How Vantae measures AI visibility.

Vantae measures how visible your brand is to the AI assistants people now ask before (or instead of) searching. We don't crawl real-user traffic — we run a repeatable synthetic measurement that makes the question testable.

For each project we track three layers:

• What AI says about you. We send your tracked queries to each enabled provider and record the answer, the brands mentioned, and (when available) the citations.
• Whether AI can read your site. We check robots.txt, sitemap.xml, llms.txt and ai.txt for every major AI crawler.
• Whether AI agents can use your site. When AI Agent testing is enabled, we let real LLM agents attempt category-typical tasks (find a price, book a meeting, locate support) and score what happens.

Each pillar is scored 0–100, then the five pillar scores are combined into an overall composite weighted by how important each pillar is to AI readiness.

Within-pillar scoring. Each AI finding for a pillar carries a priority (P1 urgent / P2 important / P3 minor) and a confidence score from the AI providers. Scores are penalised proportionally to the number and severity of findings, weighted by how important each page is to the site. A high-confidence P1 finding on the homepage docks a larger penalty than a low-confidence P3 finding on a deep blog post.

Cross-pillar composite. Pillar scores are combined using a weighted approach that penalises imbalance — a site that scores well on most pillars but poorly on one does not achieve a high overall. Trust and Conversion carry the most weight. Improving a weak pillar can have greater impact, but the result also depends on pillar weights, Trust gating, and score rounding.

Trust gating. A critically low Trust score can pull down your overall composite regardless of how well other pillars perform, because AI agents won't act on sites they don't trust. This models the analogous "governance qualifier" used by credit rating agencies: when credibility is critically low, no amount of conversion or SEO performance can fully compensate.

Scores are grouped into ten bands calibrated against the OECD/JRC composite indicator handbook. The bands are fixed thresholds, not percentiles, so improving your site always moves the needle.

We score five pillars. The first four always count; AI Agent only counts when the project enables it. Trust and Conversion carry the most weight in the composite, reflecting their outsized impact on whether AI agents recommend or act on a site.

AI Visibility answers: when someone asks an AI assistant a query you care about, how often does your brand show up in the answer?

For every tracked query, we send the prompt to every enabled provider and parse the response for brand mentions. The headline number on a project's dashboard is one of two things, depending on what the providers returned:

• AI mention rate — the percent of provider responses that mentioned your brand at all. Used when no citations were returned.
• AI share of voice — when citations are available, the percent of total brand citations across all responses that pointed at your brand.

Either way the number is a direct count over completed runs — not an estimate.

On the comparison surface, Share of Voice is the same family of metric framed relative to the competitors you've configured. We compute, across all tracked queries and all enabled providers, the percent of brand mentions or citations that went to each competitor.

A few things to keep in mind:

• Closed universe. Share is computed over the brand list you configure in Project settings → Brand. Brands not on that list aren't counted toward the denominator.
• Per-provider breakdown. The Comparison page also shows share broken down by provider, since the same query can produce very different mention patterns in ChatGPT vs Gemini vs Perplexity.

AI assistants can only cite content they're allowed to read. Every scan re-checks four files at the site root and per-bot allow/disallow rules:

• robots.txt — the canonical allow/disallow list. We parse it for every bot we track and report which crawlers are blocked, partially blocked, or allowed.
• sitemap.xml — present, parseable, and reachable.
• llms.txt — emerging convention for an LLM-friendly site map.
• ai.txt — emerging convention for AI-specific permissions.

The bots we currently track:

• · GPTBot (OpenAI)
• · OAI-SearchBot (OpenAI)
• · ChatGPT-User (OpenAI)
• · ClaudeBot (Anthropic)
• · Claude-Web (Anthropic)
• · PerplexityBot (Perplexity)
• · Google-Extended (Google)
• · Applebot-Extended (Apple)
• · CCBot (Common Crawl)

The list grows as new bots are deployed; the Crawler Signals card in the dashboard always shows the current verdict per bot.

AI Agent answers a different question from the visibility metrics: if an AI agent actually visits your site to do something, can it succeed?

When AI Agent testing is enabled for a project, our agent attempts a small set of category-typical tasks — for example "find pricing for the Pro plan", "book a demo for next Tuesday", "find support contact information". For each attempt we record whether it succeeded, where it got stuck, and what would have helped (clearer headings, exposed structured data, a less-modal contact form).

AI Agent is configured per project from Project settings → Scoring → AI Agent. Once active, it adds a fifth pillar to the score and the radar chart. The project score adjusts automatically so historical trend lines stay accurate as AI Agent data accumulates.

Not all pages carry equal weight. A broken call-to-action on the homepage costs more than a missing alt-tag on a deep blog post. We classify every page URL into one of three importance tiers and scale each finding's penalty accordingly:

• Critical — highest impact. Homepage, pricing, checkout, booking, services, demo, signup, and other high-traffic or decision-stage paths. Findings here carry the full penalty.
• Major — significant impact. About, products, portfolio, testimonials, team, and other brand/trust pages with meaningful traffic. Findings here carry a reduced penalty.
• Minor — lower impact. Blog posts, news articles, FAQs, and deep interior pages. Fixing these is good, but the score impact is intentionally lower to reflect lower expected traffic and fix leverage.

The classification is Vantae's product heuristic, based on the URL path segment (case-insensitive), not an NN/g page taxonomy. For the broader interaction principle, NN/g's Heuristic #7 is Flexibility and Efficiency of Use; it does not prescribe Vantae's page-weighting formula.

We currently query the following AI assistants for every tracked query:

Scans are scheduled per project. Manual scans run on demand from the project dashboard; scheduled scans run on the cadence configured in Project settings → Cadence (typically daily or weekly). Citation parsing, brand-mention detection, and crawler checks all run as part of the same scan.

When multiple AI providers evaluate the same page, they don't always agree. We measure how consistently different AI providers reach the same conclusions about your site — a finding flagged by every provider carries more weight than one raised by only a single model.

Agreement is reported per-pillar on the Scan Detail page:

• High agreement — all providers flagged the same issues with similar severity. High confidence; findings are reliable.
• Mixed agreement — providers are split; treat individual findings with some caution.
• Low agreement — providers are largely contradicting each other, which sometimes indicates genuinely ambiguous content.
• Insufficient data — fewer than two providers returned data for this pillar; agreement cannot be measured.

We store the raw provider response, the parsed brand-mention list and the citation list for every run, so any number on a dashboard can be drilled down to the underlying evidence. Per-run evidence is visible inside the project's Queries surface.

We don't share your scan data, your tracked queries, or your brand list with other organizations. Scoring weights, brand lists, and competitor lists are scoped to your organization.

A few things this measurement is and isn't:

• Synthetic, not panel. AI Visibility is a measurement we make by asking each provider directly — not a measurement of what real users are seeing in their personalized sessions. Personalization, region, and account state can all shift what an individual user sees.
• Provider answers move. The same query can return different answers tomorrow. That's why every metric is timestamped to a scan and trends are scan-over-scan.
• Citations aren't always available. Some providers return structured citations for some queries; some don't. When citations are missing we fall back to brand mention rate and label the headline accordingly.
• AI Agent scores real interactions, not content. Unlike the other four pillars, AI Agent measures live agent task success rather than content quality. It requires enabling per project and is the most direct signal of how usable your site is for AI agents today.