Methodology
How SurfaceGX measures, scores, and diagnoses AI brand visibility. This page explains the mechanics behind every score the platform produces — from how queries are constructed and sent to AI engines, to how individual pages are evaluated for AI readability.
Query construction
SurfaceGX generates structured prompts for each AI engine based on your brand profile. Prompts are designed to elicit the kind of responses a real buyer would receive — not developer-style queries — so that results reflect actual brand visibility in consumer-facing AI interactions.
Each query family targets a different aspect of brand presence:
- Category queries — "What are the best tools for [your category]?" — test whether your brand appears in open-ended AI recommendations.
- Brand-direct queries — "What does [Brand] do?" and "Tell me about [Brand]." — reveal what AI engines say about you specifically.
- Comparison queries — "How does [Brand] compare to [Competitor]?" — test how AI frames your brand relative to market alternatives.
- Buyer-stage queries — awareness, consideration, evaluation, and decision prompts — used by the Buyer Journey module to map full-funnel AI behavior.
All queries are sent in parallel to the configured AI providers. Responses are captured verbatim and processed by the scoring pipeline. No query results are shared with or visible to other SurfaceGX customers.
The four audit scoring dimensions
The core GEO Audit scores your brand across four dimensions. Each dimension is scored independently on a 0–100 scale, then combined into a composite score using weights you configure in your brand profile.
| Dimension | What it measures | Default weight |
|---|---|---|
| AI Presence | Whether and how prominently your brand appears in AI-generated category responses. Measures mention rate, position in the response, and depth of description. | 30% |
| Hallucination Risk | Whether AI engines produce factually incorrect claims about your brand. Scored by comparing AI output against your verified brand facts using a secondary LLM review pass. | 25% |
| Narrative Alignment | How closely the language AI uses to describe your brand matches your official positioning. Scored via semantic similarity between AI response text and your approved messaging. | 25% |
| Algorithm Readability | Whether your website is technically structured for AI crawler extraction. Derived from the Algorithm Score module's six-factor rubric. | 20% |
Weights are adjustable in your brand profile settings. If hallucination risk is a priority concern — common in regulated industries — you can increase its weight accordingly. The composite score updates automatically when weights change.
Algorithm Score: the six readability factors
The Algorithm Score module evaluates your website against six technical factors that determine how reliably AI crawlers can read, extract, and cite your pages. Each factor is scored separately with specific findings.
Semantic density
Measures what fraction of each page's text contains substantive, brand-relevant information versus navigation, legal boilerplate, ads, and UI chrome. AI systems derive value from pages where the signal-to-noise ratio is high. Pages with dense semantic content are more likely to be cited in AI-generated responses.
Common issues: Long header/footer navigation blocks, repeated boilerplate text, excessive legal disclaimers, and modal overlays that obscure primary content.
Structural clarity
Evaluates heading hierarchy (H1 → H2 → H3), presence and accuracy of structured data markup (Schema.org), and logical sectioning of content. AI crawlers use heading structure to understand the topical organization of a page and extract answers to specific questions.
Common issues: Missing H1, skipped heading levels, headings used for styling rather than semantics, absent or misconfigured Schema.org markup.
Entity authority signals
Checks whether the page establishes credible authorship and organizational identity. AI systems assign higher trust to content that attributes authorship to named, credentialed people and links to external authoritative references. Unattributed content is treated as lower-confidence by most LLMs.
Common issues: Bylines absent or lacking credentials, no organizational affiliation on About or Team pages, missing links to external citations or references.
Crawl accessibility
Verifies that your pages are accessible to AI crawlers — not blocked by robots.txt rules, paywalls, heavy JavaScript rendering dependencies, or authentication walls that prevent AI systems from reading the page content.
Common issues: Over-broad robots.txt disallow rules, JavaScript-rendered content with no server-side fallback, pages that require login to access informational content.
Canonical consistency
Checks for duplicate content issues that confuse AI systems about which version of a page to treat as authoritative. Includes canonical tag verification, URL parameter canonicalization, and detection of near-duplicate pages that may dilute authority signals.
Common issues: Missing or self-referencing canonical tags, multiple URLs serving identical or near-identical content, print versions or parameterized URLs without canonical declarations.
Freshness and update signals
Evaluates whether pages include clear publication and modification dates, whether content is demonstrably current, and whether the site signals regular updates to AI crawlers via sitemaps and last-modified headers. AI systems weight recency when deciding how much trust to place in a source.
Common issues: No publication dates visible on content, sitemap not submitted or not updated, pages with stale or absent last-modified metadata.
Hallucination detection
Hallucination detection in SurfaceGX uses a two-pass process. First, raw AI responses are collected across all queried providers. Second, a dedicated LLM reviewer — operating with your verified brand fact sheet — reads each response and identifies specific claims that are factually incorrect, fabricated, or materially distorted.
The reviewer classifies each hallucination by severity:
- Critical — Factual errors that could cause direct harm: wrong pricing, nonexistent products, false regulatory claims, misattributed quotes or endorsements.
- Significant — Distortions that materially misrepresent your brand: wrong founding date, incorrect leadership, inaccurate market position.
- Minor — Vague or slightly inaccurate language that diverges from your positioning but does not assert a false fact.
Each hallucination finding is returned with the source AI provider, the verbatim AI excerpt, and the specific fact that contradicts it. The Hallucination Risk score is derived from the count, severity, and recurrence rate of findings across providers.
To improve hallucination detection accuracy, keep your brand fact sheet in the setup profile current. The reviewer uses only the facts you have explicitly verified — if your product lineup or pricing has changed, update your profile before re-running the audit.
Narrative Alignment scoring
Narrative Alignment is scored by comparing what AI engines say about your brand against what you have declared as your official positioning. The comparison uses semantic similarity — meaning it measures conceptual alignment, not keyword matching.
The scoring captures three alignment signals:
- Positioning accuracy — Does the AI description match your declared value proposition and differentiation?
- Tone consistency — Does the register and language AI uses (e.g., technical vs. approachable, enterprise vs. SMB) match your brand voice?
- Competitor framing — When AI compares you to competitors, is the framing fair and consistent with how you position against those alternatives?
Scores are computed per AI provider, then aggregated into a single Narrative Alignment score. Provider-level breakdowns are available in the results view, so you can see which engine is most or least aligned with your messaging.
Discovery Scoring Engine
The Discovery Scoring Engine evaluates individual pages for AI visibility. Each page is assigned one of three status tiers:
| Status | What it means |
|---|---|
| AI-Visible | The page has sufficient semantic density, structural clarity, and entity authority for AI crawlers to reliably read and cite it. Qualifies for inclusion in llms.txt. |
| Partially Visible | The page has some AI-readable content but is impaired by one or more structural, authority, or crawl accessibility issues. Included in llms.txt with a [Fix Required] flag. |
| Not Visible | The page does not meet the minimum threshold for AI citation. Excluded from manifest files until issues are resolved. |
For each page that is Partially Visible or Not Visible, the engine produces a Fix Card: a specific, actionable description of what is blocking the page, which role should fix it (Content Writer or Developer), and an estimated effort level.
PESO source mapping
The Source Intel module maps the sources AI engines draw from when they discuss your brand against the PESO model: Paid, Earned, Shared, and Owned media.
- Paid — Sponsored placements, advertorial coverage, and review platforms where your brand has paid for visibility.
- Earned — Press coverage, analyst reports, and third-party editorial mentions your brand has not paid for.
- Shared — Social media, community forums (Reddit, Quora), user-generated content, and peer review sites (G2, Capterra).
- Owned — Your website, blog, documentation, and any content you publish and control directly.
Each source identified in AI responses is classified into one of these categories and assessed for authority level. The module then surfaces which PESO categories are underrepresented in your current citation profile — a gap between what AI cites and what a strong authority profile looks like for your industry.