Vantae AI Visibility Methodology

Vantae measures how AI systems understand, describe, compare, and recommend a brand across five scored dimensions. Each crawled page receives a score per pillar on a 0–100 scale; those scores roll up into a project-level composite.

The five pillars are Trust, Conversion, Clarity, SEO, and AI Agent. Trust and Conversion carry the highest weight in the composite because they most directly determine whether an AI system will recommend or cite a page.

Vantae evaluates ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google), and Perplexity independently because each provider has its own retrieval behavior, model training, response generation approach, and citation behavior.

The same brand query can produce meaningfully different results across providers. A brand that is accurately represented in Claude may be absent in Perplexity, or may be described with different facts in Gemini. Vantae records and compares results per provider rather than averaging across systems.

Vantae does not have access to proprietary provider algorithms, training data, index contents, or retrieval configurations. All evaluations are based on observing the outputs AI providers return in response to structured queries.

AI visibility is measured by sending tracked queries to each enabled AI provider and analyzing the response for brand mentions, description accuracy, comparison context, and recommendation signals.

The headline metric for a project is one of two things, depending on what the providers returned:

• AI mention rate — the percentage of provider responses that mentioned the brand at all. Used when no citation data was returned.
• AI share of voice — when citations are available, the percentage of total brand citations across all responses that pointed to the evaluated brand. Used when citation data is present.

Beyond the headline metric, Vantae records: the description the AI used for the brand, any competitors mentioned in the same response, the context in which the brand was recommended or excluded, and whether the information returned was accurate relative to the content on the site.

Per-provider breakdown is always available so teams can see exactly how representation differs between ChatGPT, Claude, Gemini, and Perplexity on the same query.

When an AI provider returns citation or source data alongside its response, Vantae records which pages were cited and whether they point to the brand being evaluated.

Citation analysis covers:

• Citation presence — whether the brand's pages appeared in the provider's cited sources for this query.
• Source identification — which specific URLs were cited, when citation data is returned by the provider.

Limitations of citation analysis. Not all AI providers return structured citation data for all queries. Citation availability depends on the provider, the query type, and current provider configuration. When citations are not returned, Vantae falls back to brand mention rate and labels the headline metric accordingly. Citation patterns can also change as provider models are updated.

AI assistants can only cite content they are permitted to read. Vantae evaluates AI crawler access by checking four publicly accessible files at the site root on every scan:

• robots.txt — the canonical allow and disallow list for crawler agents. Vantae parses each major AI crawler user agent and reports whether it is blocked, partially blocked, or allowed.
• sitemap.xml — checks for presence, accessibility, and parseability.
• llms.txt — an emerging convention for providing LLM-friendly site summaries and navigation hints.
• ai.txt — an emerging convention for AI-specific permissions and metadata.

The AI crawler user agents currently tracked by Vantae:

AI Agent is Vantae's measurement of whether an AI agent can complete category-typical tasks on a website — such as finding pricing, booking a demo, or locating support contact information.

Unlike the other four pillars, which evaluate content quality from a static analysis of the page, AI Agent testing measures actual agent behavior on the live site. The agent attempts a set of tasks and Vantae records whether each task succeeded, where the agent encountered obstacles, and what site changes would have helped.

Common obstacles that reduce AI Agent scores include:

• Forms or actions without clear, machine-readable labels
• Navigation that requires hover or JavaScript interaction to reveal options
• Pricing or booking flows blocked behind login without a visible alternative
• Key pages not reachable from the homepage within a small number of steps
• Stateful URLs that do not persist when shared or linked directly

AI Agent testing is enabled per project from Project settings → Scoring → AI Agent. When active, it adds a fifth pillar to the project score and the radar chart.

Recommendations are generated from observable evidence gathered during the scan — content on specific pages, signals in AI provider responses, and crawler access results. Every recommendation is tied to a specific observation, not to a general best-practice assumption.

Each recommendation distinguishes four components:

• Observation — what Vantae found on the page or in the provider response. For example: "The pricing page does not state a price range."
• Evidence — the specific text, signal, or provider response excerpt that supports the observation.
• Impact — the expected effect on AI visibility if the change is made. For example: "AI assistants querying about pricing are more likely to cite a page that states a price range explicitly."
• Recommendation — the specific change to make, with copy-paste-ready before/after text when applicable and a link to the page where the change should occur.

Recommendations are ranked by cross-provider agreement (findings flagged by multiple AI providers carry higher confidence) and expected lift. Priority levels are P1 (urgent), P2 (important), and P3 (minor).

AI visibility measurements reflect a specific point in time and a specific set of query prompts sent to each provider. There are several important limitations to understand before interpreting results:

• Generative AI responses are not deterministic. The same query sent to the same provider can return different results on different days. Responses can change due to model updates, retrieval behavior changes, index updates, prompt phrasing variations, user context, geographic differences, and provider configuration changes.
• Synthetic measurement, not real-user sessions. Vantae sends structured queries directly to AI providers — it does not measure what real users actually see in their personalized sessions. Personalization, account state, browser context, and regional settings can all shift individual results.
• A score is a measurement signal, not a guarantee. An AI visibility score measures observable signals at the time of the scan. It is not a guarantee of ranking, citation frequency, or recommendation in any particular user's session.
• Citations are not always available. Some providers return structured citation data for some queries; others do not. When citations are not returned, Vantae falls back to brand mention rate and labels the headline metric accordingly.
• Crawler access reflects the site at scan time. robots.txt and related files are checked as they exist at the moment of the scan. Changes to these files take effect on the next scan.

Because AI provider behavior changes over time, Vantae tracks results scan-over-scan so improvements and regressions become visible as trends rather than isolated snapshots.