks and engine re-crawl timing means they have done this before
  • Ask what the worst case looks like. Someone willing to say "technical score rises, citation rate may still be zero" has shipped real projects
  • Read the contract: is there a long lock-in before any measurable milestone?

Red flag: "Guaranteed ChatGPT recommendation." Nobody can guarantee a third party's model output.

Our own answers to those five criteria

We wrote the criteria and we are an AEO agency, which is an obvious conflict of interest you should not have to spot yourself. So here are our answers with the evidence attached — including the numbers that do not flatter us. If you think the framing is biased, the best response is to take the same five questions to everyone else and see who produces comparable evidence.

Criterion 01 — public before/after

Client case, same 8-dimension weighted rubric and the same 22 visibility queries, re-run 76 days apart: total score 46 → 67.

Per-dimension deltas and the rubric weights are all published (Schema 16, Crawlability 12, Content 16, Technical SEO 10, Performance 8, Local 14, E-E-A-T 14, UX 10).

Open the case study

Criterion 01 — we measure ourselves too

Our own site, measured on the same tool: average 41 on 2026-03-31, average 43 on 2026-09-13. Two points in 165 days.

All six diagnosis reports are published as open links, so you can check every score and breakdown yourself. We rebranded midway, which cost us accumulated brand memory.

Open the case study

Criterion 03 — file-level delivery

In the same client project the fixes were physical: domain consistency across five surfaces — canonical, og:url, JSON-LD @id, robots and sitemap — went from 0% to 100%, and personal phone-number leakage in llms.txt and related files went from 5+ instances to zero.

Before the repair, curl found 15 stale preview URLs and zero instances of the real domain. That is not a matter of opinion; you can re-run the check.

Open the case study

Criterion 05 — what we could not do

Same project, published alongside the win: the AI citation rate across those 22 queries was 0/22 at T0 and still 0/22 at T1, share of voice about 1%, and site: indexing returned zero results.

Foundation repair (a verified +21 points) and authority building (GBP, reviews, backlinks, editorial inclusion — still on the blocked list) are reported as two separate stories rather than blended into one flattering number.

Open the case study

Criterion 02 — our own site as the demo

Open https://webka.hk/llms.txt to see how we describe ourselves to AI engines, then view source on this page for its JSON-LD.

We run the same techniques on ourselves, not only on clients.

Check it yourself

Criterion 04 — a record of real AI answers

The free check is open to anyone, with no sign-up, and every run produces a shareable permanent report link.

Run it on your own site and take the report to any provider you like — including us.

Open it

Case studies are written in Traditional Chinese; the scores, dates and report links are readable in any language.

Frequently asked questions

How much does an AEO agency in Hong Kong cost?

Ours is published rather than quoted on request: a one-off onboarding fee of HK$3,800, then either HK$4,800 a month for Watch (monthly 5-engine audit, 20 hand-re-tested queries, 3 competitor baselines, a written report) or HK$12,800 a month for Grow (everything in Watch, plus 50 queries, two pieces of AEO content a month, Schema implementation, citation outreach and a 45-minute strategy call). Across the wider Hong Kong market, AI search optimisation is usually sold as a monthly retainer rather than a project fee, and most firms will not publish a number at all — which makes comparison hard and is a reason we publish ours. Treat any quote that is dramatically cheaper with the five criteria on this page: at the low end, what is usually being sold is a traditional SEO retainer with an AEO cover page.

How long until results?

Honestly: months, and the first thing to move is not the thing you care about most. Technical foundations — crawlability, entity consistency, schema, llms.txt — can be fixed in weeks and show up in a score quickly. In one published client project the total score went from 46 to 67 in 76 days on an identical rubric. But in that same project the AI citation rate across 22 queries was 0/22 before and 0/22 after, because citation depends on things code cannot reach: a Google Business Profile, genuine reviews, third-party links and editorial coverage, plus whenever each engine decides to re-crawl you. Models with live grounding, such as Perplexity, reflect changes fastest; models answering from training data are slowest of all. Anyone promising results in weeks is describing the easy half only.

What is the difference between an AEO agency and an SEO agency?

An SEO agency optimises for a results page: rankings, clicks, organic sessions. An AEO agency — answer engine optimisation, also called AI search optimisation — optimises for the answer itself: whether ChatGPT, Gemini or Perplexity name your business when a buyer asks who is good at something, how accurately they describe you, and whether they cite you as a source. The work overlaps at the technical layer (crawlability, structured data, clean entity data are good for both), and diverges at the reporting layer. That divergence is the fastest way to tell the two apart: if the monthly report shows keyword positions and traffic, you bought SEO. If it shows the share of a fixed query set where your brand was mentioned or cited, and the raw engine replies, you bought AEO. A second, checkable tell is pricing: of the 8 Hong Kong agencies selling AEO or GEO that we reviewed in September 2026, 7 published no prices at all — only one listed starting fees. Ask for the query set and the price before you ask about the framework.

Is AEO the same as GEO?

In practice, yes — they are competing labels for the same discipline. AEO stands for answer engine optimisation and GEO for generative engine optimisation; you will also see LLMO (large language model optimisation) and simply "AI search optimisation". No standards body owns any of them, and the deliverables behind them are the same: make your entity data consistent, make the site crawlable by AI user agents, publish answer-first content and structured data, and measure what the engines actually say. We use AEO as the primary term with GEO as an alias, which is partly a Hong Kong problem: in an English search here, the bare acronym "AEO" mostly returns the customs Authorized Economic Operator scheme, so the phrase is nearly always worth disambiguating. If a prospective agency insists its GEO framework is fundamentally different from AEO, ask criterion 03 — which files do you change — and see if the answer differs.

Is this the same AEO as the customs Authorized Economic Operator scheme?

No, and the collision is worth knowing about because it affects your own searching. Authorized Economic Operator is a Hong Kong Customs and Excise trade-facilitation programme for certified supply-chain operators, and it dominates the English search results here for the bare acronym — government and certification-body pages fill the first screen, and marketing results appear well below. The AEO on this page is answer engine optimisation, a digital marketing discipline with no connection to customs compliance whatsoever. Practical consequence: when researching providers, search "answer engine optimisation Hong Kong", "AI search optimisation Hong Kong" or "GEO agency Hong Kong" rather than "AEO Hong Kong", or you will spend your afternoon reading about certified importers.

Do I need an English website to work with you?

No — and for most of our clients an English site is not the priority. Our measurement is built around Cantonese probe prompts and a Hong Kong market frame, because that is how buyers here actually ask an assistant for a recommendation. A Traditional Chinese site is a perfectly good subject for the work; what matters is whether AI crawlers can read it, whether your entity data is consistent, and whether the content answers questions directly. An English version becomes worth adding when you sell across the region or to overseas buyers, in which case the correct pattern is a proper bilingual setup with reciprocal hreflang, not a machine-translated mirror. We can advise on that as part of the technical work, but nobody should rebuild their site in English as a precondition for measuring their AI visibility.

Is the report in English?

No. Both the free diagnostic report and the monthly client report are written in Traditional Chinese, and the probe prompts are generated in Cantonese. There is currently no English report and no locale switch, and we would rather state that here than have you discover it after waiting 1–2 minutes for a check to finish. The parts that travel across languages regardless are the numeric scores and their five sub-scores, the raw AI replies, the competitor names and the generated Schema.org JSON-LD — which, for a technical reader, is most of the actionable content. If you need an English-language reporting layer, say so when you get in touch: it is a reasonable request and we will tell you honestly whether and when we can do it, rather than promising it and improvising later.

Can you guarantee that ChatGPT will recommend my business?

No, and you should treat any agency that says yes as disqualified. What an engine outputs is controlled by the model provider, its training data, its retrieval layer and its re-crawl schedule — none of which any agency can direct. What can be committed to is the work and the measurement: a dated baseline, a fixed query set re-tested by hand each month, file-level technical fixes you can verify yourself with curl or view-source, and a written report showing movement in both directions. Our own published record includes a client project where the technical score rose 21 points while the AI citation rate stayed at 0 out of 22 queries, and our own site moving only 41 to 43 across 165 days. Those are the numbers an honest guarantee-free engagement produces, and they are on this site because they are true.

Start with your own number, not our pitch

Run the free check on your site, read what the engines actually say about you, then decide whether you need an agency at all.Free · No sign-up · 1–2 minutes · Report in Traditional Chinese

Check what AI says now

We reply within one working day. The contact form is in Traditional Chinese; WhatsApp in English is fine.