How to Choose Prompts to Track for AI Visibility

The prompts you track decide what every number on your dashboard means. Here is how to choose them: real questions, grouped by topic, worded several ways, mostly without your brand name.

Every number on your AI visibility dashboard means exactly what your prompt set tells it to mean. A tracked prompt is a question A2ZReach asks the AI assistants on your behalf, worded the way a buyer would ask it. Our reading: track real questions a buyer would ask, grouped by topic, worded several ways, mostly without your brand name.

If your set is a list of keyword strings, a handful of near-identical questions, or mostly prompts that name your brand, the numbers will look precise and say very little. This guide walks through how to build a set that measures something.

What Is a Tracked Prompt, and How Is It Different from a Keyword?

A tracked prompt is a natural-language question that A2ZReach asks the AI assistants for you. A keyword is a phrase you optimize a page for. The two have different shapes: "best running shoes marathon training" is a keyword, and "What are the best running shoes for marathon training under $150?" is a prompt.

Tracking in A2ZReach has three layers: your brand, then Topics, then Prompts. A Topic is a short label for a question category buyers care about. A Prompt is one specific way a buyer would ask it, in conversational language. Visibility is measured per prompt and rolled up to the topic. The Topics and Prompts guide covers the screen itself.

Questions are the right shape because that is how people use AI assistants. An NBER working paper (September 2025) ↗ (opens in a new tab), written with OpenAI's economic research team and covering ChatGPT's consumer product from November 2022 through July 2025, found that "Practical Guidance" and "Seeking Information" were two of the three most common conversation topics. With "Writing", the three made up nearly 80% of all conversations. The paper is a working paper, not peer-reviewed, and covers ChatGPT only, so it tells you about the shape of a request, not about shopping or brands.

Where Do Good Prompts Come From?

Good prompts come from the questions your buyers already ask, not from a keyword tool. Our suggestion: start with three sources.

• Sales calls and support email. The words a prospect uses to describe their problem are usually better prompts than anything written at a desk.
• Your existing content. The questions your blog posts, FAQs and help pages answer are questions someone asked. Work backwards from the headline.
• The choices buyers weigh. Buyers compare kinds of products and ways of solving a problem. If your sales team hears "what is the difference between X and Y" in your category, that is a prompt.

A2ZReach's setup suggests topics and prompts for you to review. Treat those suggestions as a starting point and check them against your own customers' language before you rely on them.

Should a Prompt Name Your Brand?

Mostly no. A prompt that names your brand is tracked for Sentiment and is left out of the Visibility Score, so keep a few of those and make most of your set unbranded.

Each prompt in A2ZReach has a Type, Visibility or Sentiment. A Visibility prompt is unbranded. It measures whether your brand shows up when someone asks a question without naming you, which is where you compete to be included. The Visibility Score is the percentage of answers to unbranded prompts in which your brand appears at least once. Share of Voice uses the same unbranded prompts.

A Sentiment prompt names your brand. It measures what the AI assistants say about you when asked directly. A prompt that names your brand is left out of the headline numbers because an answer to it will almost always mention you, which would raise the score without telling you anything. If a prompt names your brand, A2ZReach locks the Visibility option off for it.

How Many Prompts Do You Need?

There is no single right number. Start with a handful of topics and a few prompts under each, then grow the set as new angles appear. The plan you are on sets the limit.

That is how A2ZReach's documentation puts it: quality matters more than quantity. A gauge at the top of the Prompt Volumes page shows how many prompts you have used against your plan's limit.

A January 2026 study by Rand Fishkin (SparkToro) and Patrick O'Donnell (Gumshoe.ai) ↗ (opens in a new tab) points the other way on size. Its author concluded that visibility percentage across dozens to hundreds of prompts, run multiple times, is a reasonable metric, though a rough one. He also said more work is needed to learn how many different prompts a sector needs for a statistically meaningful answer. The study is informal and not peer-reviewed, and Gumshoe.ai sells AI-tracking software, which the authors disclose. Our reading: a small set is a fine place to start, a larger one gives a steadier number, and neither is a target.

Why Should You Word the Same Question Several Ways?

Because people word the same need very differently, and one phrasing is one data point. Our reading: write two to four versions of each question and let the topic roll them up.

In the same SparkToro and Gumshoe.ai study, 142 people were asked to write a prompt for the same task, headphones for a traveling relative. All 142 prompts were different. The average semantic similarity between pairs was 0.081, which the author compared to "Kung Pao Chicken and Peanut Butter". Yet across the 994 responses, a handful of well-known brands appeared in 55 to 77% of them, which the author read as the AI assistants picking up the underlying need despite the different wording.

A narrow category looked different. A prompt about brand design agencies, in the same study, produced far more varied answers, and its top brands reached only 30 to 40% visibility, against 90 to 100% for the headphone prompts. If your market is specialized, expect more spread in your numbers. The study is small and informal, so treat it as a reason to vary your wording, not as proof of a method.

How Do You Write a Good Prompt?

A good prompt reads like something a person would actually type: specific, conversational and one question long.

• Be specific. Name the situation, the use case or the constraint. "Running shoes for marathon training under $150" shows more than "running shoes".
• Add an angle. "What do podiatrists recommend for flat feet running shoes?" brings in an expert view and a specific need. Generic prompts get generic answers.
• Keep it conversational. Write the way someone talks, not the way a meta description reads.
• Avoid keyword strings. "Running shoes sustainable eco material wool" is a keyword list, not a question anyone asks.
• Ask one thing at a time. A prompt that asks two things is harder to read when the answer comes back.

A quick before and after:

The weaker versions say little about the buyer's situation. The stronger ones say more.

What Does a Good Prompt Set Look Like?

Two topics, a few unbranded prompts under each, and one brand-name prompt kept for Sentiment is a workable starting shape.

The table below is an illustration for a made-up brand, "Example Brand", that sells running shoes. The prompts are invented to show the method. They are not data, and they are not a recommended set for any real business.

How Do You Keep the Set Healthy?

Archive prompts rather than delete them, and treat any change to the set as a change to your baseline.

An archived prompt stops being tracked, and you keep what was collected before it. A2ZReach moves archived prompts to an Inactive tab, and nothing is deleted for good until you say so.

Our suggestion: note the date you add or archive prompts, and compare before and after with care. A jump in your Visibility Score after you add ten new prompts may be a real gain, or it may be a different set of questions. The date helps you tell which.

Review the set when you launch a product, when your sales team starts hearing a new question, or when a topic stops feeling relevant. You do not need a fixed schedule. You need the habit of asking whether the prompts still sound like your buyers.

Where Do You Manage Prompts in A2ZReach?

On the Prompt Volumes page, under GEO in the left panel. Topics are listed on one side, and the prompts of the topic you pick are on the other.

You can add and edit topics, add prompts, and archive, restore or delete prompts one at a time or many at once. To add many prompts at once, the page offers a template you fill in as a .xlsx or .csv file (up to 1,000 rows and 2 MB) and then import. The Topics and Prompts guide covers each step.

What Should You Do This Week?

Our suggestion for a first pass:

1. Pull three to five real questions from your sales call notes, support inbox or onboarding conversations. These are your first prompts.
2. Group them into topics. If two questions are about the same buyer concern, they belong under the same topic.
3. Write two to four versions of each question. Change the framing, the buyer's situation or the constraint. Keep each one conversational.
4. Check which prompts name your brand. Those are tracked as Sentiment only. Leave the rest as Visibility.
5. Note today's date. Any change you make later is a change to your baseline, so the date is your reference point.

FAQ

Should I track my brand name as a prompt?

Yes, but keep it apart from your Visibility prompts. A prompt that names your brand is tracked as Sentiment only, and A2ZReach locks the Visibility option off for it, because an AI assistant asked about you directly will almost always mention you. These prompts show what AI assistants say about you. They do not count toward your Visibility Score or Share of Voice.

Can I add many prompts at once?

Yes. The Prompt Volumes page lets you import a .xlsx or .csv file using a template the page provides, up to 1,000 rows and 2 MB per file. You can also add prompts one at a time from the same page.

See How Prompts Roll Up on Example Data

The prompts you choose are the foundation of every number on your dashboard. Choose them as questions a real buyer would ask, group them by topic, word each one several ways, and keep brand-name prompts in a separate Sentiment bucket.

To see how prompts roll up into topics, browse the public sample report (labeled as illustrative data, not a measurement of any real brand). To talk through your own category, book a live demo: a 30-minute visibility review with one of our specialists.