Key takeaways
- Commercial-intent prompts are questions asked close to a decision. They name a budget, a use case, a competitor or a doubt.
- Five types cover most of them: comparison, best-for, price, alternative and is-it-worth-it. Each type is won with a different kind of content.
- People phrase the same need in very different words, so a prompt list should group variations by intent instead of chasing one exact wording.
- Prioritise by closeness to purchase, fit with what you sell, and whether you appear today.
From the search box to the chat window
For years, purchase research in Korea was spread across Naver search, blog reviews, YouTube and marketplace reviews. AI chat compresses that journey. A buyer can describe a budget, a household and a constraint in one message and receive a shortlist with reasons. A brand that is missing from that shortlist is not compared at all.
Shopping is not the main thing people do with AI chat. The NBER working paper How People Use ChatGPT finds that practical guidance, seeking information and writing together account for nearly 80% of conversations. Buying questions hide inside those everyday categories: a request for help choosing a laptop is practical guidance and a purchase decision at the same time. That is why commercial prompts have to be recognised by what they ask, not by a label.
How to recognise a buying question
A prompt has commercial intent when answering it would narrow down a purchase. Look for these signals:
- A product, brand or category paired with a decision word such as recommend, compare or which is better
- A budget or price range
- A situation of use, such as a small apartment with a cat or a ten-person sales team
- A named competitor, or the product the buyer wants to replace
- A doubt about value or risk, such as downsides, durability or hidden costs
Informational questions, such as how a heat pump dryer works, still matter, but they come earlier in the journey and are won with explanatory content. Keep them on a separate list so they do not dilute the buying questions.
Five types of buying questions
The examples below are illustrations, written the way buyers tend to phrase these questions.
| Type | For example | What the engine needs | Content that helps |
|---|---|---|---|
| Comparison | Brand A or brand B for a small office? | Side-by-side facts on both | Fair comparison pages and spec tables |
| Best-for | Best CRM for a ten-person sales team | Who the product suits, with evidence | Use-case pages and customer stories |
| Price | Good air purifiers under 300,000 won | Current public prices and plans | Pricing written as text, plan tables |
| Alternative | Cheaper alternatives to a well-known tool | How you differ from the incumbent | Alternative and migration pages |
| Is it worth it | Is this model worth it? Any downsides? | Reviews, limits and guarantees | FAQs that address drawbacks honestly |
The type tells you what kind of page to build. A comparison question is rarely won by a product page alone, and a price question is rarely won by a brand story.
Where to find your buyers' questions
- Sales call notes, inquiry forms and KakaoTalk channel chats
- Site search logs and search ad keywords that contain words like recommend or compare
- Product Q&A on Smart Store and Coupang
- Community posts and blog comments in your category
- The engines themselves: ask what people usually ask when choosing your category, then check the list against real data
Expect wide variation in wording. In SparkToro research reported by Search Engine Journal, 142 participants wrote their own prompts for the same headphone need, and the prompts had a semantic similarity score of only 0.081. Yet the answers still surfaced a similar core set of brands. The lesson for a prompt list is to group phrasings by intent and track several versions of each, rather than betting on one sentence.
Decide which questions to win first
| Criterion | Question to ask |
|---|---|
| Closeness to purchase | Would the answer directly shape a shortlist? |
| Fit | Are we honestly a good answer to this question? |
| Current presence | Do we appear in the answer today, and how are we described? |
| Content gap | Do we have a page that answers it directly? |
| Business value | Does winning it matter for revenue or margin? |
A practical starting point is a few prompts for each of the five types across your main product lines, checked on a regular schedule. Questions that score high on fit and closeness but where you do not appear are the first ones to work on.
Notes for Korean brands
- Ask in Korean first, and in English too if you sell abroad. The answers can differ completely.
- Include seasonal buying questions, such as holiday gift sets or back-to-school purchases.
- For B2B, add questions about adoption cost, integrations and Korean customer references.
- Write questions the way a buyer speaks, not the way your product catalogue is organised.
Review the list every quarter. New competitors, new price points and new features change what buyers ask, and a question that was marginal six months ago can become the one that decides most shortlists in your category. Retire prompts that no longer match how people buy, and keep their history so earlier results still make sense.
Where flowgen fits
Prompt Demand collects the questions buyers ask AI in your category, groups them by intent and shows which are worth winning. GEO Insight checks whether you appear in the answers, Content Agent drafts the comparison, pricing and FAQ pages each type needs, and Analytics shows what changed after they went live.
Sources
- NBER, How People Use ChatGPT, Working Paper 34255, September 2025
- Search Engine Journal, AI Recommendations Change With Nearly Every Query: Sparktoro, 30 January 2026