Two brands can show up in the same AI answer and still lose completely different battles: one gets a footnote, the other gets the frame the model builds its whole recommendation around.
The Mention Trap: Why Citation Count Misses the Real Battle
Ask ChatGPT to recommend a project management tool, and there’s a decent chance both Asana and Monday.com show up somewhere in the answer. Same category, same prompt, same visibility on paper. But look closer at how the model talks about each one. One gets described as “built for teams that need structure and accountability.” The other gets a passing mention buried in a list of alternatives.
That’s not a tie. That’s one brand’s frame winning and the other showing up as noise.
Most AI visibility tools would score this as a wash. Both brands got mentioned, so both get credit. This is the mention trap: treating presence in an answer as if it’s the same thing as winning the answer. It’s not. Being named is table stakes. What actually matters is whose story the model is telling when it names you.
What Is Narrative Share of Voice?
Narrative share is whose frame the AI model adopts when it explains a category, not just how often a brand’s name gets typed out. It’s the difference between showing up and shaping the answer.
Think of it like this: every AI answer about your category is built on a story. Someone (or something, or some set of sources) decided that Brand A is “the enterprise-grade choice” and Brand B is “the budget option for small teams.” That story didn’t come from nowhere. It came from patterns in the sources the model was trained on and retrieves from: reviews, comparison articles, Reddit threads, G2 pages, PR, docs.
Narrative share measures which brand’s version of that story the model actually adopts. You can be mentioned in 90% of answers and still have close to zero narrative share if the model keeps framing you as an afterthought, a budget pick, or worse, describing a version of your product that doesn’t match reality anymore.
Mention count tells you if you’re in the room. Narrative share tells you if anyone’s listening to you or to the brand standing next to you.
Narrative Share vs. Search Volume Share: A Concrete Example
Picture two CRM brands, both mentioned in 80% of “best CRM for small business” answers across ChatGPT, Gemini, and Perplexity. Mention share: dead even.
Now look at the actual language. Brand A gets described as “simple, affordable, good for solopreneurs just getting started.” Brand B gets described as “the CRM that scales with you as your sales team grows, with automation that adapts to more complex pipelines.”
Same prompt. Same mention frequency. Completely different narrative. Brand B’s frame implies staying power and room to grow. Brand A’s frame implies “outgrow me eventually.” If you’re a funded SaaS company trying to land mid-market deals, Brand A just lost the recommendation battle while still showing up in the citation count.
This is why narrative share and mention share are orthogonal. They don’t move together, and treating them as the same metric hides the thing that actually decides whether AI recommends you or recommends the other guy.
How to Measure Narrative Share (The 5-Layer Framework)
You can’t measure narrative share with a single “share of voice” number. It takes layers:
- Narrative Share itself. Across the prompts that matter for your category, whose frame does the model reach for by default?
- Perception Gap. Where does the model’s version of your brand diverge from how you actually want to be seen? This is where you find outdated positioning, dead features described as current, or a persona the model invented that doesn’t match your product.
- Source Intelligence. What is the model actually reading to build that frame? Which reviews, articles, or forum threads are doing the heavy lifting?
- Explain-why. Why did the model pick this frame over another? What’s the causal chain from source to sentence?
- Narrative Graph. How does your frame connect to competitors’, and how is it drifting over time as new content enters the training and retrieval mix?
Run those five layers and you get a picture of the actual battlefield, not a mention scoreboard.
Why AI Models Adopt One Narrative Over Another
AI doesn’t rank brands the way a search engine ranks pages. It recommends from a learned story about your category, built from whichever sources spoke loudest, clearest, and most consistently. If ten comparison sites all describe your competitor as “the more reliable option for agencies,” the model absorbs that as consensus, whether or not it’s still true.
This means the model’s answer is downstream of a narrative fight that already happened somewhere else, usually in places brands aren’t watching: G2 review threads, niche newsletters, Reddit arguments from two years ago.
The Cost of Measuring Mentions Instead of Narrative
If you’re only tracking mention counts, you’ll celebrate visibility wins that don’t move revenue and miss the actual reason a competitor keeps getting the confident recommendation while you get the hedge. You’ll optimize for citations instead of fixing the sources actually shaping the frame. That’s expensive in a market where the AI answer is increasingly the first, and sometimes only, touchpoint a buyer gets.
Building Your Narrative Intelligence Practice
Narrative share isn’t a metric you check once. It drifts as sources change, competitors publish, and models retrain. Treating it as a practice, not a report, is the only way to stay ahead of the frame instead of reacting to it.
This narrative layer is exactly what Mavel was built for. Start with the free GEO report and see whose frame the models are actually running with in your category.
