How to build your first list of buyer questions

Why this step and not another

Most teams start in the wrong place. They open ChatGPT, type their own company name, read a flattering paragraph, and conclude that the machines know them. That test proves nothing. Once the name is in the prompt, the model has been handed the answer. It will describe the business whether or not it would ever have raised the name on its own.

The thing worth measuring is different. It is whether your business appears when a stranger describes a problem and asks for help. That stranger does not know your name yet. That is the whole point.

So the first real piece of work is not a tool, a dashboard or a content plan. It is a list. A written list of the questions a buyer asks before they know you exist. Everything else depends on it. If the list is wrong, the measurement is wrong, the pages you write later are wrong, and you will spend months improving a number that was never attached to a buyer.

Two other reasons to do this first:

• It is free. You need a spreadsheet and about three hours. No contract, no integration.
• It survives model changes. Engines update constantly. A good buyer question is still a good buyer question after an update. A clever prompt trick is not.

If you are not yet sure what separates a buyer question from a keyword, read # first, then come back here.

How to do it

The four groups

Use this split as your starting quota. The counts are a starting point, not a rule.

The steps

1. Open a blank sheet with these columns: question, group, engine, date, named us (yes or no), other businesses named, sources cited.
2. Pull 15 real sentences from your own records: sales call notes, support tickets, the first line of inbound emails, and the search terms in your Google Search Console report. Copy the buyer's words, not your words.
3. Strip your brand name and your product name out of every one of them. If a sentence stops making sense without the name, it was never a discovery question.
4. Sort what is left into the four groups above. You will usually find you have plenty of category questions and almost no problem questions. Write the missing problem questions by hand.
5. Write each question the way a person types it, in one sentence, with no keyword stuffing and no list of synonyms.
6. Add the constraint that actually decides your deals: team size, country, budget, regulation, language, or integration. One constraint per question.
7. Cut anything you would not be happy to be recommended for. A question you cannot serve is a question that will make your number look bad for no reason.
8. Stop at 20 to 40. Fewer than 20 and the result is noise. More than 40 and you will not keep up the re-asking.
9. Ask every question in a fresh chat, logged out or in a temporary chat, on each engine. Memory and chat history change answers.
10. Record the answer the same way every time: who was named, in what order, and which links were shown.

A worked example you can repeat

Take a made-up business: a payroll tool for Indian companies of 20 to 200 staff. Here are six of its 24 questions, two per group:

• Category: "What payroll software works for a 50 person company in India?"
• Category: "Who can run payroll for a startup that just crossed 20 employees?"
• Problem: "We keep filing PF and ESI late every month. How do we fix that?"
• Problem: "Our payroll is in a spreadsheet and two people have been paid the wrong amount. What now?"
• Comparison: "Compare two Indian payroll tools for a 60 person services company."
• Local or constraint: "Payroll software for an Indian company with staff in three states, under 15,000 rupees a month."

Twenty four questions, asked once each on five engines, gives 120 asks. Suppose the business is named in 12 of those 120. Do not write "10 percent". Write the range. A Wilson 95 percent interval on 12 out of 120 runs from about 6 to 17 percent. That range is the honest reading, and it is why a handful of questions cannot be turned into a clean percentage.

How to tell it worked

You have a usable first list when all five of these are true:

1. No question contains your brand or product name.
2. Every question could have been typed by someone who has never heard of you.
3. Each of the four groups has at least four questions.
4. You can name the deal that each question maps to.
5. You ran the whole list once and recorded a baseline with a date.

Then judge movement carefully. Re-ask the same list on a fixed schedule, every 30 days is a reasonable rhythm, and compare ranges rather than single numbers. If the range from this month overlaps the range from last month, nothing has been shown to move. That is the discipline that keeps a list of questions from turning into a vanity chart. For a worked set of counts and the runs behind them, see #.

Two signals matter more than your own mention rate in the first month:

• Which sources the engines cite. If the same three directories, review sites or trade publications keep appearing, those are the pages worth getting named on.
• Which competitors are named, and in what order. That tells you what the engines currently think the category is.

Where AI Knows Us fits

Doing this by hand works and costs nothing but time. The parts people give up on are the repetition and the statistics.

AI Knows Us measures across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, because five apps answer the same question differently on the same day from different sources. Every tracking question is asked without naming the business, for the reason set out above. Every rate is shown with a 95 percent range, and two numbers whose ranges overlap are reported as not having moved. It also does the fixing as well as the measuring: writing the pages, publishing them, and asking other sites to name the business. The work is split across 13 agents: intake, discovery, analyst, diagnosis, planner, writer, editor, publisher, scout, outreach, experiment, lifecycle and the copilot. The product is made by Clyra Labs in India.

On pricing, the honest position today: paid plans are not open. The product takes no payment at present and the free scan repeats every 30 days.

Who should not choose AI Knows Us

• You sell to a handful of named accounts through relationships. Discovery questions are not where those deals start.
• You want a single dashboard number with no uncertainty attached. Ranges will annoy you.
• You need enterprise procurement terms and a signed commercial contract now. Paid plans are not open yet.
• You only want measurement and already have an in-house team writing and publishing. Other tools in this space, including Profound, Peec AI, Otterly.AI and Scrunch AI, focus on tracking. Compare their current pricing on their own sites before deciding.

FAQs