Being quoted in an AI answer feels like winning. It isn’t. The model’s frame about your category was already decided before your name showed up.
The Mention Trap: Why You Can Be Quoted and Still Lose
Picture two project management tools. Brand A gets mentioned in 60% of ChatGPT answers about “best project management software.” Brand B gets mentioned in 30%. Brand A should be winning, right?
Not necessarily. If you actually read those answers closely, Brand A shows up in a list, once, with a generic line like “also worth considering for smaller teams.” Brand B gets the first sentence, the recommendation, and three follow-up questions steered toward it. Brand A has more mentions. Brand B has the narrative.
This is the trap almost every AI-visibility tool walks you into. They count appearances. They track share of voice like it’s 2015 SEO. But an AI answer isn’t a search results page where position and click-through are the whole game. It’s a generated recommendation, built on a frame the model already holds about your category before it ever writes a sentence about you. Mentions are what you see. Frame is what decided what you’d see.
How AI Models Build Their Category Frame (It Happens Before Your Mention)
When someone asks ChatGPT “what’s the best CRM for a 20-person startup,” the model isn’t scanning the web in that moment and picking a winner. It’s pulling from a learned representation of your category, shaped by training data, reinforced or adjusted by retrieval and live citations, and expressed through whatever framing has the strongest pull in that representation.
That frame gets built the same way any consensus gets built: repetition, authority, and consistency across the sources the model actually weighted heavily. If Reddit threads, G2 comparison pages, and three widely-cited review sites all describe your competitor as “the enterprise-grade option” and describe you as “budget-friendly but limited,” that’s the story the model learned. It’ll keep telling that story regardless of how many times your name shows up elsewhere.
This is why asking “why does ChatGPT recommend my competitor instead of me?” is the wrong first question. The better question is: whose description of the category did the model absorb as true?
Source Authority vs. Mention Volume: Where Consensus Actually Lives
Mention volume is easy to fake and easy to game. Press releases, sponsored roundups, a flood of low-authority blog posts, all of that can bump your name count without moving an inch of actual consensus.
Source authority is different. It’s about which handful of documents the model treats as ground truth for your category. A single well-cited Gartner comparison, a heavily-linked Reddit thread with hundreds of upvotes, or a Wirecutter-style deep review can outweigh fifty mentions from sites the model doesn’t trust.
Citation density matters too, not just whether a source mentions you, but how often that specific source gets cited by other sources the model also trusts. That’s how semantic dominance compounds. It’s not one article. It’s a network of sources all repeating the same frame until it becomes the default answer.
Mapping Your Narrative Gap: Where the Model Sees You vs. Where You Want to Be
Try this: ask ChatGPT, Perplexity, and Google AI Overviews the same three or four questions a buyer in your category would actually ask. Not “tell me about [your brand].” Ask “what’s the best tool for X,” “how does [category] pricing usually work,” “what should I watch out for when choosing a [category] vendor.”
Now write down, in plain language, the story each answer tells about you. Not whether you’re named. What role you’re cast in. Are you the safe default? The niche player? The one with the asterisk? Compare that to how you actually want to be described, the language your sales team uses, the positioning you’ve built your whole GTM around.
The distance between those two descriptions is your narrative gap. It’s rarely about visibility. It’s almost always about which sources the model is treating as authoritative on your category, and whether your version of the story lives in any of them.
The Three Levers to Shift Consensus (Without Chasing Mentions)
1. Identify the Sources the Model Trusts (Citation Archaeology)
Before you can change the frame, you have to find out what built it. That means tracing the specific sources showing up in AI answers about your category, repeatedly, across different prompts and different models. Some will surprise you: a niche subreddit, an old comparison post, a review aggregator you’ve never optimized for. Those are the documents actually shaping the recommendation.
2. Compete for Semantic Dominance in the Frame (Not Just Keywords)
Once you know which sources matter, the work isn’t “get a backlink from them.” It’s making sure the language used to describe your category on those sources reflects the frame you want. That might mean engaging directly where those conversations happen, briefing analysts on the comparison points you actually win, or making sure your own explanation of the category exists somewhere those trusted sources will cite it.
3. Place Authority Signals Where the Model Looks (Not Where You Publish)
Your blog post about why you’re the best choice does very little if the model isn’t weighting your domain heavily for this category. The signals need to land in the third-party, high-trust sources that already have pull, comparison sites, community threads, review aggregators, industry roundups, not just your own properties.
A Real Example: SaaS Category Comparison Pages
Take any competitive SaaS category where a handful of comparison sites dominate the “X vs Y” search intent. If those pages consistently frame one vendor as “built for scale” and another as “good for solo founders,” that’s the language models will echo almost verbatim, because it’s the most repeated, most cross-cited framing available. A vendor could publish a hundred blog posts arguing they scale just fine and barely move the needle, because those blog posts aren’t the sources the model treats as authoritative for that comparison. The fix isn’t more content. It’s getting the frame corrected at the source the model actually trusts.
Why Dashboards Miss This (And What to Do Instead)
Most AI-visibility tools will tell you your mention count went up 12% this month. That’s a symptom, not a cause. It doesn’t tell you whose frame the model is using, why it picked that frame, or which sources you’d need to influence to change it.
This is the layer Mavel was built to read. We measure narrative share, whose story the model actually tells about your category, and we trace it back to the sources and citations building that story. Not another dashboard number to interpret. A prioritized answer to what to ship if you want the frame to change.
If you’re staring at a mention count that keeps climbing while the recommendation still goes to someone else, come talk to us about what your category’s narrative gap actually looks like.
