Methodology

Lumos measures brand visibility inside AI-powered search engines — ChatGPT, Gemini, Claude, and Perplexity — by running a recurring set of category prompts daily and scoring the responses on four pillars: Visibility, Citability, Sentiment, and Position. This page documents the full scoring rubric, the prompt categories, the metrics surfaced on every dashboard, and the technical-audit layer that explains why a page might be invisible to AI even when it ranks on Google.

How Lumos measures and scores your brand's visibility in AI-powered search engines.

Visibility Score

The visibility score measures the percentage of AI-generated responses that mention your brand. It's the core metric for understanding how often AI search engines recommend you to users.

Total

Mentions across all prompts — branded and non-branded combined. Gives a high-level overview of your overall AI presence.

Branded

Mentions from prompts that include your company name. Typically very high (often 100%) since LLMs echo entities from the query.

Non-Branded (Organic)

Mentions from prompts that don't contain your company name. The most meaningful metric — measures whether AI recommends you unprompted.

visibility_score = (mentions / total_evaluations) × 100

The Echo Effect

When a user asks "What do you think about [Company X]?", the model will almost always mention that company — it's echoing back the entity from the query. This inflates the visibility score artificially. By separating branded and non-branded prompts, we distinguish between two fundamentally different types of mentions:

Prompted Mentions

AI mentions you because you were in the query. Less meaningful — nearly any brand achieves high scores here.

Organic Mentions

AI mentions you unprompted, based on training data and web search results. Highly meaningful — this is true AI visibility.

A prompt is classified as "branded" if it contains the company name or any known alias (case-insensitive match).

Scoring Rubric: The Four Pillars

Every response Lumos collects feeds four orthogonal pillars. The Visibility Score on every dashboard is a weighted sum of these — chosen so that brand mention rate dominates, but a brand that is mentioned without citation, with bad sentiment, or buried at position #8 cannot coast on raw mention count alone.

Pillar	Weight	What it measures
Visibility	40%	Did the brand get mentioned at all in the AI response? Averaged across the prompt set, then split into branded and non-branded buckets. Non-branded is the meaningful signal.
Citability	25%	When the brand was mentioned, did the AI cite it with a source URL (the brand domain or a third-party that names the brand)? Captures whether the answer is verifiable and click-throughable.
Sentiment	20%	Are the mentions positive, neutral, or negative? Scored from −1 (uniformly negative) to +1 (uniformly positive) per response, then averaged across the prompt set.
Position	15%	When the AI answer ranks multiple brands, where does this brand land? #1 = full credit, decaying linearly to 0 at the bottom of any list longer than 5. Brands that appear but aren't ranked get a neutral mid-position credit.
Total	100%	Aggregated to a single 0–100 score, refreshed daily.

Worked Example

Consider a B2B SaaS brand running 80 category prompts across ChatGPT and Gemini. Over a 7-day window the brand is mentioned in 36 of 80 non-branded prompts (45%), cited with a source URL in 22 of those 36 mentions (61%), achieves an average sentiment of +0.52, and lands at an average position of 2.8 across answers that ranked multiple competitors. Plugging those into the weights:

Visibility   45 × 0.40  = 18.0
Citability   61 × 0.25  = 15.3
Sentiment   76 × 0.20   = 15.2   // (0.52 + 1) / 2 × 100 = 76
Position    73 × 0.15   = 11.0   // ((5 − 2.8) / (5 − 1)) × 100 = 55, clamped against fallback
                          ─────
Visibility Score          59.5

A score of 59.5 places the brand in the “Strong” tier (50–74). The immediate lift opportunity is Visibility: doubling non-branded mention rate from 45% to 80% would add about 14 points to the headline score — more than any other single pillar can deliver from its current value.

Methodology last reviewed: May 18, 2026. Pillar weights were last recalibrated in March 2026 after a cohort study of 40 Lumos accounts comparing pillar contributions to self-reported buyer-mention outcomes.

Methodology FAQ

How does Lumos score AI visibility?

Lumos computes a Visibility Score per brand by running 70+ category-specific prompts daily against ChatGPT, Gemini, Claude, and Perplexity. Each response is evaluated for four pillars — Visibility (was the brand mentioned at all), Citability (was the brand cited with a source link), Sentiment (was the mention positive, neutral, or negative), and Position (where in the ranked list the brand appeared). The pillars are weighted Visibility 40%, Citability 25%, Sentiment 20%, Position 15%, summed to a 0–100 score.

Which AI engines does Lumos query?

Lumos runs prompts daily against ChatGPT (OpenAI), Gemini (Google), Claude (Anthropic), and Perplexity. Each engine is scored independently, then aggregated into a multi-engine Visibility Score. Customers can drill into per-engine performance to see, for example, whether their brand is strong in ChatGPT but invisible in Gemini for the same category prompts.

How often is the data refreshed?

Lumos refreshes its prompt evaluations on a daily cadence for every customer account. Technical-audit data (bot access, schema, page citability) is refreshed weekly. Customers with large prompt sets can opt into 6-hour or hourly cadences for category-critical prompts.

What is GEO (Generative Engine Optimization)?

Generative Engine Optimization is the discipline of making your brand and content discoverable, citable, and accurately described inside AI-generated answers from ChatGPT, Gemini, Claude, and Perplexity. It overlaps with classic SEO (crawlability, structured data, content quality) but adds new surfaces: how a brand is named, what it is associated with, and which sources AI engines cite when discussing it.

How does Lumos define a "branded" versus "non-branded" prompt?

A branded prompt names the brand explicitly ("What does Lumos do?"). A non-branded prompt asks about the category without naming any brand ("What is the best GEO platform for LATAM?"). Branded prompts measure how AI engines describe a brand once asked; non-branded prompts measure share-of-voice — whether the brand surfaces at all when the user is shopping the category. Lumos splits visibility into branded and non-branded buckets so customers can see both.

Why do my scores change day-to-day?

AI engines are stochastic: the same prompt run twice can yield slightly different answers. Lumos averages across 70+ prompts per category and uses 7-day rolling windows for headline scores, which dampens noise. Larger week-over-week swings usually trace to a real underlying cause: a competitor launched a piece of content the AI now cites, a model was retrained, or your own site changed (robots.txt, schema, content).