Showing up in an AI answer and getting recommended by it are two different outcomes, and most brands are optimizing for the wrong one.
Mentions vs. Narrative: Why You’re Invisible in Plain Sight
Try this. Ask ChatGPT “what’s the best project management tool for a 20-person startup?” If you run a project management company, you might show up in the list. Good. Now ask a follow-up: “which one would you actually pick and why?” Watch what happens. The model picks one, gives you two or three sentences of reasoning, and that reasoning almost never mentions the five other tools it just listed.
That’s the gap. Being on the list is a mention. Getting picked, with a reason attached, is a recommendation. Companies chase the first and wonder why revenue doesn’t follow.
Most AI-visibility tools count the first thing. They tell you how often your name shows up across a set of prompts, track that number over time, and call it visibility. But a brand can appear in 80% of category answers and still lose almost every recommendation, because appearing and being the answer are governed by completely different mechanics. One is retrieval. The other is narrative.
What Consensus Actually Is (And Why Models Optimize for It)
Language models don’t reason about your product from scratch every time someone asks a question. They reproduce a pattern they’ve absorbed from millions of documents: reviews, comparison posts, Reddit threads, analyst write-ups, docs, forums. When enough of those sources agree on a framing, “X is best for enterprises, Y is best for solo founders, Z is the budget option,” that framing becomes the default answer. That’s consensus.
Consensus is stable, cheap to reproduce, and low-risk for the model. Making up a fresh, balanced comparison every time is expensive and inconsistent. Repeating the frame the internet already agrees on is fast and defensible. So models optimize for consensus the same way a student optimizes for the answer that shows up in every textbook: not because it’s necessarily right, but because it’s the safest bet with the least effort.
This is why two brands with similar feature sets get wildly different treatment. It’s not that the model did a deeper analysis of one over the other. It’s that one of them already has a settled story in the source material, and the other doesn’t, or worse, has a story that doesn’t match how the company sees itself.
How AI Learns Consensus: The Sources Behind the Frame
The frame comes from somewhere specific. It’s built from a mix of comparison articles, community threads, G2 and Capterra copy, review aggregators, product docs, and enough repetition across those sources that the pattern becomes learnable.
Here’s a concrete version. Say you run a mid-market HR platform. There’s a five-year-old Reddit thread where someone recommended a competitor for “startups that don’t want to deal with a sales team.” That thread got quoted in a Zapier comparison post. That comparison post got cited in three other “best of” roundups. Now that single framing, “good for startups, avoid the sales process,” is baked into how models describe your competitor, whether or not it’s still true.
Meanwhile your own positioning, the one your sales team pitches, might live almost entirely on your website and in your own content. If it isn’t echoed across third-party sources, the model has nothing to learn it from. You can be mentioned constantly and still be narratively absent, because the model never picked up your frame from anywhere outside your own four walls.
The Consensus Gap: Present But Not Recommended
This is the exact shape of the problem: present in the answer, absent from the reasoning. A brand shows up on the list because it’s popular enough, well-known enough, or indexed enough to get pulled into retrieval. But when the model explains its pick, it reaches for whatever frame has the most third-party agreement behind it, and that frame belongs to someone else.
You end up mentioned in the same breath as the winner, described in flatter, more generic terms, while a competitor gets the specific, confident language: “best for,” “known for,” “the go-to choice when.” That asymmetry is invisible if you’re only counting how often your name appears. It’s the whole story if you’re reading why the model said what it said.
Breaking Consensus: Shifting the Narrative Sources AI Learns From
You don’t fix this by producing more content that says the same thing about yourself. Consensus shifts when the sources feeding the model start repeating a different frame, consistently, across enough places that it becomes the pattern worth learning.
That means finding out which sources are actually anchoring the current consensus in your category (which comparison posts, which forums, which analyst pieces get cited over and over) and figuring out what would need to change in those sources, or what new ones would need to exist, for the model to learn a different story about you. It’s slower than publishing another blog post. It’s also the only thing that actually moves the recommendation instead of the mention count.
Measuring Narrative Share, Not Just Mentions
This is why Mavel doesn’t stop at counting appearances. Narrative Share measures whose frame the model actually adopts when it explains a recommendation, not just who got named. Alongside that, Mavel maps the Perception Gap between how you want to be described and how AI currently describes you, and traces the Source Intelligence behind it: the specific sources feeding the consensus, so you know exactly where the story is being written and what it would take to change it.
Mentions tell you that you exist. Narrative Share tells you whether you’re winning.
If you’re tired of watching a competitor get recommended while you get listed, that’s the gap worth measuring. Talk to Mavel about what’s actually building consensus in your category, and what it would take to shift it.
