Prompt Discovery: Building the Fifty Questions Your Buyers Ask AI

Why the prompt set decides everything

Every AI visibility number is a count over a set of prompts. Change the set and the number changes, whether or not anything about your visibility did. A team tracking twenty brand-name prompts will report excellent visibility and learn nothing. A team tracking the questions buyers ask before they have heard of anyone will report something uncomfortable and useful.

Prompt discovery is the work of choosing that set on purpose. It comes before share of voice measurement, and it decides whether that measurement is worth doing.

What a good prompt looks like

• Written the way a buyer writes. Full questions, plain language, with the constraints real buyers mention: team size, budget, industry, the tool they use now.
• One question at a time. A prompt that asks three things gets an answer that half-covers each.
• No brand name, except in comparison prompts where the buyer would genuinely name competitors.
• Capable of naming you. The answer should plausibly list vendors, recommend tools, or cite sources of the kind you publish.

The five stages

Aim for around ten prompts per stage. The examples use our own category, AI visibility tools, so they are real questions rather than invented ones.

The shortlist and comparison stages are where an assistant most directly decides which vendors a buyer considers, so weight them if you have to cut.

Where to find the phrasing

Do not write prompts from your own vocabulary. Collect the buyer's.

• Sales calls and notes. The questions prospects ask in the first call are the questions they asked an assistant the night before.
• Support tickets. Especially from new customers, who were recently buyers.
• Search Console. The queries that already bring people to your site, particularly long, question-shaped ones.
• People Also Ask. The questions Google itself groups around your key terms. Semrush found People Also Ask alongside 90.03% of AI Overviews, so these are close to the questions Google's own AI answers are built around.
• Community threads. Reddit and industry forums, where people ask for recommendations in their own words.
• Review sites. The phrases buyers use when they explain why they chose or left a product.

Filter: keep the prompts that can name you

Run each candidate prompt once before adding it to the set, and look at the answer.

• Keep prompts whose answers name vendors, recommend products or cite sources of the kind you publish.
• Move to your content plan prompts that return only a definition. They are worth a glossary page or a guide, not a tracking slot.
• Cut prompts that return nothing relevant to your category, and near-duplicates that differ only in wording.

Neighbouring questions matter

Assistants do not always answer the literal question alone. Google describes AI Mode as "breaking down your question into subtopics and issuing a multitude of queries simultaneously on your behalf." A buyer's shortlist question may be answered using pages written for the implementation and comparison questions around it.

That is an argument for covering each stage with several related prompts rather than one perfect one, and for writing content that answers the follow-up questions as well as the main one.

Keep it stable

The value of the set is in comparing it with itself over time. Freeze it for at least a quarter. When the market changes and new questions appear, add them as a separate cohort, report them separately, and fold them in once they have a few periods of history. Retire a prompt only when it no longer reflects how anyone buys.

Turning gaps into work

For every prompt where you are absent, look at what the answer cites instead. The sources tell you what to do:

• A competitor's page is cited: you need a page that answers the question better, with sourced claims. See How to Get Cited.
• A third-party list is cited: you need to be on that list, which is outreach rather than content.
• A community thread is cited: you need genuine participation where the thread lives.
• Nothing relevant is cited: there may be no good answer yet, which is the easiest gap to close.

Doing it in Veritas

Prompt Discovery in Veritas finds the buyer questions in your category where your brand is absent, and AI visibility tracking runs the set across ChatGPT, Perplexity, Gemini and Google AI Overviews. Tracking allowances depend on the plan: 10 prompts on Starter, 25 on Growth and 50 on Business. Start with the prompts closest to purchase, mostly from the shortlist and comparison stages. For dedicated tracking tools, see the best AI visibility tools.