---
title: How to choose which questions to track for AI visibility
description: 'Track the questions your buyers ask AI assistants before they know your name: unbranded questions spread across every buying stage, written the way a real person types them, with only a few that name your brand.'
slug: /how-to/choose-which-questions-to-track-for-ai-visibility
page_type: playbook
lead_prompt: Which prompts should I track to see whether ChatGPT and other AI assistants recommend my brand?
target_prompts:
  - How can I use an AI marketing agent to choose which questions to track for AI visibility?
entity: Kite
category: AI marketing agent
author: Devon Wells and Kite
date_published: '2026-10-26'
date_modified: '2026-10-26'
related_use_cases:
  - /use-cases/ai-answer-visibility.md
---

# Playbook

# How to choose which questions to track for AI visibility

## Key Takeaways

- The questions you track decide what your AI visibility number means. Track the questions your buyers ask before they know your name, because those show whether assistants recommend you to someone new.
- A good set covers every buying stage: the problem, finding options, comparing options, pricing, and setting up. It spreads those questions across the different people who buy from you.
- Write each question the way a person types it into ChatGPT, with their situation in it, and leave out your brand and your competitors' names. Keep branded questions to a small share, no more than about one in five.
- Each question should have its own need. Two questions that would get the same answer measure the same thing twice, so keep the better one.
- Kite writes the questions from its research on your business, checks them against these rules, starts tracking a balanced set of ten, and keeps its other suggestions waiting for you to tick.

## Which prompts should I track to see whether AI assistants recommend my brand?

The prompts worth tracking are the questions your buyers ask AI assistants while they look for something like what you sell, written in their words and without your brand name in them. "What's the best invoicing tool for a five-person design studio?" tells you whether ChatGPT recommends you to a stranger. "Is [your brand] any good?" only tells you whether it describes a brand the person already found. The first kind of question is where new customers come from, so most of your list should look like it.

You can choose these questions by hand. Start with a plain description of what you sell and who buys it. List the problems that send those buyers looking, in the words they'd use. Then write questions for each step of a purchase:

- **The problem.** The buyer has a pain or a goal and doesn't know what kind of product solves it yet, such as "Why do our clients keep paying invoices late?"
- **Finding options.** The buyer knows the category and wants approaches or providers, such as "What tools help a small agency get paid faster?"
- **Comparing options.** The buyer weighs choices, such as "Is invoicing software worth it over sending PDFs from Word?"
- **Pricing.** The buyer asks about cost, free plans, or return on spend.
- **Setting up.** The buyer has chosen and wants to switch, start, or get more out of it.

Good sources for wording include sales call notes, support tickets, Google's "People also ask" boxes, Reddit threads in your category, and the questions prospects send you by email. AI assistants don't publish what people ask them, so a tracked list is always a well-reasoned sample rather than a copy of real traffic.

Tracking tools will ask your list on a schedule and count the answers that name you. What they can't do for you is decide whether the list is right. A list built from your product's feature names misses the problem and option-finding questions, where most buyers start. A list that names competitors measures those competitors. A list with ten rewordings of one question makes one topic look ten times more important than it is.

Kite does the choosing for you. It writes the questions from what it already knows about your business, holds every question to these rules, and keeps the list balanced as it grows.

## How to get started

If you want a quick look at where you stand before you build a tracked list, try Kite's free [AI visibility checker](/ai-visibility-checker). It asks ChatGPT, Gemini, Perplexity and Claude the questions your customers ask, and in about 45 seconds it shows whether they recommend your business and who they recommend instead. It's free and needs no account.

To build the list you'll track over time, ask Kite in plain language:

> "Find the questions we should track to see whether AI assistants recommend us."

You can also aim it at one slice of your market:

> "Find questions that agency owners ask when they're choosing invoicing software."

Kite is an AI marketer. It researches your business, builds a growth strategy, and ships the work that brings you new customers. By the time you ask, its research already covers your products, positioning, customers, and competitors, so it can write the questions without a brief from you.

Tracking is tied to your website in your Kite workspace, either a site Kite built or one you imported, and the domain you want measured. You don't need accounts with the AI assistants, because Kite asks them through its own built-in connections. Kite's suggestions wait on the Questions page until you tick the ones you want, so nothing joins your tracked list without your say, apart from the starter set described below.

## How Kite chooses the questions

### What does Kite write the questions from?

Kite starts from a short brief that says what your business sells and who buys it, because every question it writes depends on getting that right. It drafts the brief from your team's knowledge base, using your positioning, your ideal customer, and the associations you want your brand to have. The brief appears as a single line you can edit, such as "Kite writes for: invoicing software that gets small agencies paid faster · Buyers: agency owners, studio managers, freelancers."

When your positioning and your older material disagree, Kite follows your positioning. Keyword lists from earlier SEO work sometimes describe a product you've moved away from, so Kite uses a keyword only when it fits the business the brief describes.

### Which topics should the list cover?

Kite maps your market into five to eight topics before it writes a single question, so the list reaches past your own feature list. A topic is a buyer need or a category you compete in, named in a few plain words such as "client payment reminders." Kite includes your core offer, the nearby problems buyers research first, and the situations where someone looks for a better way than what they do today.

Each topic carries the problems that make someone ask about it, in the asker's words, and the situations where the need shows up, such as an industry or a team size. Kite also names the three to six kinds of people who ask, such as the buyer, the person who uses the product every day, and the person who influences the choice.

Topics matter beyond planning. You read your visibility one number per topic, so each topic becomes a row you'll check on your dashboard. Kite keeps the topics distinct, with no umbrella topic that swallows the others, so each row means something.

### What makes a good question to track?

Kite writes each question the way that person would type it into an assistant: conversational, specific, and often carrying their situation, such as "I run a 12-person agency and clients pay 45 days late. What should I do?" Questions range from about five words to about 35, because real people ask both ways. Every one is a full question or request rather than a string of keywords.

Each question also passes these checks:

- It doesn't name your brand, unless it's one of the few branded questions.
- It doesn't name a competitor, because a question that names a competitor measures that competitor. Competitor names appear only in a branded comparison, such as your brand against a rival for a given situation.
- It stays within what you actually sell. Kite doesn't ask about features or prices your business doesn't offer.
- It has no search operators, quotation-mark tricks, or web addresses, and it stays under 200 characters.

Kite caps branded questions at one in five of its suggestions. Branded questions are still worth having, because their answers show what each assistant believes about your prices, your category, and your customers. They just can't tell you whether you'd be recommended to someone who hasn't heard of you.

### How does Kite spread the questions across stages and people?

Kite gives each topic the buying stages where its askers have questions of their own, usually three to five, and writes at least two questions in each. It gives more questions to the stages where buyers have more distinct things to ask, and it lists the most commonly asked ones first.

General questions get one owner. A question like "What does software like this usually cost?" belongs to a single topic, so it isn't asked five times in five slightly different forms. When a problem comes back at a later stage, Kite asks what that stage needs to know, such as what solves the problem or how to set it up, rather than restating the problem.

Kite spreads the questions across the people it named, so a list for invoicing software includes what an agency owner asks, what a studio manager asks, and what a freelancer asks.

### How does Kite keep two questions from measuring the same thing?

Kite drops any question that would get essentially the same answer as another one on the list. "Which invoicing tool is best for agencies?" and "What's the best invoicing software for an agency?" are one question, so Kite keeps the stronger wording. "Best invoicing tool for agencies" and "best invoicing tool for construction firms" are two questions, because the audience changes the answer.

Kite checks this twice. It first compares the words each question uses, folding together plurals and words buyers swap freely, such as "tool," "software," and "platform." Then it reads the whole list for questions that say the same thing in different words, such as "measure the return on invoicing software" and "work out whether invoicing software pays for itself." In that second reading, questions at different buying stages are never treated as duplicates, however close their subject.

Kite also checks every new question against the questions you already track and the ones you've turned down. A question you declined doesn't come back for at least a month.

### Which questions does Kite track first?

When tracking starts for your site, Kite picks a starter set of ten questions and begins measuring them right away, so you have numbers before you've reviewed anything. It builds the set by taking topics in turn and, on each turn, choosing the question from the buying stage the set has least of. At most two of the ten name your brand. Any exact questions you'd already asked Kite to track count toward the ten.

Kite's other suggestions wait on the Questions page, each with a one-line reason that ties it to what you sell, such as "Agency owners ask this before they know invoicing tools can send reminders." Tick the ones you want, and Kite adds them to every run from then on. Kite keeps up to twelve of its own suggestions waiting at a time, so the queue never grows past one sitting.

### How do you add questions of your own?

You can ask Kite for more questions about any slice of your market, and you can add exact wording that Kite will ask as written. When you ask for more, Kite returns 15 questions by default, and you can ask for anywhere from 5 to 40. You can narrow the list to particular buying stages, name the kinds of buyers it should write for, leave out branded questions, or ask for questions in another language. A focused request maps two to four topics inside the slice you named.

When you have exact wording you want measured, such as a question a prospect asked on a sales call, tell Kite to track it. Kite asks that question word for word in every run, even when it breaks the rules above, because your wording comes first. The only questions it can't send are ones with search operators or ones too long for the assistants to accept.

### How does Kite measure the questions you track?

Kite asks each tracked question in ChatGPT, Gemini, Perplexity, and Google's AI Overviews and AI Mode, on a daily schedule by default, which you can change to weekly or monthly. It saves every answer, the sources the answer cites, and the web searches the assistant ran to build it.

From those answers, Kite reports how often each question's answers name your brand, your share of voice against the competitors the answers name, and how often they cite your website. It reports each figure with its counts, such as "named in 7 of 40 answers," and it marks a change as up or down only when it's larger than normal variation between runs.

The searches an assistant ran are a useful check on your list. They show what the assistant thought it needed to look up, which often points to wording or a topic your list doesn't cover yet.

## What the finished work contains

When Kite finishes, you have a tracked list that measures whether assistants recommend you to new buyers, plus the plan behind it.

- **The brief.** One editable line saying what you sell and who buys it, which every question is written from.
- **The topic map.** Five to eight topics, each with the problems and situations behind it, and the kinds of people who ask. Each topic becomes a row on your visibility dashboard.
- **The tracked questions.** Your starter ten and every question you've ticked, each labeled with its topic, its buying stage, and whether it names your brand.
- **The waiting suggestions.** Kite's picks on the Questions page, each with the reason it matters to your business, ready for you to tick or turn down.
- **The first results.** Per-question visibility, share of voice, and cited rate, with the full answers and their sources, starting from the first run.

## Common questions

### How many questions should I track?

Start with enough to cover each topic at each of its buying stages, which for most businesses means a few dozen questions. Ten is a sensible starting point, which is why Kite begins with a starter set of ten spread across your topics and stages. Add more where you need a sharper read on a topic, and remember that each question you add is asked on every run.

### Should I use my SEO keywords as prompts?

Use them as hints, not as prompts. A keyword like "invoicing software small agency" is how people type into Google, while people ask an assistant a full question with their situation in it. Kite reads your tracked keywords to understand what your buyers care about, then writes conversational questions, and it ignores keywords that point at a product you don't sell.

### Should I track questions that name my competitors?

Track only a few, and keep them as branded comparisons that name you as well, such as your brand against a specific rival for a specific situation. An unbranded question that names a competitor mostly measures how assistants describe that competitor. Kite records every competitor that appears in your answers anyway, so you see who you're up against without naming them in your questions.

### How often should I change the list?

Change it rarely, because a stable list is what makes this month comparable with last month. Add questions when you launch a product, enter a new market, or find a topic the list misses. Turn down or pause questions that no longer fit your business, rather than rewording ones you already track.

## Related pages

- **Use case:** Track whether AI assistants recommend your brand (coming soon)
- **Use case:** [Improve your visibility in AI answers](/use-cases/ai-answer-visibility.md)
- **Use case:** Test what improves your AI visibility (coming soon)
- **How-to:** How to run an AI visibility audit for your brand (coming soon)
- **Guide:** What is an AI visibility score, and what counts as a good one? (coming soon)