# Which Prompts Should You Actually Track?
How to build an AI prompt tracking set that gives useful data: the four prompt families, where to source prompts, how many to track, and when to cut.
**Published:** September 30, 2026
**Author:** Kate Starr & Connor Lahey
---
An AI visibility dashboard can show 83% visibility and still tell you very little if most of that visibility comes from prompts that already mention your brand.
Understanding your AI visibility is only useful if you’re measuring it with the right prompts.
In marketing, prompt tracking means monitoring how your brand appears when AI engines answer questions about your market.
In this article, you’ll learn how to build a prompt set that gives you data you can act on.
## What is prompt tracking, and why does prompt selection matter?
Prompt tracking is the process of running the same questions across AI engines over time and [measuring how your brand appears in the answers](https://www.searchable.com/blog/why-ai-visibility). You can track whether your brand is mentioned, which competitors appear, and how your brand’s visibility changes over time.
The tracking process is easy to automate. The harder part is choosing which prompts to track, because each prompt measures something different.
For example, “What are the best CRMs for a small sales team?” shows whether your brand appears when someone is looking for products in your category. “Is HubSpot worth it?” measures something else, because the prompt already names a brand.
Both prompts can be useful, but if your list includes too many questions that measure the same thing, your overall visibility score will be less useful. Instead, build a prompt set that covers the questions you care about without filling it with duplicates or prompts that give you a distorted view of your visibility.
## Why your starting prompt set needs a review
When you start using an AI visibility tool, it usually gives you a list of prompts to track. That list becomes your prompt set, and in many cases, it stays largely unchanged after setup.
Across more than 1.6 million tracked prompts in nearly 39,000 Searchable projects, about 53% came from onboarding. This is because the onboarding list gives users a useful starting point without having to build everything from scratch.
Where tracked prompts come from
1,641,972 tracked prompts across 38,826 Searchable projects, as of September 30, 2026.
53.0%
47.0%
From the onboarding listAdded outside onboarding
That first list is created before you have any tracking data.
At that point, you don’t know which prompts bring up your competitors, which ones lead to similar answers, or which questions are most relevant to buyers. You also don’t know which important topics might be missing from your list.
Those patterns only become clear after you’ve tracked the prompts for a while.
For example, two prompts may turn out to be the same question in different words, so you only need one. Another may rarely mention companies in your category at all. You may also spot important questions where your competitors are mentioned that your original list missed.
That’s why you should treat your first prompt set as a starting point and refine it as you go.
Once you have some data, review the prompt list again. Keep the prompts that tell you something useful and remove near-duplicates. Then, add any important questions the first list missed.
## Branded vs. unbranded prompts: why branded prompts inflate your AI visibility score
Branded prompts can inflate your AI visibility score because they already include your brand name. For example, “Is Salesforce good for small businesses?” tells ChatGPT which brand to talk about. But “What’s the best CRM for a small business?” leaves the choice open.
We pulled the data from Searchable to show how different the results can be. Over seven days in ChatGPT (September 23 to 29, 2026), brands appeared in 82.9% of 379,602 responses to prompts that **named the brand**. Across 1,889,364 responses to **category prompts** that didn’t name the brand, visibility was 15.3%.
Brand visibility in ChatGPT: branded vs. category prompts
Share of ChatGPT responses that mentioned the brand, September 23 to 29, 2026.
Prompts that name the brand (379,602 responses)82.9%
Category prompts (1,889,364 responses)15.3%
The chart above shows that branded and category prompts measure different things.
A branded prompt shows what happens when someone already knows your company and asks about it. You can use it to track how AI engines describe your product, which claims they repeat, and whether sentiment changes.
A category prompt shows whether your brand appears before the buyer has chosen a company. That makes it the better measure of discovery, because the AI engine chooses which brands to include in the answer.
If you combine both types into one visibility score, branded prompts can make the number look stronger than your real category visibility. Your score may stay high even if your brand rarely appears in broader category questions.
Across more than 1.6 million tracked prompts, only about 16% were branded, so branded prompts don’t dominate most sets. Mixing them into the same score can still make the overall number harder to interpret.
Share of tracked prompts that are branded
15.7% of 1,641,972 tracked prompts name a brand, as of September 30, 2026. For your own set, aim for roughly 10%, and report branded visibility separately from category visibility.
Branded prompts are still worth tracking, but keep them separate.
As a simple rule, keep branded prompts to roughly 10% of your prompt set, and report their visibility separately from category prompts.
## The 4 types of prompts to track
A useful prompt set should cover four types of questions: category, comparison, problem, and branded prompts. Each one tells you something different about how buyers find and evaluate companies like yours.
The four prompt families, with an example of each and what tracking it tells you:
Prompt family
Example prompt
What it tells you
Category
“What are the best CRMs for a small sales team?”
Whether new buyers discover your brand. Give these the most space.
Comparison
“HubSpot vs Salesforce for a small business”
Whether you make the shortlist, and which competitors appear alongside you
Problem
“How do I stop my bike chain from rusting in the winter?”
Whether AI suggests your type of product before the buyer names a category
Branded
“Is Trek a good bike brand for beginners?”
How AI describes your product and whether sentiment changes. Keep to roughly 10%.
### Category prompts: where new buyers discover you
Category prompts ask for products or companies in your market without naming a specific brand. For example, “What are the best CRMs for a small sales team?”
These prompts show whether your brand appears when someone knows what kind of product they need but hasn’t decided which company to choose yet.
Category prompts should make up the largest part of your set. They show whether buyers can discover your brand when they ask about your category.
Focus on the main categories, use cases, and customer types that matter to your business. Skip small wording variations of the same question.
### Comparison prompts: where you make the shortlist
Comparison prompts are questions buyers ask when they’re comparing products, brands, or options. For example, “HubSpot vs Salesforce for a small business.”
These prompts show whether you appear when a buyer is deciding between a few options. They can also show which competitors AI engines mention alongside your brand.
Include the comparisons that reflect the real alternatives buyers consider before making a purchase decision. You don’t need to track every possible brand combination.
### Problem prompts: before the buyer names your category
Problem prompts describe something the buyer wants to solve without naming the type of product they need. For example, “How do I stop my bike chain from rusting in the winter?”
These prompts are important because buyers often start their search with the problem. An AI engine might suggest chain lubricant, a rust-prevention product, or a maintenance routine in the answer.
Include the prompts for the main problems your product can solve. If the answers rarely mention your type of product or cite your page as a solution, the prompt may be too broad to be useful.
### Branded prompts: how AI engines describe your brand
Branded prompts include your company or product name. For example, “Is Trek a good bike brand for beginners?”
These prompts won’t tell you much about discovery because the buyer already knows the brand. They do, however, show how AI engines describe your product, whether the information they give is accurate, and whether sentiment changes over time.
Keep branded prompts to roughly 10% of the set, and report them separately from category prompts. That way, you can still monitor how AI engines talk about your brand without making your broader visibility look stronger than it really is.
## How to find prompts to track
You can build your prompt set from three places: conversations with buyers, existing search data, and prompt research tools.
We recommend starting manually so you understand the language people use. Then use your existing data and tools to fill the gaps and expand the list without adding duplicate prompts.
### 1. Start with the questions buyers already ask
Collect the exact questions customers and prospects ask when they’re researching, comparing, or trying to solve a problem.
Go through sales calls, support tickets, customer interviews, demo notes, and community conversations. Pull out questions that fit the four prompt families.
For example:
* Category: “What’s the best CRM for a small team?”
* Comparison: “HubSpot or Salesforce for a startup?”
* Problem: “How do I increase free trials?”
* Branded: “Does Salesforce integrate with Slack?”
You can also ask ChatGPT or another AI engine questions about your category and look for recurring questions, comparisons, and problems that come up repeatedly. Use those ideas to expand your list, but compare them against what real buyers are already asking. That way, your prompt set reflects the questions buyers actually ask.
### 2. Turn Google Search Console queries into prompts
Your SEO data is another useful source because it shows which questions already bring people to your site.
Open Google Search Console and look for the queries your pages appear for. You can often turn questions and longer searches into natural prompts with minimal tweaks.
For example, you can turn a query like “best CRM for agencies” into “What are the best CRMs for a marketing agency?”
You don’t need to turn every query into a prompt. Focus on the ones that fit one of the four prompt families and cover a topic you want to track.
### 3. Use prompt research to find coverage gaps
When your prompt set gets bigger, it’s easy to miss or over-cover certain topics.
Searchable’s [prompt tracking and prompt research](https://www.searchable.com/features/prompts) suggests prompts by topic and shows how much coverage each topic already has.
You can also look at [the sub-questions engines actually ask](https://www.searchable.com/features/aeo-insights/query-fanout) to find related angles your current set may be missing.
This makes it easier to spot areas where you have plenty of prompts and areas you’ve barely covered.
## How to choose which prompts to track first
When several prompts seem worth tracking, use an estimated prompt volume as a demand signal. In Searchable, the Volume column shows the estimated level of interest behind each prompt, which helps you compare prompts against each other.
For example, “Top beginner bikes for casual riding” shows higher volume than “What to look for in a first mountain bike.” That suggests there’s more overall interest around the first topic.
Keep in mind that volume is an estimate. AI engines don’t share how often people type a specific prompt, so volume can’t tell you exactly how many people asked it.
So, use demand to compare prompts in your own list, then check these four things:
**Does the prompt fit your category?** If the answer is unlikely to mention products like yours, it probably isn’t worth prioritizing.
**Do your competitors appear?** If competitors show up and you don’t, that’s a useful gap to track. If nobody from your category appears, the prompt may not be relevant enough.
**Can you influence the sources behind the answer?** Check which websites the AI engine cites. If your brand could realistically appear on those sites, you have something you can work on.
**Is the prompt close to a buying decision?** A lower-demand prompt can still matter more if someone is using it to compare products before choosing one.
## How many prompts should you track?
The right number depends on how many topics you want to track. Start by listing the main topics you want your brand to appear for, like your product categories, use cases, or customer types.
Then give each topic enough prompts to show a reliable picture. One or two prompts per topic won’t tell you much, because AI answers change from one run to the next.
[A 2026 State of AI Search report](https://www.airops.com/report/the-2026-state-of-ai-search) found that only 30% of brands visible in one answer were still visible in the next, and just 20% stayed visible across five runs.
So if a topic only has one or two prompts, a single changed answer can make your visibility for that topic look much better or worse than it really is.
As a simple rule, track roughly 15 prompts for each important topic. Broader topics with more subtopics may need more.
Within each topic, most prompts should be category prompts. Add comparison and problem prompts for the questions buyers ask about that topic. Then keep branded prompts to roughly 10% of your full set.
For example, a bike brand tracking six topics would need 90 topic prompts and about 10 branded prompts, for 100 prompts in total.
Example: sizing a prompt set for a bike brand
Six priority topics at roughly 15 prompts each, plus branded prompts at about 10% of the set.
Road bikes15
Gravel bikes15
Mountain bikes15
Electric bikes15
Commuter bikes15
Accessories15
Topic prompts (6 topics × 15)90
Branded prompts (about 10% of the set)10
Total prompt set100
If your plan doesn’t allow that many prompts yet, start with your most important topics and cover them properly. You can add more topics later.
Once your set is in place, add new prompts when you launch a product, enter a new category, or see a new competitor showing up in the answers.
## When to remove a prompt from your tracking set
Your prompt set should change over time. Remove prompts that no longer give you useful information, and keep the ones that still help you understand your visibility, competitors, sources, or sentiment.
Review the set regularly and cut prompts when:
* You appear every time and the result has barely changed. If there’s nothing left to learn or improve, you may not need to keep measuring it.
* You never appear for a reason you can’t realistically change. For example, the answer may consistently rely on sources or types of companies where your brand doesn’t fit.
* The topic has very little demand. If there’s little interest in the topic and the prompt isn’t important for another reason, the prompt may not be worth the attention.
* Another prompt asks the same question in slightly different words. Keep the version closest to how buyers phrase it.
Keep or cut this prompt?
Work down the questions. Stop at the first yes.
Are AI answers starting to describe your brand more negatively?Yes: keep
No ↓
Do you appear every time, with results that have barely changed?Yes: cut
No ↓
Do you never appear, for a reason you can’t realistically change?Yes: cut
No ↓
Does the topic have very little demand, with no other reason to keep it?Yes: cut
No ↓
Is it a near-copy of another prompt, asking the same question in different words?Yes: cut one
No ↓
Keep tracking it
## How to build and maintain your prompt set in Searchable
In [Searchable](https://www.searchable.com/), you can find prompts, choose which ones to track, and review the results in one place. Then you can use what you learn to improve the prompt set over time.
Start with prompt research. Searchable suggests prompts by topic, which helps you find questions your current set is missing.
Then narrow down the list. Use the volume indicator to compare interest between topics and check intent to understand what kind of question each prompt represents.
Keep the prompts that fit the topics, competitors, and buyer questions you care about.
Once you start tracking them, check the answers behind the scores:
When [tracking brand mentions in AI answers](https://www.searchable.com/blog/tracking-brand-ai-responses), check which brands appear, [which sources AI engines cite](https://www.searchable.com/features/aeo-insights/sources), and how they describe your brand. This gives you more context than the visibility score alone.
For example, your score might stay the same while a competitor starts appearing more often, or the sentiment around your brand gets worse. That gives you something specific to act on.
Finally, review your prompt set every quarter. Remove near-duplicate prompts and ones that no longer matter. Add new ones when your products, competitors, or buyer questions change.
That way, your set stays focused instead of growing with prompts you no longer need.
The workflow is simple: find prompts, filter them, track them, read the answers, and update the set regularly.
## Frequently asked questions
## Start with the prompt set you already have
Open the prompt set you’re tracking and check whether every prompt helps you answer a specific question about your visibility. If it doesn’t, remove it or replace it with something more useful.
While you can do this manually, it’s usually inefficient and unreliable, especially when your prompt set grows. The same prompt can return different answers across runs and AI engines, which makes it harder to compare results and spot real changes over time.
Searchable helps you handle that in one place, from researching prompts to tracking them consistently and reviewing the answers behind your visibility data.
---
[Back to Blog](https://www.searchable.com/blog) | [Searchable Homepage](https://www.searchable.com)