Your AI visibility numbers are only as good as the prompts behind them. Track questions nobody asks and you measure the wrong thing. Track questions that name your brand and you look great, while new customers never hear of you. This playbook shows how to build a prompt set that reflects what your customers actually ask, and how to keep it useful as it grows.
Start from real customer questions
The best prompts are questions people already ask you, or ask about your category. Good places to find them:
- Sales calls and demos. What prospects ask before they buy: what it costs, how it compares, whether it fits their situation.
- Support tickets and chat logs. Especially questions from people who haven't bought yet.
- Google Search Console. Your search queries, rewritten as full questions. People type "family hotel lisbon pool" into Google, but ask an assistant "Which family hotels in Lisbon have a pool and are close to the beach?"
- People Also Ask and related searches on Google for your main keywords.
- Community threads where people ask for recommendations in your category.
Write each prompt the way a customer would say it: a full question in plain words, with the context people add, such as a budget, a group size or a place. "Coffee subscription" and "Which coffee subscription is best for someone who drinks two espressos a day?" can get quite different answers, and the second is closer to how people ask.
Cover the whole buying journey
People ask different questions at different stages, and different brands win each stage. A useful prompt set covers all of them. The examples below are for Northwind Coffee, a made-up brand that sells coffee subscriptions.
| Stage | What the customer wants | Example prompt |
|---|---|---|
| Problem | To solve a problem, before looking for a product | "How do I keep coffee beans fresh at home?" |
| Category | The best option for their situation | "Best coffee subscription for espresso drinkers" |
| Comparison | One kind of solution against another | "Is a coffee subscription cheaper than buying beans at the store?" |
| Alternatives | Options besides a brand they know | "Alternatives to Copperline Roasters" (an example competitor) |
| Brand | Facts about your brand by name | "Is Northwind Coffee worth it?" |
Most of your prompts should sit in the category, comparison and alternatives rows. That is where an assistant decides which brands to recommend. Problem prompts show whether you are seen as an expert. Brand prompts do a different job, covered next.
Keep your brand out of discovery prompts
If a prompt names your brand, the answer will almost always mention you. That says nothing about whether a customer who doesn't know you yet would find you, and it inflates your visibility.
So keep two kinds of prompts apart:
- Discovery prompts don't name you. They measure whether AI recommends you to people who are still choosing.
- Brand prompts name you. They show what AI says about you: whether the prices, features and reputation it describes are accurate, and in what tone.
Give brand prompts their own tag or topic, so you can read them separately from discovery prompts. Prompts that name a competitor, such as alternatives questions, belong in the discovery set: they show whether you are offered as the alternative.
Group prompts by topic
Topics are the themes you compete on. For Northwind Coffee they might be espresso, decaf, gifts and subscriptions. Grouping prompts this way shows which themes you lead and which a competitor owns, something a single overall number hides.
Use tags for things that cut across topics: the stage of the journey, a customer type, or brand versus discovery.
How many prompts per topic
Five to ten prompts per topic is a practical start. With fewer, one unusual answer swings the topic's numbers. With many more, you pay for near-duplicates and won't have time to read the answers.
Make variations that differ in intent, not just in word order. "Best espresso beans" and "best beans for espresso" will often get similar answers. "Best espresso beans for milk drinks" or "best low-acid espresso beans" add something new.
Three to six topics with five to ten prompts each is enough to see patterns. Expand once you have read the first answers and know where the gaps are.
Choose markets and languages
Track each prompt in the markets where you sell, in the language customers use there. A Dutch prompt asked from the Netherlands is a different question from an English prompt asked from the Netherlands, and it gets different brands and sources.
- Translate for meaning, not word for word. Use local phrasing, local currencies and units, and the competitors people know there.
- Run local prompts only where they apply. "Best coffee roaster in Amsterdam" belongs in the Netherlands, not in every market.
- Prioritize markets by what they are worth to you. Each market multiplies the work: a prompt checked daily in three markets uses three prompt units. Plans are sized in prompt units, so the markets you add decide what you pay.
Use prompt volumes to prioritize
Not every prompt is asked equally often. Two volumes help you decide which ones matter most:
- AI search volume: how often a prompt is asked in AI tools in a month, in that market.
- Google volume: how often it is searched on Google in that market.
Keyword difficulty and search intent add context. Use the volumes to rank prompts within a topic and to decide which topics deserve more prompts. Keep a few lower-volume prompts that matter commercially, such as the questions people ask right before they buy.
Long, specific questions often show low or no volume, because few people type exactly the same sentence. Read volumes as a way to compare prompts, not as a forecast of traffic.
Group prompts by the page that should win them
For every prompt, ask which page on your site should be the answer. Several prompts often lead to the same page. "Best coffee subscription for espresso drinkers" and "espresso bean subscription for home baristas" can probably be won by one article.
A good test is overlap. If two prompts share many of the same Google results and AI sources, one article can usually win both. GoGetCited's Pages view suggests groups this way, from prompts that share at least three of the URLs Google ranks and AI answers cite. It flags when two of your pages compete for the same prompt, and lists the pages cited instead: what your article has to beat.
Prompts without a suitable page on your site are content gaps. They are often the clearest place to start writing.
Review the set every month
A prompt set is never finished, but it shouldn't change every week either. Once a month:
- add new questions from sales, support and Search Console
- review suggested prompts and keep the ones that fit
- move prompts that turned out to be irrelevant to inactive
- make sure new prompts have a topic and a page
Keep a stable core so your trends stay comparable. Adding or removing many prompts at once changes your metrics even if nothing else changed. AI visibility metrics explained covers how to read those trends.
How GoGetCited helps
The setup wizard reads your website and proposes topics, prompts and competitors, about five prompts per topic, in each market's language. You can import Search Console queries rewritten as prompts, paste a list or a CSV, or ask for more suggestions from People Also Ask, related questions and your site. Each prompt shows its AI search volume and Google volume per market, refreshed monthly, and Pages groups prompts by the article that should win them.