This workflow dives deep into the prompts you track, uncovers competitive gaps, examines the citations large language models rely on for recommendations, and reveals your brand’s true visibility in Moz Pro.

Marketing professionals are used to being overwhelmed by data. We're used to sifting through data points from disparate systems and building a narrative that proves we are doing a good job for our clients.

However, prompt tracking tests marketers and SEOs in a way that feels closer to the fabled Madison Avenue advertising executives of the 1960s:

• No direct conversion data ties to an individual tracked prompt.
• No public-facing search volume data exists for prompts.
• Prompt responses only offer directional guidance. Responses are customized based on each user’s own habits and past chat history.

The measurement playbook that SEOs have memorized over the last 15 to 20 years has been thrown out the window.

Fortunately, you can analyze directional AI Visibility data and turn it into an action plan. Unlike most digital marketing programs, there are no easy, quick fixes. Brand marketing? You better believe that’s back on the menu. Buying non-branded links to manipulate search results? Yeah, not so much, unless you’re into burn & churn projects.

The goal isn't to manufacture tidy attribution stories from sample data. The goal is to connect the questions your customers ask with the answers they receive, the brands surfaced in those recommendations, and the sources that influence those answers.

That connection will not produce a tidy conversion rate. It will show you where to investigate and what to improve next.

Tracking your brand's visibility without a clear objective is like setting up an analytics dashboard without knowing which conversion actions most impact your business.

Define what you are trying to solve and connect the analysis to a business outcome, enabling you to interpret data in a way that drives strategic decisions.

For example:

• A product marketing team may want to know whether its brand appears as a recommendation when potential buyers ask LLMs to compare the best products for a specific use case.
• A content marketing team may want to understand which content types influence LLM recommendations and which related fan-out queries emerge from a priority topic. Fan-out queries are the related searches or subquestions an AI system may use to explore a wider request.
• A digital PR team may want to identify third-party sources that create credible opportunities for brand coverage, inclusion, or outreach.
• An SEO & GEO team may want to show executives how their brand appears in AI alongside traditional search performance.
• An executive may want a focused dashboard that tracks how the brand is discussed across a defined set of high-value prompts.

Once you know the main purpose for setting up a dashboard, choosing what to track gets much easier.

Define:

• Your target audience or ideal customer profile (ICP).
• The products or services that can credibly satisfy the need.
• Your priority topics.

The target market and language. Where customers are in the purchasing cycle.The potential work your team expects to do after reviewing the data (e.g., reaching out to authors, creating new content, etc.)

Now each team in your marketing organization has something it can use, rather than starting from a blank slate. Content writers can plan around it; product marketing can sharpen positioning; media buyers can create ads around specific features and pain points; PR can identify where stronger third-party coverage may matter; and SEO/GEO can build a reporting view tied to a real objective.

It also prevents a common mistake: treating every missing brand mention as a problem. If a prompt does not fit your audience or product offering, appearing in the answer may have little commercial value.

Start with top-level metrics, but don't stop there.

• Visibility measures coverage. How many tracked responses included your brand?
• Average rank measures position when present. Absences are excluded, so read rank beside visibility.
• Share of voice measures presence within a defined competitive set. It does not represent the entire market or every brand named in the response text.

The AI Visibility dashboard in Moz outlines these data points in a digestible format, highlighting how often brands are mentioned, how early the brands are mentioned in responses, and how often each brand “wins” the race to be shown first. 

These metrics answer different questions. 

• A strong average rank with low visibility means the brand performs well when selected, but doesn’t appear often. 
• High visibility with a weaker rank means the brand appears regularly but may not hold a leading position. 

This is why I’d never put one of these numbers on a slide by itself and call it the story.

Example: Coverage and rank can point in different directions.

In this post, I will reference a prompt-tracking dashboard I set up to compare my favorite headphone brand against its fiercest competitors. As a heavy metal enthusiast, I have tested a wide range of headphones and earbuds over the last decade, and Bose has always come out on top. P.S. This is not sponsored in any way, but I cannot express my gratitude enough for Bose creating headphones that turn the cacophonous sound of heavy metal into sweet, bass-induced sonic delight. Ok, enough of that; let’s get back to the data & the example:

In a supplied set of 72 headphone prompts, Bose appeared in 61.1% of ChatGPT responses and 44.4% of Perplexity responses. However, when present in prompts, the average rank was 1.95 in ChatGPT and 1.84 in Perplexity.

Bose appeared less often in Perplexity, but tended to hold a slightly stronger position when included in the prompt response. That is a reason to investigate the missing appearances, not declare one model universally stronger or weaker.

The overall visibility score is only the starting point. Export the data, then break it apart until you can see where the pattern is actually coming from.

Start by segmenting prompts into a few useful buckets:

• Topic: Where is the brand strongest or weakest across product categories, use cases, features, pricing, and support?
• Intent: Does the brand appear in informational, commercial, transactional, or navigational prompts?
• Prompt format: Are gaps concentrated in recommendations, comparisons, how-to questions, or feature constraints?
• Model: Does the same question produce a different brand set or recommendation pattern across ChatGPT, Gemini, Google AI Mode, and Perplexity?
• Competitor presence: Which tracked competitors appear when your brand does not?

On Intent:

You can then internally label your prompts based on traditional search intent.

Pro tip! Keep prompt counts beside percentages. If a topic contains six prompts, one response can materially change the result, so you do not want outliers to paint a story that might lead you down a rabbit hole.

Don't segment everything just because you can. Look for three findings:

1. A recurring strength worth defending.
2. A model-specific difference worth investigating.
3. A repeated gap across important prompts or models.

Your goal at this stage is to convert a giant number of exports into a handful of patterns worth investigating. Segmentation should reveal what the overall score hides, not give you six more charts to explain. Consider which patterns affect key customer segments, have the greatest potential business impact, and which you can realistically address with available resources. This approach helps your team focus on the patterns that matter most instead of getting lost in interesting but less impactful findings.

If your current prompt set is too thin to reveal any themes or actionable data, Moz’s Prompt Suggestions can surface additional prompts based on topic sets. Review the baseline first, then add relevant suggestions and label them separately.

Visibility tells you whether your brand appeared. Sentiment helps you understand how favorably it was discussed.

Let’s take a look at how you can analyze brand sentiment in the Moz tools.

Within AI Visibility, open the Sentiment tab for the model and prompt list you want to analyze. Start with the Overall Score and its confidence label, but use it as a starting point, not the verdict. It reflects discussion within the tracked prompts, not universal brand perception.

Next, review the Mentions Breakdown. The total number of mentions and the positive, neutral, and negative mix give essential context when comparing brands. Two brands can land on similar scores with very different mention counts, so don't treat those scores as equivalent without checking the volume behind them.

Then move from the aggregate score to Top Themes and Sentiment Drivers:

• Use Top Themes to identify perceived strengths and weaknesses.
• Select your brand or a competitor under Sentiment Drivers, then expand individual themes to inspect the summaries and relevant response snippets.
• Investigate mixed or weak themes before proposing a fix. Determine which prompts, claims, and sources contribute to the pattern.

Example: The overall score is only the first layer

On the ChatGPT dashboard, Bose had an Overall Score of 73 with high confidence across 169 mentions: 79 positive, 90 neutral, and none negative. Noise cancellation scored 88, and listening comfort scored 95 among its top themes.

That doesn't mean the internet universally loves Bose. It means those strengths showed up consistently within this tracked prompt set. The 90 neutral mentions are also a reason to keep digging rather than stop at the headline score.

Not every mention is grouped into a theme, so the theme counts may not perfectly match the headline metrics. Use the in-app view to spot the pattern, then read the response text to understand what's driving it.

Once you find a pattern, move from the metric to the actual response.

Compare all of the following:

• The products or services recommended.
• The criteria used to make the recommendation.
• The competitors included.
• The claims made about each option.
• The cited sources.
• Whether your brand credibly fits the need.

Example: Similar metrics can hide different problems

For “best over-ear headphones for glasses wearers,” Bose held an average rank of 1 in three models and was absent from Google AI Mode. For “best headphones for video calls,” Bose appeared only in Perplexity, at exported rank 2.

The first gap points you toward comfort criteria and the sources supporting them. The second points you toward microphone and call-quality evidence. Neither proves you need to dedicate resources to build a new page.

Ignore the basic instinct to create a 2,000-word blog post to address not showing up for a single prompt. A missing appearance may point to out-of-date information, weak third-party coverage, unclear product positioning, a genuine product limitation, or poor prompt fit.

A missing appearance isn't the diagnosis. It's the clue that tells you where to investigate next: the prompt, the response, the product fit, and the sources directing the recommendation.

Citations show the sources LLMs reference when answering specific prompts. Marketing teams can review the list of cited pages in a given dashboard to identify possible opportunities for PR campaigns and SEO efforts. I use a simple checklist when analyzing cited pages:

First, classify each relevant source: Is it an editorial review, comparison page, retailer, community, video, social media post, research report, owned resource, or competitor-owned page?

Then analyze how that citation is used and how it relates to your brand:

1. Does your brand appear on the cited source page?

2. Does your brand appear in the LLM response that cites it?

Those combinations lead to different actions:

Example: An aggregate citation field is not the final answer

Cited pages from Perplexity showed a TechRadar over-ear headphone guide as a source across 20 prompts, with Bose mentioned in half of them. The live page already contained a substantive Bose section, which likely influenced Perplexity’s responses.

The wrong move would be sending a generic pitch to TechRadar to add more information about Bose. The defensible move is to identify the associated prompts and responses, confirm which attributes are highlighted, and find a specific, evidence-supported improvement you can demonstrate with new content.

Important clarification: a citation isn't the same as a backlink, and it certainly isn’t an outreach permission slip. A page can mention your brand without linking to you. That can create an outreach opportunity, but only when a link would genuinely improve the page. Big publications and popular authors are drowning in templated pitches; “you mentioned us, please link” is not a sustainable strategy.

Studying the data is only half the job. Most reports end up in purgatory because they never tell the team what to do next. The typical report barely explains what happened, adds multiple fancy-looking charts, and quietly expires in a shared folder after a 30-second executive review. Your AI Visibility report should finish with a short list of work the team can actually start.

Turn each verified finding into one of four types of work:

• Fix the facts: Correct any obsolete information on both owned pages and the third-party sources from the citations report.
• Earn consideration: Craft pitches for your brand to be mentioned on cited pages where competitors are featured, and you’re not.
• Protect a strength: Keep the supporting claims and pages accurate and current.
• Keep digging: Investigate when product fit, prompt value, or source evidence remains murky.

Not every finding deserves its own project. Some patterns can turn into work that actually moves revenue; other tasks may be intriguing but not worth pursuing. To prioritize effectively, use this checklist:

• Assess business impact
• Consider the frequency with which the pattern appears.
• Evaluate your confidence in the supporting evidence.
• Estimate the effort required.

Pick three to five priorities based on these criteria. Don’t mark every potential task as a high priority unless you want to stress out your colleagues and burn the midnight oil with your project management system.

Assign each project an owner, a concrete deliverable, a supporting prompt or URL, a review date, and a status. For most teams, a shared spreadsheet is enough; larger teams can move the work into their existing project management system.

One last rule: don't keep moving the goalposts. Preserve the existing prompts, exports, topics, cited pages, and collection settings. Compare the same scope next time and report any new prompts separately. Otherwise, you can manufacture a dramatic increase in brand visibility simply by changing what you track.

For example, let’s say you want to pivot your brand visibility dashboard to track only comparison prompts against competitors, using only branded terms. Your next collection may reflect a notable increase in brand visibility, but that’s because you guided the system to mention your brand name in every single prompt. Based on your original business objective, you may want to track non-branded prompts to gauge better whether LLMs recommend your products without being prompted by the brand name. On the other hand, a large number of branded prompts can give you more evidence to collect intel on how your brand is perceived across the web.

Steps 1 through 7 show you how to analyze your AI visibility within Moz Pro and the corresponding exports. If you’re managing multiple dashboards, the true problem is scale. Once you're reviewing hundreds of prompts across multiple models, topics, responses, and cited pages, the manual work gets ugly fast.

That is why I built the AI Visibility Analysis skill. It applies the same workflow logic in bulk: comparing models and topics, spotting recurring strengths and gaps, summarizing response themes, building a citation review queue, and turning the strongest verified findings into prioritized actions. 

It then packages the analysis into:

• An executive overview that gives leaders the scorecard, the most important gaps, and a short list of priorities.
• A detailed report that explains the findings across models, topics, competitors, prompts, response themes, and citations.
• An editable action workbook that maps prompts and sources to actions, owners, effort, confidence, status, and next steps while keeping the supporting evidence attached.
• Supporting CSV and JSON files that preserve the prompt appendix, cited-page inventory, citation-review queue, and calculated metrics for deeper analysis.

Example: The analysis matters more than the file count

In the Bose example, the skill processed the same prompt and citation exports used throughout this post, segmented performance across models and topics, surfaced repeated gaps, built a citation review queue, and translated the strongest findings into prioritized actions. It then packaged that work into a three-page executive summary, an eight-page detailed report, a 13-sheet workbook, and supporting evidence files.

The skill is platform-agnostic and can run in tools such as Claude Code, Codex, or Cursor/Grok. Because the skill uses a reasoning-heavy workflow, stronger models generally produce deeper analysis. While new models keep launching, it’s best to stick with high or extra high reasoning, regardless of the model, to ensure the best-quality output.

Note: You can download the AI Visibility Analysis skill as part of a wider Marketing Skills public repository I have created here. In addition to the AI Visibility Analysis skill, you can download the Backlink Gap Analysis & SEO Workbook Builder, both of which use Moz data to help you in everyday SEO work.

Start with the scorecard or use the skill to scale the analysis, but don't run it once and call it a day. Immerse yourself in the data, unearth any potential patterns, and try to find 3-5 major opportunities for your brand over the next 90 days.

The ultimate goal with this data is to answer the three questions:

• Where is your brand consistently winning (Defend your position)
• Where are competitors being recommended instead of you (Build your visibility)
• What evidence appears to be influencing that decision (Understand the market)

That is where AI Visibility data stops being another score and starts becoming useful.

You may discover a positioning problem, a gap in third-party coverage, outdated information, weak product evidence, or simply prompts where your brand was never a good fit in the first place. The important part is knowing the difference.

AI Visibility will not give you the clean attribution marketers have spent the last 20 years chasing. However, these reports can show you something traditional ranking reports never could: how AI systems understand your brand, who they compare you against, and which sources are shaping those recommendations.

The brands that get the most value from this data won’t be the ones obsessing over whether their visibility score moved three points this month. They’ll be the ones using the data to figure out why they are or are not part of the answer, and what they can do about it.