Chasing mentions in AI answers is a lagging strategy. The narrative that gets your brand mentioned was decided upstream, months before the model ever answered a prompt.
Every agency managing brand presence in AI search is running the same play right now: track mentions, count citations, try to nudge the number up. It feels like progress because it’s measurable. But it’s measuring the wrong layer.
By the time your brand shows up in a ChatGPT answer or an AI Overview, the model has already decided the story. It’s decided who the leader is, who the “budget option” is, who gets recommended for which use case. Mentions are just the visible output of a narrative decision that happened earlier, somewhere in the training data and the sources the model trusts. If you’re only tracking presence, you’re reading the scoreboard after the game’s been decided.
Why presence ≠ visibility (the symptom vs. the cause)
Ask ChatGPT “what’s the best project management tool for a 10-person startup” and you’ll get two or three names, usually with a clear favorite. That favorite isn’t chosen at random. It’s built from a frame: a story about who’s easy to set up, who scales, who’s “for small teams” versus “for enterprise.” That frame came from somewhere: review sites, comparison posts, Reddit threads, G2 categories, product-led content that got picked up and repeated until it became consensus.
Presence tools will tell you your brand showed up in the answer, or didn’t. That’s a symptom. It tells you the score, not why you’re losing or winning. If your brand is absent, the real question isn’t “how do we get mentioned more.” It’s “whose frame is the model running on, and why isn’t ours in it.”
Agencies that report mention counts to clients are reporting the temperature, not the diagnosis. Clients want to know why a competitor gets recommended and they don’t. Mention tracking can’t answer that question. It can only confirm the symptom exists.
The three narrative layers agencies miss
Most agency AI-visibility work stops at one layer: is the brand mentioned. There are three layers underneath that actually determine the outcome.
Layer one: the frame. What story is the model telling about the category? “Best for ease of use,” “best for enterprise compliance,” “best for solo founders.” Every category has 3 to 5 competing frames, and the model has picked one as dominant.
Layer two: the sources. Which sites, reviews, comparisons and forums is the model actually drawing from to build that frame? This is rarely the brand’s own website. It’s third-party consensus.
Layer three: the drift. Frames shift over time as new content gets published and re-ingested. A brand that was “the expensive one” a year ago can become “the enterprise standard” if enough sources start repeating that framing. Nobody’s watching for that shift until it’s already locked in.
Agencies that skip these layers are optimizing blind. They can push a client’s name into more answers without ever touching the reason the model prefers a competitor.
Mapping your category’s prompt universe: where is your frame absent?
Buyers don’t search “best CRM.” They ask “what CRM should a 20-person sales team use if they’re moving off spreadsheets.” That’s a prompt, not a keyword, and it comes with its own implied criteria.
Every category has a prompt universe: the real range of questions buyers and AI systems ask, from broad (“what’s the best [category] for [use case]”) to comparative (“why do people recommend [competitor] over [brand]”) to skeptical (“is [brand] worth it for a small team”). Mapping that universe means running the actual prompts, not guessing at keywords, and checking where your frame shows up, where it’s missing, and where a competitor’s frame has taken the space you should own.
This is where agencies can build a real audit for clients: not “we’re mentioned in 40% of answers” but “here’s the exact set of questions where your narrative is absent, and here’s whose frame is filling that gap instead.”
Source intelligence: audit the sources the model actually draws on
If the frame comes from sources, the fix has to start with sources. A dashboard that says “you’re losing” doesn’t tell you where to act. Source intelligence traces the specific citations, comparison pages, and forum threads the model is pulling from to build its answer about your category.
Picture two competing SaaS tools in the same space. One gets recommended for “best value,” the other for “best support.” Trace the citations behind each answer and you’ll usually find a handful of repeated sources: a G2 comparison, a Reddit thread from 18 months ago, a blog post that got picked up by three other sites. That’s the actual input shaping the output. Fix the source, and the frame eventually shifts. Ignore it, and no amount of new content on the brand’s own site will move the needle, because the model isn’t citing the brand’s site. It’s citing everyone talking about the brand.
The playbook: 5 moves to own the frame before the model updates
- Map the prompt universe for the category. Run the real questions buyers ask, not guessed keywords.
- Identify the dominant frame the model has adopted, and name it plainly: “budget pick,” “enterprise-safe,” “easiest onboarding.”
- Trace the sources feeding that frame. Find the 5 to 10 pieces of content actually doing the work.
- Find the gaps where your frame is absent and a competitor’s is filling the space uncontested.
- Ship source-level changes (new comparisons, corrected claims, updated third-party content) before the next training or indexing cycle locks the current frame in further.
Models update. Indexes refresh. The frame you own today isn’t guaranteed tomorrow, and neither is the one you’re losing to. The agencies that treat this as a one-time audit will get outpaced by the ones treating it as ongoing narrative maintenance.
Proof: tracking narrative share, not just mentions
Imagine two SaaS brands in the same category, both showing up in AI answers at similar rates: call it 35% mention frequency each. Mention tracking says they’re neck and neck. But look at whose frame the model is actually using: one is described as “the reliable choice for teams that need compliance,” the other as “a decent option if you’re on a budget.” Same mention rate, completely different recommendation weight. The first brand is winning the category. The second is a footnote that happens to get named.
That’s narrative share: whose story the model is telling, not just whether your name appears in it. It’s the metric that actually explains why a client is losing deals to a competitor even when both brands “show up” in AI search at similar rates. And it’s the number agencies need if they’re going to prove ROI on AI visibility work that goes beyond a mentions graph trending up and to the right.
If you’re an agency trying to explain to a client why a competitor keeps getting recommended over them, mentions won’t get you there. You need to see the frame, the sources, and the gap. That’s what Mavel is built to show. Worth a look before your next client review.
