Learn how to research the questions your customers actually ask AI, build a smarter prompt tracking list, and improve your brand’s visibility in ChatGPT, Gemini, and Google AI search.

Getting the questions and prompts right is the foundation of your AEO work, so it's worth having a thorough process for it. That process starts with your business and your customers' real language, not with a tool's suggested prompts. You already know what you sell and who buys it. The work is translating that into the way customers describe their problem when they ask AI, and picking the two or three services that matter most commercially, so your list starts where the money is.

In practice, it comes down to five steps, the same ones we use with our own clients:

  • Start with the problems customers have, and the questions they ask about them.
  • Rewrite them the way a customer would type them into AI.
  • Cut anything you don't sell, or that won't get a business recommended.
  • Track the same list every month, and only change it once a quarter.
  • Act on what it shows you.

Why tool-suggested prompts fall short

Most AI search tools start by reading your website, writing their own summary of your business, and suggesting prompts from that summary. So the list reflects how your website describes you, not how customers ask, and tracking it mostly measures how well AI repeats your marketing back to you.

Run with those suggestions, and you'll get vanilla, sausage-factory results: much the same generic questions as everyone else in your category, measured against your own marketing.

What you need instead is a reliable prompt research process, one you can repeat, that starts with how your customers actually search, the words they use and the questions they ask, and ends with a short list of prompts worth tracking every month. Everything else you do is built on that list: the content you write, the fixes you make, and the results you report to the business. Get the list wrong, and all of it is aimed at the wrong questions.

Start with the customer's problem

A good list starts with the customer's problem, not your product, and covers the whole buying journey from there. Customers ask three kinds of questions:

    • What to do about their problem, and whether they need help: “do I need a lawyer to write a will, or will an online kit do?”
    • Which option to choose: “best wills lawyer for a blended family”
    • How to buy: “fixed-price will in Auckland”

Tools mostly suggest the last two, because they start from your product. Customers often start somewhere else: a dairy farmer asks why their cell count is rising, not which teat spray to buy. That first stage is where AI recommends your whole category or talks people out of it. We've seen this first-hand. In one recent trial, AI answered a “do it yourself or get help?” question by recommending a professional, then named the business we were working with as one to use. That's a recommendation made before the person has even started comparing providers.

The anatomy of a prompt

A good prompt has four parts. Take “Recommend a marathon shoe for a heavy heel-striker under $150”:

    • What the person wants: “recommend”. It's usually recommend, compare, or find, and it comes straight from the questions you collected.
    • The product or service: “a marathon shoe”. Use the words your customers use, not your own product names. People Also Ask, and your Search Console data show you what those are.
    • Their situation: “a heavy heel-striker”. Often it's the problem they're trying to solve. It's the part most businesses leave out, and your reviews supply it, because they tell you who bought and why.
    • Their limits: “under $150”, or timing, or location. Only add the ones customers actually type.

When you write each prompt, check it has all four parts, and fill any gap from your research rather than your own assumptions.

Filtering and tracking your list

Cut any prompt that fails these checks. Does the answer recommend businesses, or is it just a fact or a how-to? Those belong on your website, not your tracking list. Is it something you sell? Is it different from the prompts you already have? Would a customer actually type it?

Before you lock a prompt in, run it once and look at who AI names. If the answer only cites government sites and official bodies, you'll struggle to break in, so cover it on your website and leave it off the list. If it names your competitors, directories or comparison sites, that's a gap you can close, and it belongs on the list even though you're not there yet. How easily you can close each gap decides the order you work on them.

Build up to 20 to 30 prompts, spread from “do I need this” to ready to buy. Run them monthly, starting with ChatGPT, Google's AI Overviews and Gemini, where most of the use is, and note who's named. Keep the same list for a quarter, because AI answers vary: even across the 80 or so prompts we track each month, the overall score moves about five points either way between runs. A slightly imperfect set you've tracked for three months is worth far more than a perfect one you've only checked once.

Where to start, and what to measure

Getting started comes down to three steps: build a baseline, act on what it shows, and track what moves each month. The baseline matters because it shows who AI recommends instead of you and which websites it trusts, which is where your gaps are. Tracking the same prompts each month then gives you evidence you can take to the business, not just a one-off check.

Step 1: Build your baseline

It takes about an hour, and you don't need a tool:

    • List what you sell that matters most.
    • Pull questions from People Also Ask, Google autocomplete, your Search Console data, and your reviews.
    • Ask whoever answers your phones what customers ask them.
    • Write your first 10 to 15 prompts the way customers would type them, including two or three “do I even need this” questions, without naming your business.
    • Run them in fresh chats while logged out, and record in a spreadsheet which businesses were named and which websites were cited.

Step 2: Act on what it shows

Your baseline shows who you're really up against, often publishers, comparison sites, and government pages, and which websites AI trusts. Work through it like this:

    • If AI answers without you, check whether you already have a page it's ignoring, and if not, write one.
    • If you're missing from the review sites AI cites, ask customers to review you there.
    • If AI has your details wrong, fix them on the website it's taking them from.
    • If a competitor is named first, look at the websites cited for them.
    • If AI sends customers elsewhere on topics you already pay for in Google Ads, that's a budget conversation.

Step 3: Track what moves

Run the same prompts every month and keep an eye on three things:

    • Which businesses get named across the same prompts over time. It's the most reliable signal, and the first to move.
    • Visits from Google Analytics' AI Assistant channel, added in May 2026. It shows some of the visits AI sends you, but not all: someone who researches in ChatGPT and then types your name often shows up as Direct, as our earlier Direct traffic piece explained.
    • Searches for your business name in Search Console, and answers to your “how did you hear about us” question once you add “an AI assistant like ChatGPT” as an option.

There's no reliable data on how often people ask a given prompt, so treat any tool's prompt volumes as estimates. Start with the hour and the spreadsheet.

Frequently asked questions

We found these questions the same way this article suggests you find yours. Two came from Google's People Also Ask and two from Google autocomplete, both searched from New Zealand, and we've kept the wording close to how people typed them.

Are search queries and AI prompts the same thing?

No. A Google search is usually a few words, like “CRM for small business”. A prompt reads like a conversation and carries the person's situation, such as their team size, their budget or the problem they're trying to solve. The same person can type both and get completely different answers from Google and ChatGPT.

How do I track brand mentions in ChatGPT?

Run the same set of customer questions in ChatGPT each month, in fresh chats while logged out, and record whether you're named, who's named instead and which websites are cited. Google Analytics' AI Assistant channel will show some of the visits ChatGPT sends you, though not all of them.

What tools track AI visibility for a brand?

A growing number of platforms run your prompts across ChatGPT, Gemini, Claude, Perplexity, and Google's AI Overviews on a schedule, including add-ons from SEO platforms such as Semrush and Ahrefs. They earn their place once you're tracking more prompts than you can check by hand, but you don't need one to start.

What is the difference between brand awareness and brand visibility?

Awareness is whether people already know your business. Visibility is whether you show up when they go looking, and that now includes AI answers. A well-known brand can still be missing from ChatGPT's recommendations, which is why it's worth checking the questions customers actually ask rather than assuming awareness carries over.

See where you stand

If you'd like a head start on your baseline, ADMATIC offers a free first look at how AI search treats your business. We run a set of real customer questions about your category through the main AI platforms and show you where you're named, who's named instead, and what's worth looking at next. Request your free first look.

Author: Rod Russell, Managing Partner, ADMATIC, 11th November 2026