How Faro Index measures AI brand accuracy

What queries the scan sends

Queries are drawn from a library of 800+ buyer-intent patterns organized by industry. Each industry library covers four query types: category definition queries (“what is [industry] software?”), vendor evaluation queries (“what are the best [category] tools?”), specific comparison queries (“how does [brand] compare to [competitor]?”), and feature or capability queries (“which [category] tools include [feature]?”).

Faro Index does not use branded queries (“tell me about [company]”) in the standard scan methodology. Branded queries show how AI describes a company in isolation. Buyer-intent queries show whether AI recommends a company when buyers are actively in the market. That distinction matters for how the results are used.

Each plan tier sends a different query volume and platform set: Free scan sends 10 queries to 2 platforms (ChatGPT, Perplexity). Starter sends 25 queries to the same 2 platforms. Growth sends 50 queries across 3 platforms (adds Gemini). Pro sends the full query library per industry (50+ queries) across all 4 platforms (adds Claude) with 2 repetitions per query. Repetitions detect inconsistency: if AI gives different answers to the same question, that inconsistency is itself a finding.

The same query library is sent to each platform API. ChatGPT (OpenAI) and Claude (Anthropic) return generated prose; Perplexity may include retrieved URLs; Gemini (Google) returns synthesized answers with varying citation behavior. Faro Index stores the full response text, any cited URLs, and extracted company mentions per platform so results are comparable across models even when citation formats differ.

How AI visibility is scored

Visibility measures how often AI mentions a brand when buyers ask about its category. For each query, the model response is analyzed to determine: whether the company name was mentioned, at what prominence (primary recommendation, supporting mention, or passing reference), and whether the citation included a sourced URL.

The Visibility Score (0–100) combines how often the brand appears in buyer-intent answers, how prominently it is positioned (primary recommendation versus passing mention), and how many of the scanned AI platforms included it. A score of 90+ means AI consistently recommends the company. A score of 50 means AI sometimes mentions it. A score of 20 means AI rarely includes it.

How accuracy is measured (Brand Accuracy Rate)

The Brand Accuracy Rate measures the percentage of AI's factual claims that are correct. Measuring this requires two steps: extracting claims and comparing them to ground truth.

Claim extraction:For each AI response that mentions the company, the system uses OpenAI to identify and extract verifiable factual claims. A verifiable claim is a specific assertion that can be checked against the company's website: pricing, founding year, product names, leadership, company size, technology used, geographic presence, category positioning.

Ground-truth comparison: Each extracted claim is scored as Correct (matches the website or a trusted source), Outdated (was once correct, now stale), Incorrect (factually wrong), or Unverifiable (cannot be confirmed from available sources). The Brand Accuracy Rate is the percentage of scored claims marked Correct. Ground truth is the company's own website content from the same crawl used for GEO scoring, not third-party news or social posts.

Benchmark Brand Accuracy Rate assessments use a 15% random sample flagged for manual review, with a stored reproducible seed. The automated-vs-manual agreement rate will be published after the 350-company benchmark audit completes.

How GEO score is calculated

GEO Score (0–100) measures how well a website communicates with AI. The scanner crawls up to 50 pages per domain and blends six weighted signals (configured in Faro Method v1, asserted to sum to 100% at deploy). Question-style FAQ headings are used only for page-type detection in the crawler — they are not a separate GEO dimension. llms.txt presence is not scored (weight 0): tested across 400+ keywords with no measurable citation impact.

• Schema markup coverage: Share of crawled pages that include structured data in JSON-LD or Microdata format. Schema gives AI direct access to entity facts without having to parse prose. Pages with no schema are the primary driver of low GEO scores.
• Content readability: Flesch Reading Ease score measured on the first 500 words after the H1 heading. AI models extract facts more reliably from clearly structured, lower-complexity prose. Content above reading level 60 scores well; dense technical jargon scores poorly regardless of accuracy.
• Answer block presence: Whether the page contains a 30–80 word declarative paragraph placed directly after the H1 that answers a buyer question. Answer blocks are the primary source for AI citations and direct answers. Pages without them are rarely cited verbatim.
• Content depth: Share of pages with 500+ words of substantive body text. Thin pages are rarely treated as authoritative sources by AI models.
• Structured tables and lists: Presence of HTML tables or substantive ordered/unordered lists — formats AI models cite more often in comparative answers.
• Accessibility markup: Heading hierarchy, image alt text, ARIA roles, and transcript/caption signals that help automated extraction.

Content recency applies a small bonus only when JSON-LD dateModified or datePublished confirms content updated in the last 12 months. Missing or unreliable dates never reduce the score.

How content leakage is detected

Content leakage has two meanings in Faro Index: (1) AI uses your positioning or facts without citing your website, and (2) AI misrepresents your brand narrative—wrong category, competitor confusion, or negative framing. After Pillar 1 collects raw AI responses, Pillar 3 analysis identifies the primary narrative AI uses, narrative gaps versus your website copy, competitor associations, negative signals, and uncited content overlap where response text mirrors your site without attribution.

The Leakage Protection score (0–100, higher is better) combines four weighted signals: primary narrative alignment with your site (50%), absence of negative signals (25%), absence of competitor misassociation (15%), and attribution accuracy (10%). A score of 70 means AI protects your narrative inconsistently. A score of 90 means AI consistently represents the brand accurately and favorably. This is not a social listening metric—it measures factual and narrative fidelity in AI answers, not media mention volume.

A worked example: scanning ourselves

When we built Faro Index, the first scan we ran was our own. The tool returned a Brand Accuracy Rate of 20 percent. AI was resolving “Faro” to a 3D measurement company (FARO Technologies), a city in Portugal, and a card-shuffling technique. Only one of five checkable factual claims about the actual company was correct.

The fix was unglamorous. Organization schema with explicit disambiguation on the homepage, FAQPage schema on the pricing page, and a few direct-answer paragraphs written for the exact questions buyers ask. One afternoon of work. Perplexity reflected the change within 48 hours. ChatGPT took longer, which matches the broader pattern: Perplexity updates in about two weeks, ChatGPT in four to six.

Per-claim breakdowns are available for any company in the benchmark, including ours, so the scoring is inspectable rather than a black box.

Benchmark report methodology

The 2026 benchmark report covers 244 companies across 11 industries, four AI platforms — Faro Method v1 (v7.1/method-v1). Industries include ad tech, fintech, SaaS, martech, cybersecurity, healthcare, insurtech, edtech, legal tech, real estate, and e-commerce. Companies were selected using a stratified approach to include market leaders, challengers, and independent vendors within each category.

Name ambiguity (short names, common English words, rebrand lag) was the dominant driver of low Brand Accuracy Rate in the benchmark.Fintech worst at 82.4 percent, one in five facts wrong. GEO gaps matter, but they are a secondary structural factor: companies with weak schema and low GEO scores were more likely to be described from third-party sources, not the primary cause of identity confusion. Homepage schema coverage from the benchmark cohort will be published after the 350-company scan completes with updated detection (C6 hold). Average AI Visibility 96 out of 100 (all four pillars scored 0–100).

Each company was scanned at the Pro tier: 50 queries, 4 platforms, 2 repetitions per query, for up to 400 platform responses per company. Overall average Brand Accuracy Rate across the 244 companies was 88.8 percent — roughly one in nine facts AI stated about the average company was wrong or outdated. The benchmark excludes failed scans and any scan that returned fewer than 30% of expected platform responses. Industry averages use the same scoring methodology as live user scans — no adjustments were applied to benchmark metrics after the fact.

Scans reflect AI platform responses at the time of the scan. AI models update continuously. We recommend weekly monitoring (available on Starter plans and above) to track changes over time. A 10+ point change in Visibility or Brand Accuracy Rate triggers an automatic email alert on all paid plans.