Promptologie
Prompts are to GEO what keywords were to SEO, but they behave differently: longer, conversational, and shaped by who is asking and where.
From keywords to prompts
Prompts are to GEO what keywords were to SEO, but they behave differently: longer, conversational, and shaped by who is asking and where. A keyword was “coffee machine recommendation”. The prompt is “Recommend a fully automatic coffee machine under one million won that suits a family of four and is easy to maintain.” Price, household, use and maintenance arrive in one sentence, and the engine answers all of them at once.
That changes what you optimise for. One page stuffed with one keyword cannot answer a prompt like that. A page that states the price, the household size it suits, the cleaning routine and a comparison with the obvious alternative can.
Questions change along the journey
Customers ask different questions at each stage, and most of them never mention a brand. The brand-free questions are where discovery happens, so they matter more than “What is LumiHome like?”.
| Stage | Question |
|---|---|
| Problem awareness | Why does my vacuum lose suction after a few minutes? |
| Exploration | What kinds of cordless vacuums are there? |
| Comparison | Cordless or robot vacuum for a small apartment? |
| Shortlist | Which cordless vacuum under 300,000 won is best for pet hair? |
| Decision | LumiClean Pro or HomeMate X2: which is easier to maintain? |
| After purchase | How often should I replace the LumiClean Pro filter? |
Choose prompts
- Start from real demand instead of guesses: query reports, support tickets, the questions sales hears every week, and what flowgen AI Demand finds for your category.
- Cover the whole journey: inspiration, comparison, decision, and after the purchase. Teams tend to over-invest in the shortlist stage and ignore the after-purchase questions that build trust.
- Include prompts that never name your brand. Most discovery happens there.
- Include the questions you would rather not answer: price objections, known weaknesses, comparisons with the market leader. If you do not answer them, a review site will.
- Start small. Twenty to thirty prompts for one product line is enough to learn from; widen the set once the first pages are cited.
Structure prompts
Group prompts into topics, then tag each by intent, persona and funnel stage. A useful structure is Topic → Persona → Intent → Funnel → Prompt.
| Field | Example |
|---|---|
| Topic | Cordless vacuums for small apartments |
| Persona | Renter in a studio flat with a cat |
| Intent | Choose between two models |
| Funnel | Decision |
| Prompt | Which is quieter for a studio, LumiClean Pro or HomeMate X2? |
Managed this way, prompts stop being a list of questions and become data that shows the real buying journey: which personas you win, which stages you lose, and which topics nobody owns yet.
Track prompts
Run every prompt on a steady cadence across the platforms and regions you care about. Consistency matters more than frequency: a daily reading is usually enough to see the trend, and a weekly one is enough for most teams to act on. Record the full answer, not only whether you were named, because the wording of the answer is what Brand Knowledge will read later.
Keep the prompt set stable for at least a quarter. Changing the questions every month makes before-and-after comparisons meaningless, and the point of tracking is to see what changed after you published.