From keywords to prompts: How to right-size your AI search tracking

takes
Learn how to calculate how many prompts you should track for your AI search strategy using keywords as your guide.

“How many prompts should I track for AI search?”

The honest answer is: It depends.

On your company footprint. Your business initiatives. Your products and services and competitive landscape.

But a nuanced answer isn’t always the most helpful one. Especially when you’re trying to stand up a new AEO/GEO program.

Which is why we’ve come up with a practical approach brands may find useful: X [# of topic clusters] x Y [12-15 questions related to each topic cluster] = Z [# of AI search prompts to track].

This is less of a prescriptive formula and more of a ballpark figure. But we’ve found it to be a helpful heuristic for brands that are unsure of what good looks like when it comes to prompt tracking.

Here’s the reasoning behind the recommendation—and how to get the most out of your prompts.

Keywords aren’t prompts (but they are helpful guides)

Your brand is most likely already tracking keywords across Google Search Console and Google Ads.

These are a natural starting point for AI search tracking.

At the same time, it’s important to remember that keywords aren’t a one-to-one match for prompts:

  • Prompts are longer and more detailed: The average keyword is somewhere between 2-5 words. The average prompt is approximately 5x that length (and can be hundreds of words long depending on how detailed the user gets—or if they’re talking instead of typing).
  • Prompts have more intent specificity: Users generally make it clear what their intent is with a prompt (e.g., “What should I consider when buying an X?”, What’s the best Y?”, “Compare the pros and cons of Z”, etc.)
  • Prompts provide more context: Users tend to get granular with their prompts to weed out overly broad results, which provides context clues for audience segments, use cases, decision-making criteria, and more.

Say your brand is optimizing for the keyword “project management software.”

In AI search, that short, fragmented phrase could go down a lot of different roads:

  • What project management software features are most important for teams that use Google Workspace?
  • Which project management software is best for medium-sized SaaS companies with remote employees in different countries?
  • Compare ClickUp to Asana—which project management software is better for companies with over 1,000 employees in regulated industries and also works for both marketing and engineering teams?
  • You get the idea.

AI platforms like ChatGPT, Gemini, and Perplexity will break prompts up into sub-queries (aka a query fan-out), scour the web, synthesize information from multiple sources (including, hopefully, your website), and pull the bits and pieces back together to return an answer to the user.

The natural next question: Does that mean I need to track every possible prompt variation?

Thankfully, no. The goal is not to monitor every single version of a prompt you can think of—it’s to get a representative sampling you can use to track your performance for business-relevant topics over time.

As Scrunch data scientist Michael Iannelli put it in his post on measuring prompt performance:

“Just like meteorologists don’t need to measure the temperature of every square foot of a city to report the weather, marketers don’t need to track each and every prompt to measure brand perception and competitor movement.”

So the question then becomes: How many prompts do I need to track to get a representative sampling?

Keywords are your starting point. Next comes transforming them into topic clusters that give you the coverage you need.

Right-size for a representative sampling

Here’s the workflow we recommend:

Step 1: Identify your topic clusters

Put together a list of SEO keywords, paid search keywords, and the major product and content categories of your website (check out our AI search guide for tips on sourcing prompts).

Next, sort that list based on common themes (e.g, product use cases, competitors and alternatives, etc.) and group related keywords together.

These are your topic clusters (which will become your prompt clusters).

For instance, maybe you have a cluster all about a specific competitor. It would likely include keywords like:

  • [Competitor] alternatives
  • [Competitor] vs. [your brand]
  • Enterprise version of [Competitor]
  • And so on

Each cluster will usually have somewhere between 5-15 keywords associated with it. If the number of keywords goes beyond that, it may be a sign that you need to break a single topic into multiple clusters.

Step 2: Formulate questions for topic clusters

Next map out 12-15 questions for each topic cluster that made the cut (you can use our free Prompt Generator tool to get an idea of the types of prompts you should be tracking).

Make sure your questions cover the full customer journey, from advice and awareness to evaluation and comparison. You’ll also want to include both branded and non-branded variations of questions to cover all of your bases.

If you’re a Scrunch customer, our AI can generate prompts for you.

We scan your brand context and produce prompt options in the same topic areas you care about. This can be very helpful for mid-funnel questions your team may not think to write, persona-specific angles, topic variants that capture long-tail phrasing, and competitive and comparison questions.

Not a Scrunch customer? Don’t be afraid to ask your AI of choice for help.

It’s a good idea to inform your question set based on the questions that regularly appear in SEO research, PPC queries, and customer outreach to your sales, customer success, and customer support teams. If it’s a question your customers actually ask, you know it matters.

Step 3: Estimate your total number of prompts

Finally, multiply the number of topic clusters by the number of questions. The result is approximately how many prompts you may want to track.

Remember: This isn’t a magic number—every company will have its own unique needs. Some topic clusters may call for more questions. Some may require less.

Either way, it’s a solid starting point to build on over time.

Here’s an application based on estimated keyword tracking volume for different company sizes:

Company sizeTraitsTopic clustersQuestionsEstimated prompts
Startup/SMBNarrow niche, limited content, lean or solo marketing team~3-10012-15x~36-1,500
Growth/Mid-marketMultiple product lines or audience segments, significant amount of content, SEO resources~33-40012-15x~396-6,000
EnterpriseBroad product portfolios in multiple regions or languages, complex website architecture with expansive content, SEO team~133-4,000+12-15x~1,596-60,000+

Make your prompts work harder

Here are some tips to help you get the most out of your prompt tracking:

Organize your prompts

The larger the number of prompts you’re tracking, the more important organization becomes.

That’s why Scrunch offers multiple ways to group and analyze prompts, including:

  • Key Topics: auto-categorization based on the major themes your brand wants to monitor (think product, category, pricing, etc.)
  • Tags: User-defined labels for things like campaigns, product lines, geographies, seasonal prompts, etc.
  • Filters: Dimensions for things like prompt topic, funnel stage, persona, branded versus non-branded prompts, AI platform, etc.

Not using Scrunch? You may need to break out the spreadsheets. Our prompt tracking framework can help you make sense of what you’re tracking.

Keep prompts AI-friendly

Remember: AI models interpret intent based on phrasing.

For best results:

  • Use simple, plain language.
  • Avoid marketing speak.
  • Include the variations real users ask.
  • Ask questions, not statements.

It’s less “online banking benefits” and more “What are the best online banks for small businesses?”

Also be sure to add geographic context when you need city or regional specificity.

Maintain clean prompt data

The entire point of tracking prompts is collecting data that you can rely on.

To do that, we recommend:

  • Keeping prompts neutral and repeatable (i.e., avoid leading words or assumption-based language)
  • Keeping prompts consistent (i.e., don’t update prompt phrasing every week or else you won’t be able to accurately compare how performance improves or declines over time)

Cover your branded and non-branded bases

Branded questions tell you how accurately AI describes you when someone asks about you directly.

Non-branded questions tell you where AI places you relative to competitors when someone asks about a business-relevant product, category, etc.

A good prompt tracking strategy will account for both.

That said, we recommend that you prioritize non-branded prompts to get a better sense of how AI platforms recommend (or don’t recommend) your brand when users don’t specifically ask for it.

It pays to routinely sanity check your prompt tracking strategy based on what customers in your industry are actually asking about.

This is where tools like Scrunch’s Trends feature come in. You can reveal which topics are gaining traction across your industry to help you make sure that what you’re monitoring reflects what people are really interested in.

It can also help you uncover new categories, new competitors, and seasonal demand shifts you might otherwise miss.

Make sure your prompt strategy is driving real results

Scrunch’s AI Referrals and Agent Traffic features are helpful here.

Once you know which of your webpages are seeing human visits from AI platforms and which ones AI crawlers retrieve content from most often, you’ll know if you’re tracking the right questions.

You can compare your top visited or most crawled URLs to your prompt list to identify any gaps (as well as use model-specific referral patterns to decide which AI platforms should be your top priority for monitoring).

Edit and update prompts over time

Customer behavior isn’t static. The prompts you’re tracking shouldn’t be either.

Our advice:

  • Monthly: Review AI search trends for new topics and add relevant prompts
  • Quarterly: Add new prompts based on product launches, campaigns, or competitor moves
  • Ongoing: Archive prompts that no longer matter or replace them with updated variants

Prompt tracking isn’t set-and-forget

The "formula" is simple, but the work is iterative. Start simple and refine over time.

The shift from keyword tracking to prompt tracking is still early. Many brands are figuring it out on the fly.

That presents a real opportunity for the ones that buckle down and take a thoughtful approach.

Your objective isn't perfection—it's a representative picture of how AI sees your brand.

And that picture gets clearer every time you look.

See how AI sees your brand with Scrunch

Build your prompt tracking strategy with help from Scrunch. Try Scrunch free for 7 days or schedule a demo to see it in action first.