The Two-Part GEO Audit: Exposure × Citation
Our GEO audit is written in two parts, because a brand can fail in AI answers in two completely different ways that need different fixes. Part One, Exposure, asks whether AI knows you and talks about you: brand awareness and factual accuracy, share of voice against competitors, and the sentiment of what is said. Part Two, Citation Probability, asks whether AI uses your pages as its evidence: your citation rate, and how well you are presented when you do appear. Blending the two into one score hides the only thing an audit exists to reveal — which lever to pull. This post sets out the framework and the anonymized case that forced us to build it.
Why one report became two
Early audits reported a single visibility picture, and it kept producing an unanswerable meeting. A client could be famous in AI answers and absent from their sources, or well-cited and never recommended, and the blended view made both situations look like the same moderate problem with the same moderate prescription: more content, more signals, more everything. The fixes, though, live in different departments. Exposure is earned in the third-party record — directories, trade press, reviews, community discussion. Citation is earned on your own pages — facts stated in passages a model can lift. An audit that cannot tell the client which of those two budgets to spend has failed at its only job. So the report split in two, and the split is now the framework: 上篇 exposure, 下篇 citation probability — published in Chinese as the two-part framework our clients receive.
Part One: Exposure
Exposure measures whether the platforms know the brand at all, and what they believe about it. Three components. Awareness and accuracy: asked directly in buyer language, does the platform identify the brand correctly — right industry, right products, right business model — or does it improvise, as in the cognition case elsewhere in this blog? Share of voice: across the category's buyer questions, how much of the answer space carries the brand's name versus named competitors? Sentiment: when the brand is described, is the framing positive, neutral, or cautionary — and is it built from facts or from the absence of them? Exposure is the brand layer of the audit, and its raw material is everything published about the brand by everyone except the brand.
Part Two: Citation probability
Citation probability measures whether the brand's own pages function as evidence. Two components. Citation rate: in how many answers do the brand's pages appear in the sources? Presentation quality: when they do appear, how are they presented — cited for the claims that matter or for trivia, positioned among the top sources or buried, quoted accurately or paraphrased into someone else's point? A citation that misrepresents you is a liability wearing a metric's clothing. The product layer dominates this part: platforms cite pages, not companies, so every product or SKU is tested on the strength of its own page — its facts stated answer-first, self-contained, and quotable.
The 2×2 grid, and the case that forced it
The two parts cross the two objects of measurement into a grid. Brand × Mention: does the answer name the company? Brand × Citation: are the company's own pages used as sources? Product × Mention: does the answer name the product a buyer would actually buy? Product × Citation: is the product page itself cited and recommended? Recommendation sits across the grid as the outcome the other four cells predict — the sentence where the decision happens.
The case that forced the framework is the anonymized Cixi home-appliance OEM documented in this blog's Failures section. Its demo scores: mention rate 100%, citation rate 0%, share of voice 54.5%. Read through a one-part audit, that brand looks half-healthy — better than most. Read through the 2×2, the diagnosis is exact: maximum exposure, zero citation, and — because nothing anchored its descriptions to its own pages — three AI systems describing three different wrong companies. The fix list wrote itself, in the framework's order: cognition and citation first, because exposure was already amplifying errors. That is what the two-part structure is for. Not a tidier report — a report whose bad news arrives with its cause attached.
Sources & method. The two-part framework (Exposure: awareness, share of voice, sentiment; Citation Probability: citation rate, presentation quality; 2×2 of brand/product × mention/citation) is our own audit standard, documented in the GEO two-part audit framework used across this practice's engagements. Case figures (mention 100%, citation 0%, share of voice 54.5%) are from our demo measurement of an anonymized home-appliance OEM in Cixi, Zhejiang, detailed in “What AI Thinks Your Factory Makes” in this blog.
© 2026 Jun Liu · JUN & GEO. This article is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License (CC BY-SA 4.0) — https://creativecommons.org/licenses/by-sa/4.0/. You may share and adapt it with attribution and under the same license.