How to Build a Cantonese Keyword List Using Real Customer Conversations

Most Hong Kong marketers spend hours inside keyword tools trying to guess what their customers are searching for. They filter by Traditional Chinese, pick suggested volumes, and build content briefs around terms that look right on a spreadsheet. Then the content goes live and sits. The problem is not the content itself. The problem is that no keyword tool was in the room when a real customer described their issue in their own words, at their own pace, in the language they use every day.

Your customers tell you exactly what they search for. They do it every time they join a discovery call, a product demo, or a support session on Microsoft Teams. This workflow shows you how to pull those phrases out systematically and turn them into a Cantonese keyword list that reflects how people in Hong Kong actually type and search.

Why Cantonese Keyword Research Is a Different Problem Entirely

The reason keyword tools fail in the Hong Kong market is straightforward. Cantonese has its own colloquial written register, including characters, particles, and tonal markers that carry meaning specific to how speakers write the way they talk. A search query typed by someone in Wan Chai or Tuen Mun can look completely different from the same question typed in Beijing or Taipei, even when the topic is identical.

Most popular keyword research platforms aggregate data primarily from Simplified Chinese or English queries. Their suggestion engines are trained on datasets where Cantonese speakers are a minority. The most natural, high-intent phrases your Hong Kong audience actually types tend to surface as low-volume or simply do not appear at all.

Standard keyword tools are poor at capturing that register. Going directly to customer conversations gives you a structural advantage over competitors who rely on those tools alone. You get access to phrasing that has never appeared in a keyword planner report because it lives exclusively in spoken Cantonese, and in the typed messages people send when they search after those conversations.

The Most Underused Research Asset Most HK Marketers Already Have

Sales teams, account managers, and customer success staff at most Hong Kong SMBs and agencies run multiple video calls every week. Many of those calls happen over Microsoft Teams. Teams records sessions automatically when hosts enable the feature, and those recordings sit in Microsoft Stream or SharePoint long after the call ends.

Nobody mines them for marketing research. That is the gap this workflow fills.

The raw material is already there. You do not need to commission surveys or run separate user-interview programs. You need a process for getting the audio into a format you can search and annotate, and a framework for knowing what to look for once you have it. Five to ten transcripts from customer calls are enough to surface patterns that months of keyword tool use would never reveal.

Converting Recordings into Text You Can Actually Search

Audio is not searchable. Transcripts are. The first step in this workflow is converting your Teams recordings into accurate text files that preserve speaker labels and timestamps, so you can trace phrases back to context later.

Tools built specifically for this step handle both the transcription and the speaker separation. Running your recordings through a teams to text converter gives you a clean, structured transcript file you can paste into a spreadsheet, search in a text editor, or annotate with tags. The quality of the transcript matters because Cantonese mixed with English loanwords, brand names, and industry jargon needs accurate recognition to be useful for keyword work. Test with a short recording first and check how it handles code-switching, which is extremely common in Hong Kong business conversations.

Once you have clean transcripts, store them in a shared folder with consistent naming: client type, date, and product or service area. That structure lets you pull transcripts from specific segments later when you are building keyword lists for different service pages or audience groups.

How to Spot the Phrases Worth Keeping

Not every sentence in a customer call is a keyword opportunity. You are looking for specific patterns that signal search intent. These are the moments where a customer describes their problem before they know your solution, asks a comparison question, or uses a local phrase that a keyword planner would never suggest.

Pay attention to these patterns as you read through your transcripts:

• Problem-first descriptions: Phrases like "我唔知點解..." (I do not understand why...) or "之前試過但係..." (I tried before but...) flag the clause that follows as a search-ready keyword seed.
• Question particles: Patterns ending in "係咪" (is it), "點做" (how to do), or "有冇" (is there / do you have) map directly to conversational search queries that Google sees from Hong Kong users.
• Comparison language: Customers who ask "同...有咩分別" (what is the difference between... and...) are using exact phrasing that mirrors high-intent comparison searches.
• Colloquial product names: Customers rarely use your official product name when they first describe what they are looking for. The informal term they use is often what they typed into Google before finding you.
• Objection and anxiety phrases: Language like "係咪安全" (is it safe) or "會唔會..." (will it...) points toward FAQ content and trust-building keywords with genuine local search volume.

Flag each phrase with a simple tag as you go: problem, question, comparison, colloquial name, or objection. You will use those tags to sort your list in the next stage.

Building a Structured Keyword List from Raw Conversation Data

Once you have tagged phrases from five to ten transcripts, patterns emerge fast. You will see the same colloquial terms appear across multiple calls. You will notice customers describing the same product three or four different ways depending on their background or industry. Each variation is a keyword cluster waiting to be built out.

Move your flagged phrases into a spreadsheet with four columns: raw phrase exactly as the customer said it, cleaned keyword version normalised for search, tag category, and the source transcript reference. Keep the raw phrase intact even if it feels rough. That exact phrasing is often the most natural search query you will ever find, because a real person under real conditions produced it.

At this stage, cross-reference your phrases against Google Search Console data for your own site. If customers are using terms that already drive some impressions but have low click-through rates, those are priority targets. They are already in the index. They just need better content mapped to them.

Run each phrase through Google Trends set to Hong Kong and Traditional Chinese to confirm whether interest is stable, seasonal, or trending upward. You do not need high volume to justify creating content around a Cantonese phrase. In the Hong Kong market, a tightly targeted page that answers a specific local question can rank with relatively modest competition because most competitors are simply not targeting that query at all.

Mapping Keywords into a Content Brief Your Writer Can Use

A keyword list is not a brief. A writer handed a list of Cantonese phrases without context will default to whatever structure feels familiar, which usually produces something generic that could have been written for any market. The value of mining customer calls is not only in the phrases themselves. It is in the context surrounding them: the objections that follow, the comparisons the customer draws, and the emotional framing they use to explain their situation.

For each target keyword cluster, write a brief that includes the primary keyword phrase, two or three related phrases from the same conversation cluster, the customer context where the phrase appeared (paraphrased, not a direct quote), the intent behind the search (informational, comparison, or ready-to-act), and one or two objections the customer raised immediately after using the phrase.

That last element is particularly powerful. If a customer says "我想知道...[keyword]...但係係咪..." the concern they raise right after is the content your page must address to convert, not only to rank. Writers who understand that context produce pages that answer the full question, not just the surface query. That depth is exactly what separates pages that rank and convert from pages that rank and bounce.

Keep each brief tight. One page per cluster. The writer should read it in two minutes and know exactly what the article needs to say, in what register, and for whom. Cantonese readers notice immediately when content was written for a general Chinese audience and then adjusted for Hong Kong afterward. Authentic local phrasing grounded in real customer language makes that difference apparent from the first paragraph.

Turning Customer Voices into Your Most Competitive Keyword Asset

The marketers who consistently outperform in Hong Kong search are not the ones with the biggest keyword budgets. They are the ones who understand what local customers are actually saying and build content around that understanding with discipline and consistency.

Your call recordings are a running primary research study. Every week, customers are telling you in Cantonese, in their own register, with their own colloquialisms, exactly what they need and how they describe it. Converting those conversations into structured keyword data is a workflow that takes a few hours to set up and then compounds in value every time a new transcript is processed.

The keyword tools built for other markets will keep missing these phrases. That is not a problem you fix by subscribing to a different tool. It is a problem you solve by going closer to the source, which in Hong Kong means going back to the conversations your customers are already having with you, and finally listening to them as a marketer, not just as a salesperson.