Wedge Visibility — The AI Search Metric Beyond Surface Rate

Most "AI visibility" dashboards report one number: surface rate. How often does the model mention your brand at all. It's the right number for about a month after launch, and the wrong headline forever after. Strong brands saturate it fast. We ran a 50-cell Brand Radar against Linear across ChatGPT, Claude, Gemini, AI Overviews, and Perplexity. Surface rate: 94%. We ran the same protocol against Notion: 85%. At those levels the question "are we visible to AI search" is answered. Yes. You're in the room.

So stop asking it. The second-order question is the one that decides outcomes: when the model names you, does it also say what makes you different? Or does it list you as one of five interchangeable options and move on?

That's wedge visibility. It moves independently of surface rate, almost nobody measures it, and it's where AI-search positioning is actually won or lost.

What a wedge is, and what wedge visibility measures

A wedge is your single sharpest differentiator — the one thing a customer says when they recommend you without a script. Linear's wedge is speed: the keyboard-driven, sub-100ms, no-loading-spinner experience. Stripe's wedge is developer-first. Notion's wedge is flexibility. Not a feature list. The one attribute that, if a buyer remembers nothing else, is the reason they pick you.

Wedge visibility is the rate at which AI answers surface that differentiator when they name your brand. It's conditional. The denominator is "answers that mentioned you at all." The numerator is "answers that also carried your wedge." See /glossary/wedge-visibility for the formal definition.

This is a different axis from surface rate — the metric most tools call citation rate — and the two move independently. Surface rate asks: are you named? Wedge visibility asks: named as what? You can be named everywhere and characterized as nothing. A model can list you in every comparison and never once say the thing that makes a buyer choose you. High on one axis, flat on the other. That gap is invisible on a surface-rate dashboard, and it's the most common failure mode we see.

The Linear worked example

Linear's speed wedge surfaced in 77% of the cells where Linear was mentioned, aggregated across the full audit. Respectable headline. But the aggregate hides the entire lesson. Break it down by query type and the number swings violently:

Category queries ("best modern issue tracker") — 93% wedge visibility. Ask the open category question, the model reaches for Linear and reaches for "fast" in the same breath.
Comparison queries ("Linear vs Jira") — 93%. Head-to-head framing pulls the differentiator out reliably; that's the shape of query where contrast is the whole point.
Branded queries ("is Linear worth it") — 50%. Name the brand directly and half the time the answer recites features without naming the one that matters.
JTBD migration ("how do I migrate from Jira to a faster issue tracker") — jumped 0% → 80% between rounds. The reversal happened when the query phrasing itself carried a speed signal. Put the wedge word in the question and the answer surfaces it.
JTBD workflow ("best sprint planning workflow") — 0%. Linear surfaced as a tool. The reason to choose Linear did not.

Same brand. Same week. Same five engines. Wedge visibility ranges from 0% to 93% depending purely on the shape of the question. And surface rate hid all of it — Linear was surfaced at roughly 90%+ across every one of those query types. A surface-rate dashboard would have shown a flat green line over the exact week where the brand's differentiator was vanishing on half its highest-intent queries. Full numbers and methodology at /audits/linear.

Why the two metrics diverge

Surface rate and wedge visibility come from different machinery, which is why they drift apart.

Surface rate is driven by brand-name frequency: how often your name appears in training data and how reliably retrieval pulls a page that mentions you. Get mentioned enough places and you get named. That's a reach game, and strong brands win it almost automatically.

Wedge visibility is driven by something narrower — whether the indexed sources connect your brand to the differentiator in phrasings a model is likely to lift. It's not enough for "Linear" and "fast" to both exist on the internet. They have to co-occur, attributively, in citation-likely sentences. "Linear is the fast one" does the work. "Linear has many features including performance improvements" does not.

The mechanism behind low wedge visibility is almost always the same. Your own marketing pages bury the differentiator behind a feature grid — the wedge is in there, three bullets down, hedged. Meanwhile the sources that carry it cleanly are third-party: the Hacker News comment that says "switched for the speed," the review that leads with it, the comparison post built around it. Those phrasings feed the model's knowledge graph far more directly than your homepage does. When third parties don't say why you're different, nothing does — and the model defaults to listing you as a generic category member.

The four wedge-visibility states

Plot surface rate against wedge visibility and you get a 2x2. Each quadrant is a different problem with a different fix.

High surface + high wedge visibility — Positioned correctly. The model names you and characterizes you the way you'd characterize yourself. This is the goal state. Defend it; don't assume it's permanent.
High surface + low wedge visibility — Mentioned as an interchangeable category member. The model lists you and says nothing about why you'd be chosen. This is the most common and most dangerous state, because it looks perfectly healthy on any surface-rate dashboard. You're "visible." You're also forgettable. The fix is differentiator-content investment, not more reach.
Low surface + high wedge visibility — Niche but sharp. When you do come up, the model knows exactly what you're for. The fix is reach: get named in more places.
Low surface + low wedge visibility — Category-level repositioning. The model neither names you reliably nor knows what you'd be for. This is foundational work, not a content sprint.

Most brands tracking AI visibility only ever see the horizontal axis. They miss the vertical one entirely — which means they can't tell the difference between the top-right quadrant and the second one, the trap that feels like winning.

How to measure your own wedge visibility

Reproducible, no special tooling required.

Pick your wedge. Not the one your positioning deck claims — the one customers use unprompted. Read your testimonials, your sales-call transcripts, your Reddit and HN threads. The word that recurs when someone recommends you is your wedge. Usually one word.
Run 10 anti-prime queries across all five engines — ChatGPT, Claude, Gemini, AI Overviews, Perplexity. Anti-prime means the query describes the buyer's problem without naming your brand or your wedge. (See /glossary/anti-prime-queries.) Fresh session each time.
For every cell where your brand surfaces, score one thing: did the differentiator surface too? Yes or no.
Compute wedge visibility = cells-where-wedge-surfaced ÷ cells-where-brand-surfaced. That conditional is the whole point — you're only scoring answers that named you.
Break it down by query type — category, comparison, branded, JTBD. As the Linear numbers show, the aggregate lies. The variance between query types is the actionable part; that's where you find the 0% cells hiding inside a 77% average.

Free path: score answers by hand with the LLM quote extractor, which parses raw responses into structured citations so your yes/no scoring stays consistent. Scaled path: Brand Radar runs the whole protocol weekly and tracks both axes over time.

The content prescription

Low wedge visibility is almost never a writing problem. It's a distribution problem. Your blog post saying "we're fast" is one signal in the index. Forty Hacker News comments saying it are forty calibration signals — and they outweigh you, because the model trusts the crowd's characterization over yours. So the work isn't writing better copy on your own pages. It's getting the differentiator carried elsewhere, in the right words.

Three moves, in order of leverage:

Get third-party sources to carry the wedge. Reviews, comparison content, community threads. Each one that leads with your differentiator is a calibration signal the model weights heavily. This is the highest-leverage and the slowest.
Ship JTBD landing pages whose H1 carries the differentiator vocabulary. The Linear migration reversal — 0% → 80% — proves the mechanism: when query phrasing carries the wedge word, the answer surfaces it. Pages built around "faster issue tracking" teach the index that the JTBD and your wedge belong together.
Stop optimizing your own pages for keyword density. Optimize for differentiator-density across other people's pages. The wedge needs to live where the model trusts it, not where you control it.

Close

Surface rate tells you if you're in the room. Wedge visibility tells you whether the room knows why you matter. Past launch, the first metric saturates and the second is the one that moves revenue. Track only the first and you'll celebrate a flat green line over the exact weeks your positioning is dissolving.

Brand Radar measures both, free tier covers one brand — start here, or score a single answer by hand with the LLM quote extractor. Either way: measure the axis that's actually contested.

FAQs

What is wedge visibility? The rate at which AI answers surface your single most important differentiator when they name your brand. It's conditional on being mentioned — the denominator is answers that named you, the numerator is answers that also carried your wedge. It is distinct from surface rate, which only asks whether you were named at all.

How is wedge visibility different from surface rate? Surface rate asks "are you named?" Wedge visibility asks "named as what?" Surface rate is driven by brand-name frequency in training data and retrieval. Wedge visibility is driven by whether sources connect your brand to your differentiator in citation-likely phrasings. They move independently — you can be named everywhere and characterized as nothing.

What was Linear's wedge visibility in the audit? 77% aggregated across mentioned cells, but the breakdown is the lesson: 93% on category and comparison queries, 50% on branded queries, and 0% on a generic JTBD workflow query — all in the same week. A migration query jumped from 0% to 80% once the phrasing carried a speed signal. Full numbers at /audits/linear.

Why is high surface rate plus low wedge visibility the dangerous state? Because it looks healthy on every surface-rate dashboard. You appear "visible" while the model lists you as an interchangeable category member and never says why a buyer would choose you. You're named and forgettable at the same time, and the standard metric can't see the problem.

How do I raise wedge visibility? Treat it as distribution, not copywriting. Get third-party sources — reviews, comparisons, community threads — to carry your differentiator in their own words. Ship JTBD landing pages whose headlines use the differentiator vocabulary. Stop optimizing your own pages for keyword density and start raising differentiator-density across other people's pages, where the model trusts the characterization more.