A brand mention gap analysis answers one question with receipts: on which buying questions does AI recommend your competitors and skip you — and which specific pages are causing it?
It's the AI-era equivalent of googling yourself, and most teams have never done it. The output isn't a vanity score — it's a prioritized to-do list of real URLs. Here's the full walkthrough; the manual version takes about an hour.
Take your tracked prompt set (or build one — method here; 10 questions is enough for a first pass). Make a grid: questions down the side, one column per brand — you plus every competitor you care about, with a few empty columns for brands you didn't expect. Ask each question in ChatGPT (and ideally Gemini and Perplexity — the engines disagree more than people assume1) and tick every brand named. Ask each question twice on different days if you can: single runs are noisy.2
The mentions grid
Questions × brands — filled where AI names the brand · 2 = named consistently, 1 = sometimes, 0 = never
A typical first grid: total gaps on the head terms, partial gaps mid-list, and one defensible question at the bottom. Every empty cell is a to-do with a URL behind it.
Total gaps— questions where competitors appear and you never do. These are lost deals you never saw; they're also your highest-priority fixes.
Partial gaps — you appear sometimes, or only on some engines. Usually an evidence-density problem: the corpus knows you, but not confidently. These move fastest.
Position gaps— you're named, but consistently last, hedged, or described worse than rivals. Read the actual sentences: how AI frames you is a preview of how buyers meet you.
Also note the reverse: questions where youshow and a competitor doesn't. That's defensible ground — you'll want to know what's feeding it so you don't accidentally stop.
For each total gap, interrogate the answer: “Why those brands? Which sources say they're the best? Cite them.” When live search fired, you'll get actual URLs; when it didn't, ask what sources wouldsupport the claim and verify them yourself. Across ten questions you'll converge on a shockingly short list — typically a handful of listicles, one or two review platforms, and a few community threads doing most of the damage. That concentration is the good news: the public citation studies show a small set of trusted pages supplies an outsized share of AI's evidence,3 which means a small set of placements can flip a lot of answers.
For each gap source, score two things 1–5: impact (how many of your tracked questions cite this page × how commercial those questions are) and effort(getting added to a listicle = pitch an author; review-platform presence = a program; an old Reddit thread = one honest reply). Work the high-impact, low-effort corner first. A typical first sprint: answer two community threads, pitch three listicle authors, kick off a review push on the one platform your category's answers keep citing.
Re-run the same grid on a cadence and watch gaps change state: total → partial → won. Because you traced sources first, movement is attributable — when a mention appears two weeks after a listicle adds you, you know what worked and what to do more of. Log the reverse too: mentions you lose flag sources that changed under you.
Manually this is an hour per pass, which is exactly why most teams do it once and stop — and why we built the automated version: Beacon runs the grid daily across five engines, extracts the citing sources, and turns every gap into a ranked action. But automated or manual, do the analysis. The brands beating you in AI answers are mostly beating you on pages you've never audited.
Beacon grids your buyers' questions across five AI engines daily, traces the citing sources, and ranks every gap by impact.