Showing up in an AI answer isn’t the same as winning it: a real audit checks whose story the model is actually telling, not just whether your name appears.
The Visibility Trap: Why Mention Counts Miss the Real Story
Ask ChatGPT “what’s the best project management tool for a remote team” and you’ll probably get a list. Your brand might be on it. So might three competitors. Most teams stop right there, screenshot the answer, and call it a win.
That’s the trap. Getting mentioned tells you almost nothing about why the model picked that order, which brand it framed as the default, or which one it quietly positioned as the safe choice versus the niche alternative. You can appear in the answer and still lose the recommendation.
Think about it like a press roundup. Five companies get quoted in an article about the state of an industry, but only one gets described as “the company defining the category.” The other four are mentioned. Only one is winning the narrative. AI answers work the same way, except the “article” gets rewritten every time someone asks a question, based on whatever the model learned about your category from the sources it trained on and retrieves from.
A visibility audit that only counts mentions is measuring attendance, not influence.
The Three Layers of an AI Visibility Audit
Most AI visibility checks stop at layer one. A real audit goes three layers deep.
Layer 1: Presence Audit (What existing tools show you)
This is the baseline work, and it’s not useless, it’s just incomplete. You’re checking: does your brand show up in ChatGPT, Perplexity, Gemini, Copilot, and Google AI Overviews for the prompts that matter to your category? How often? In what position? Tools like Profound and Peec are built for exactly this. They’ll tell you your mention rate went up 12% this quarter.
What they won’t tell you is whether that 12% increase actually changed anyone’s mind.
Layer 2: Narrative Audit (Whose frame the answer adopts)
This is where the real work starts. For every prompt where your brand appears, read the actual language the model uses. Is it describing you in your own terms, the way you’d want a customer to describe you? Or is it using a competitor’s framing, with you as the footnote?
Picture two project management tools, both mentioned in the same AI answer. One gets described as “ideal for teams that need lightweight, flexible workflows.” The other gets “the enterprise standard for cross-functional project tracking.” Both brands are present. Only one owns a frame that sounds like a decision. The other sounds like an also-ran.
The narrative audit asks: whose story is this answer actually built on? Not who’s in it. Whose frame decided the outcome.
Layer 3: Source Audit (Which citations drive the recommendation)
Once you know whose narrative is winning, you need to know why. AI answers aren’t invented from nothing. They’re built from the sources the model has been trained on or retrieves from: review sites, comparison articles, Reddit threads, analyst reports, your own site.
Trace the citations behind the answer. If a competitor’s frame keeps winning, there’s usually a source pattern behind it: a G2 category page that ranks them first, a “best tools for X” roundup that uses their language verbatim, a Wikipedia entry that undersells your positioning. This is the layer that turns an audit from a diagnosis into a to-do list.
How to Map Your Category’s Prompt Universe
Buyers don’t search in keywords anymore, they ask questions. “What’s the best CRM for a 10-person sales team” is a different prompt than “what’s the best CRM,” and it’ll surface a different answer.
Start by listing every real question a buyer might ask at each stage: discovery (“what tools exist for X”), comparison (“X vs Y”), and validation (“is X worth it”). Pull from actual sales call transcripts, support tickets, and Reddit or G2 questions in your category. Run each one across ChatGPT, Perplexity, Gemini, and Copilot. You’ll quickly see that your narrative might be strong in discovery prompts and completely absent in comparison prompts, which is exactly where deals get won or lost.
Running the Audit Yourself: A 5-Step Framework
- Build your prompt list. Pull 30-50 real prompts across discovery, comparison, and validation stages.
- Run them across engines. Check ChatGPT, Perplexity, Gemini, Copilot, and AI Overviews. Log presence, position, and exact wording.
- Read the frame, not just the name. For each mention, write down the one-sentence story the model tells about your brand versus competitors.
- Trace the sources. Where possible, check what the model cites or seems to draw from. Look for patterns across multiple prompts.
- Score the gap. Compare where you’re present to where you’re actually winning the frame. That gap is your action list.
What a Strong Audit Reveals (and Why Most Tools Miss It)
A strong audit usually surfaces something uncomfortable: a brand with great mention rates but a weak frame, or worse, an AI-invented version of the company that doesn’t match reality at all. Maybe the model consistently describes you as “budget-friendly” when your actual positioning is premium. Maybe it recommends a competitor first in 8 out of 10 comparison prompts, even though your mention rate looks fine on a dashboard.
Presence-tracking tools miss this because they’re built to count, not to interpret. Counting is countable. Reading whose story wins requires judgment, not just crawling.
From Audit to Action: Moving from Visibility to Narrative Share
The output of a real audit isn’t a score. It’s a short list of what to ship: which sources to fix, which comparison pages need your language in them, which prompts need a stronger frame behind your brand. That’s the difference between tracking visibility and building narrative share, whose story the model actually tells when someone asks the question that matters.
If you’re serious about knowing whose frame is winning in your category, not just whether you showed up, that’s the audit Mavel is built to run.
