Is Your Company Credible, or Just Cited?

VeritasLinks is not a GEO or "AI visibility" tool [F4904]. This distinction matters because traditional SEO measurement tools are failing to capture brand visibility when brands are cited in AI answers without generating clicks [F4208]. A company can be highly visible in model outputs yet remain barely preferred by actual buyers. Citation is not recommendation. Being mentioned is not the same as being judged credible.

The gap between visibility and preference is where risk hides. High visibility does not guarantee that a model will select you when the stakes are high. It means you are in the room, but it does not mean you are being chosen. This is the "ghost ranking" problem. You see your name in the text, but the model’s final decision leans toward a competitor with lower visibility but higher judgment scores.

The data confirms this divergence. In July 2026, VeritasLinks analyzed several AI-visibility vendors to separate brand visibility from simulated buyer preference. Semrush showed high visibility at 135 but low preference at just 3. Profound followed with visibility of 130 and preference of 5. AthenaHQ had visibility of 129 but preference of only 3. Ahrefs Brand Radar posted visibility of 127 against a preference of 2. Nightwatch recorded visibility of 125 with preference of 6. Bluefish showed visibility of 121 and preference of 5.

The chart shows that high visibility vendors often have low preference, proving the gap between citation and recommendation.

Model disagreement is not statistical noise. It is a structural risk for teams making high-stakes decisions based on stochastic outputs. When an M&A analyst or KYB reviewer asks an AI to evaluate a vendor, the tool does not return a single truth. It returns a distribution. VeritasLinks measures this judgment layer directly, capturing credibility and recommendation signals rather than just counting citations [F634]. This approach reveals where models agree and where they fracture [F419].

The phenomenon is known as Split-Perception. Different models judge the same company differently, and that gap is a measurable risk signal. VeritasLinks founder Andrew Pomazkov published the underlying research on SSRN [C6]. A company might be highly visible in one model’s answers yet barely preferred in another’s. Being mentioned is not the same as being recommended [C2]. If your due diligence relies on a single-pass snapshot, you may be betting on a random draw rather than a stable fact [C3].

This disagreement matters because models project their perception onto companies, which then influences purchases and partnerships. A company not on the short list of a key model does not just rank low. It is effectively not in the room. VeritasLinks tracks this across ChatGPT, Claude, Grok, Perplexity, and DeepSeek, showing separate scores for each [C2]. The spread between these scores indicates the stability of the perception. A wide spread suggests the company’s reputation is fragile, dependent on which model is asked.

The chart illustrates the magnitude of this risk. It plots the gap between the highest and lowest model scores for the same company on the VeritasScore scale of 300 to 870. For a typical company, the median gap is 100 points. At the 90th percentile, one in ten companies shows a gap of 134 points. This data confirms that model disagreement is not an edge case. It is a standard feature of the current AI landscape. The chart shows how far apart AI models score the same company, with a median gap of 100 points and a 90th percentile gap of 134 points.

Ask the same question to an AI model ten times and you will likely get ten slightly different answers [C2]. This is not a bug in the technology. It is the nature of stochastic generation. Most visibility tools run a single pass and report the result as if it were a fixed fact. They count citations and call it visibility. But citation is not recommendation. A model can mention your name while silently preferring a competitor.

VeritasLinks measures the judgment layer underneath the text. We do not rely on one snapshot. We run multiple prompts across several models, including ChatGPT, Claude, Grok, Perplexity, and DeepSeek [C2]. Each model answers repeatedly. This allows us to measure the spread of responses [C3]. We separate the signal from the noise. We track where models agree and where they diverge. This divergence, or Split-Perception, is a risk signal [C5]. If one model recommends you and another remains neutral, the gap matters.

The result is a VeritasScore on a 300 to 870 scale [C4]. This score reflects consistency, not just frequency. It captures whether models consider your company credible and where they disagree. The analysis also includes AI-simulated focus groups where synthetic buyer personas discuss the company and surface objections [C7]. It includes an AI Investor Review that reads the company the way an investor or credit officer would [C8]. These are not static metrics. They are dynamic perceptions.

Scores are independent. There is no paid placement, and scores cannot be bought [C13]. This matters for due diligence. A high score means something because it is earned through repeated measurement, not purchased. The research behind Split-Perception is published on SSRN [C6]. It confirms that different models project different perceptions onto the same entity. VeritasLinks quantifies this gap, giving you a clearer view of your actual standing in the AI landscape.

The chart above shows where companies land on the VeritasScore scale, with a median of 437 among the 139 companies analyzed as of July 2026.