Building Your Question Panel
AI visibility is measured in questions, not keywords. This chapter shows you how to build a fixed panel of fifteen to twenty-five buyer questions across five families, run it consistently across the engines, and avoid the comprehensive audit you'll never repeat.
Every measurement system needs a fixed instrument. In classic SEO, that instrument was a keyword list. In AI visibility, it's a question panel: a fixed set of questions, asked the same way, across the same engines, at regular intervals. This chapter is about building yours, and it's the single highest-leverage afternoon in this book.
Why questions, not keywords
People don't type keywords at an assistant. They ask it things, in sentences, the way they'd ask a knowledgeable friend. "Physio Brunswick" becomes "who's a good physio in Brunswick for a running injury?" The engine's answer to the full question is what your buyer sees, so the full question is what you measure. Your keyword data still helps you find the questions. It just stopped being the unit.
The five question families
A good panel covers five families, because they map to five different moments in the buying journey and engines answer each differently.
Category questions ask for a shortlist: "best commercial cleaning company in Melbourne", "who should I use for X". These are the recommendation moments, the highest-stakes answers to be in. Cost questions ask what things cost: "how much does a rebrand cost in Australia". Buyers ask these constantly, businesses answer them rarely, and the gap between the two is where citations are won cheaply. Comparison questions weigh options: "X vs Y", "is A better than B for C". Brand questions are about you by name: "is [your business] any good", "[your business] reviews", "who owns [your business]". Problem questions describe the situation without naming the service: "my website doesn't show up on Google, who can help". They're the hardest to win and the most valuable, because the engine is doing the diagnosing and whoever it names gets a buyer who hasn't started comparing yet.
Writing the panel
Aim for fifteen to twenty-five questions across the five families, and write them in your buyer's language, not yours. If your industry says "remedial works" and your customers say "fixing the cracks", the panel says cracks. Add locality wherever the buying decision is local, because "best accountant" and "best accountant in Geelong" produce different answers with different names in them. Pull the raw material from three places: what people say on the phone before they become clients, the questions on your enquiry forms, and your search query data.
Then freeze it. The panel only works as a measurement instrument if it stays fixed, because the thing you're tracking is movement in the answers, not variety in the questions. Change the questions every month and you've measured nothing. Review the panel twice a year, retire questions deliberately, and log the change when you do.
The constraint worth writing down: do not build the two-hundred-question panel. You won't run it monthly, and a big panel you abandon is worth less than a small one you run every month for a year. Twenty questions, run consistently, will tell you more than any one-off comprehensive audit ever could.
Running it
Ask every question in each engine that matters for your market. For most Australian businesses right now, that's ChatGPT, Google's AI Overviews, and Perplexity at minimum, with Gemini and Copilot worth adding if your buyers live in Google Workspace or Microsoft 365. Use a clean session, not your logged-in personal account with months of history, because the engines personalise and you want the answer a stranger gets. Save every answer verbatim with a date. What you now have is a grid: questions down the side, engines across the top, answers in the cells. The next chapter runs your brand through it, and chapter six turns the grid into a baseline you can act on.