Choosing your first keyword

The keywords you track in Outercite are not search terms. They are the full questions real customers type into AI engines when they want to buy something or find a service. Choosing them well makes the difference between data that drives action and data that sits unused.

What you'll learn

• Why buying-intent prompts outperform generic terms
• How to write prompts that match what customers actually type
• How many keywords to start with and when to add more

---

Keywords are prompts, not search terms

Traditional SEO tools track short-tail keywords: "accountant", "coffee machine", "physio Sydney". AI search works differently. When someone opens ChatGPT or Perplexity, they type full sentences. "Find me a good accountant for a small e-commerce business in Melbourne." "Which coffee machine makes the best flat white under $500?"

Outercite submits these full sentences to each of the six AI engines it monitors (ChatGPT, Claude, Perplexity, Google AI, Grok, and DeepSeek) and checks whether your business appears in the response. That means your keywords need to be full, realistic questions, not fragments.

---

Why intent matters

Not every prompt produces a citation. When someone asks "how does compound interest work?", AI engines typically answer with an explanation, not a list of businesses. When someone asks "which bank has the best savings rate right now?", they get a list of providers.

The second type of prompt is what you want to track. These are called buying-intent queries: prompts where the person is close to making a decision and the AI is likely to cite specific businesses or products.

Intent classification in Outercite separates your keywords into buckets so you can see at a glance which ones are generating citations and which are informational. The aggregate citation rate across all tracked prompts is 48%. Buying-intent prompts typically sit above that; informational prompts sit below.

---

How to write good buying-intent prompts

Think about the moment just before a customer finds you. What would they actually type?

Think like a customer, not a marketer

Marketers often write prompts that describe their product: "best enterprise CRM software". Customers ask questions that describe their problem: "what CRM should I use if my team is 10 people and we sell to retail buyers?"

The more specific the prompt, the better. Specific prompts surface you when it matters most.

Include context where relevant

AI engines use context to narrow their recommendations. If your business has a location, a niche, or a price point that customers care about, include it in the prompt.

Examples:

• "affordable physio near the CBD for sports injuries" (local + niche)
• "which accounting software is easiest for a sole trader in Australia?" (problem-specific)
• "where can I get same-day flower delivery in Brisbane?" (local + urgency)

Use comparison language

Comparison prompts ("best X vs Y", "which X is right for me") are strong citation triggers because AI engines often name specific options. For example: "should I use a mortgage broker or go direct to my bank?"

---

Practical examples (generic)

The table below shows the difference between weak and strong prompts for three types of business.

Notice that the strong prompts are questions a real person would type. They include a setting, a decision point, and sometimes a constraint (price, location, timing).

---

How many keywords to start with

Start with between 5 and 15 prompts. This is enough to get meaningful data without overwhelming your first report.

Here is a rough breakdown:

• 3-5 prompts that describe your core product or service with buying intent
• 2-4 prompts that are location-specific (if relevant)
• 1-3 comparison prompts where customers weigh up alternatives
• 1-2 branded prompts (your company name alone, or "reviews of [brand]")

---

Where to find prompt ideas

You do not need to invent these from scratch. Good sources include:

• Your sales team. Ask what questions prospects ask before signing up.
• Your support inbox. Pre-purchase questions often map directly to AI prompts.
• Your existing SEO data. Long-tail queries in Google Search Console are close to AI prompts. Connect your account via Google Search Console integration to import suggestions directly.

After a week of data, go to /prompts to edit, pause, or replace keywords that are not generating useful data. If a prompt you expected to be buying-intent has been classified as informational, rewrite it to be more decision-oriented.

---

Try this in Outercite

Go to /prompts and add your first five keywords now. Use the examples in this article as a starting point, then swap in your own product type and location.

Related