Guides
Choosing prompts
What makes a good prompt, what mix of types to keep, and the mistakes that skew the numbers.
Your prompts are your sample. If they are badly chosen, the numbers are correct, but they measure something other than what you care about.
A good prompt
- Sounds like a customer who doesn't know you yet. "What invoicing software is good for a freelancer?", not "What do you think of [my brand]?".
- Is specific. "Best courier for large parcels in Bucharest" says more than "good couriers". Assistants answer broad and concrete questions differently, and customers ask concrete questions.
- Is in your market's language, the same as the brand's language.
- Is a question, not a keyword. People talk to assistants in full sentences.
A balanced mix
| Type | Example | What you learn |
|---|---|---|
| Organic | "Which travel agency is good for Greece?" | whether you appear on your own, without being named in the question |
| Competitor comparison | "X or Y for holidays in Greece?" | how you are described next to competitors |
| Brand-specific | "Is agency X trustworthy?" | what assistants say about you when someone looks you up |
Keep most of your prompts organic: they answer the main question — do assistants recommend me when they don't know about me? Brand-specific prompts name you almost every time, so they inflate visibility; keep them few and keep them in mind when you read total visibility.
Cover every stage: informational questions ("how do I choose…"), comparison questions and buying questions ("where do I buy…"). The Intent mix card shows whether your set leans toward just one.
Where to get ideas
- Google Search Console — the searches that already bring you visitors. See Google Search Console.
- What people search for — related searches and questions from Google.
- Related questions — you start from a prompt that does well and find variants.
- Where you're missing → Generate probe prompts — topics where competitors already appear.
They are all in Prompt Discovery.
Common mistakes
- The brand name in organic prompts. The assistant will name you because you asked it about you.
- Near-identical prompts. Ten variants of the same question give the same number ten times and weigh too much in the total. Ten different questions are better.
- Changing the set often. Numbers are compared from one period to the next on the same prompts. If you replace half of them, the change next to the number says more about the prompts than about you. Add gradually, and pause instead of rewriting.
- Conclusions from a single prompt. On one prompt, the number comes from few answers and jumps easily. Read the whole set, or several weeks.