Is Your Company’s AI Reputation Stable or Random?

Most AI visibility tools count citations. They report how often a model mentions your name. This is not the same as credibility. Being mentioned is passive. Being judged is active. VeritasLinks measures the judgment layer underneath the citations [F4532].

A company can be highly visible yet barely preferred. It might appear in answers, but not be chosen. VeritasLinks separates consistent credibility from random citation [F4532]. It runs multi-model, repeated analyses to distinguish signal from stochastic noise and model bias [F4533]. This approach reveals where AI models disagree about the same company [F4535].

The gap between models is a risk signal. Different models project different perceptions onto a company. This Split-Perception means the score is not a single fact. It is a spread of opinions. VeritasLinks scores are independent of paid placement or model bias [F4534]. Goldman Sachs, Citigroup, and JPMorgan Chase are warning investors about risks from AI software hallucinations [F4528]. If your due diligence relies on a single model pass, you might be trusting a random draw.

Consider the AI visibility vendors analyzed by VeritasLinks in July 2026. Semrush showed a visibility score of 135, but a buyer preference of only 3. Profound had a visibility of 130, yet a preference of 5. AthenaHQ recorded a visibility of 129 with a preference of 3. Ahrefs Brand Radar reached a visibility of 127 but lagged with a preference of 2. Nightwatch maintained a visibility of 125 and a preference of 6. Bluefish held a visibility of 121 with a preference of 5.

The chart shows that high visibility does not guarantee buyer preference, revealing the true cost of model disagreement.

VeritasLinks avoids the single-pass trap by running repeated analyses across multiple large language models. A single question asked to ChatGPT yields one answer. VeritasLinks asks it many times, measuring the spread of those answers to separate signal from stochastic noise [F4533]. This method distinguishes consistent credibility from random citation, which is critical for risk teams who cannot afford to act on a hallucination [F4532]. The platform analyzes the company across ChatGPT, Claude, Grok, Perplexity, and DeepSeek, providing a separate score for each model [C2].

The stability of these scores matters because retrieval-augmented generation can increase hallucination rates by biasing token statistics locally [F4531]. While purpose-built GenAI tools for law produce hallucinations 17% of the time according to a Stanford study, general-purpose models introduce their own variance [F4530]. VeritasLinks measures this variance by repeating prompts and calculating the difference between the highest and lowest model scores. This gap, or Split-Perception, is not just a metric. It is a risk signal that reveals whether a company’s reputation is solid or merely a momentary alignment of random weights.

The data shows that model disagreement is significant. For a typical company, the median gap between the highest and lowest model scores is 100 points on the 300 to 870 scale [cross_model_gap]. For one in ten companies, that gap widens to 134 points [cross_model_gap]. This means two different AI agents could look at the same due diligence report and arrive at different conclusions about the same vendor. VeritasLinks quantifies this disagreement so you know when to trust the consensus and when to dig deeper.

The chart illustrates the scale of this uncertainty by showing how far apart AI models score the same company, with the typical median gap at 100 points and the 90th percentile reaching 134 points on the VeritasScore scale.

VeritasLinks scores are independent of paid placement or model bias [F4534]. This independence is a structural feature, not a marketing promise. The score cannot be bought, and it does not rely on sponsored visibility metrics. For enterprise risk and compliance teams, this distinction is critical. You are evaluating the judgment of the model, not its advertising budget.

The VeritasScore operates on a 300 to 870 scale [C4]. It is built from several components that measure how large language models perceive, describe and judge a specific company [C1]. These components include multi-model analysis across ChatGPT, Claude, Grok, Perplexity and DeepSeek [C2]. Each model receives a separate score, revealing where perception diverges. The platform measures the spread of answers by running many prompts with repeats, rather than reporting a single pass as fact [C3]. This method separates signal from noise in stochastic outputs.

The scoring aligns with major trust frameworks, including the EU AI Trust Act, ISO 42001, NIST RMF 600.1, and OWASP Top 10 LLMs [F4539]. As regulatory bodies enforce compliance for systems generating synthetic content, the stability of these scores becomes a risk indicator. Providers placing systems before August 2, 2026, must meet marking requirements by December 2, 2026 [F4544]. General-purpose AI models face substantive obligations until August 2, 2027 [F4543]. VeritasLinks provides the measurable baseline for these deadlines.

The chart above shows where companies currently land on this scale. Among 139 companies with fully scored analyses in July 2026, the median score is 437, with only about one in eight scoring 500 or more and none reaching 700.