Rankio calculates the Overall Score (0–100) as a two-tier composite: Content Quality (80%) blended with GEO Readiness (20%). Content Quality is computed from 7 weighted metrics per AI response: Presence (25%), Citation quality (20%), Position (15%), Recommendation strength (15%), Sentiment (10%), Cross-model consistency (10%), and Frequency (5%). GEO Readiness is the automated GEO Content Audit score that checks 10 structural elements AI models need to cite a page. The model is calibrated against real referral traffic from AI search.
Rankio sends the same prompts to multiple AI models, parses each response for brand citations, scores 7 content quality dimensions, then blends with a GEO Readiness score (10 structural checks) into a single 0–100 composite. Weights are calibrated quarterly against real traffic outcomes. Every data point is auditable.
How Rankio calculates the Visibility Score
The Visibility Score answers one question: how visible is your brand when people ask AI models about your market? It is computed in five steps:
Prompt generation
You provide a brand name, URL, or topic. Rankio generates (or you define) a prompt set that covers four intent types: discovery, comparison, branded, and intent-based queries. A minimum of 30 prompts is recommended for statistical reliability.
Multi-model querying
Every prompt is sent simultaneously to ChatGPT (OpenAI), Gemini (Google), and Perplexity. Responses are captured with timestamp, model version, and full raw text. The same prompt set is used across all models to ensure comparability.
Response parsing
Each response is analyzed for brand mentions (exact match, fuzzy match, entity recognition, URL detection), citation position, sentiment, and recommendation language. This produces a per-prompt, per-model metric vector.
Metric scoring
Raw detections are converted into normalised scores (0–1) for each of the 7 metric categories (see Scoring Model below). Each score is multiplied by its category weight.
Aggregation
Weighted scores are summed per prompt–model pair, then averaged across all prompts and models to produce the final Visibility Score (0–100). The same pipeline, applied to competitors, produces your AI Share of Voice.
The scoring model: Content Quality + GEO Readiness
Rankio uses a two-tier composite scoring model. The first tier — Content Quality (80% of the final score) — is a weighted linear model of 7 metrics. The second tier — GEO Readiness (20%) — is the GEO Content Audit score measuring how well the page is structured for AI extraction.
Tier 1: Content Quality (80%)
For each prompt–model pair, every metric category produces a normalised score between 0 and 1. These scores are multiplied by their weight, summed, and scaled to 0–100.
| Metric | Weight | What it measures | How it is scored (0→1) |
|---|---|---|---|
| Presence | 25 % | Is the brand mentioned at all in the AI response? | Binary: 1.0 if mentioned, 0.0 if not. Averaged across the full prompt set. |
| Citation quality | 20 % | How is the brand referenced? | URL citation = 1.0 · Direct name mention = 0.6 · Contextual/implied = 0.3 |
| Position | 15 % | Where in the response does the brand appear? | 1st position = 1.0 · 2nd = 0.7 · 3rd+ = 0.4 |
| Recommendation | 15 % | Is the brand actively endorsed by the AI? | "We recommend" / "top choice" = 1.0 · "is an option" = 0.4 · No rec. = 0.0 |
| Sentiment | 10 % | What is the tone of the mention? | Positive = 1.0 · Neutral = 0.5 · Negative = 0.1 |
| Consistency | 10 % | Does the brand appear across multiple models and prompt types? | Ratio of (model × prompt) pairs where brand is cited vs. total pairs |
| Frequency | 5 % | How many times is the brand mentioned in a single response? | 1 mention = 0.5 · 2 = 0.7 · 3+ = 1.0 |
Score formula
The final Overall Score is a two-tier composite that blends Content Quality (the 7 metrics above) with GEO Readiness (the GEO Content Audit score):
Overall Score = Content Quality × 0.80 + GEO Readiness × 0.20
Where Content Quality = Σ (metric_i × weight_i) × 100 for the 7 metrics above.
And GEO Readiness = the overall GEO score (0–100) from the automated GEO Content Audit that checks 10 structural elements (Direct Answer, tables, FAQ, JSON-LD, headings, lists, internal links, meta description, entity clarity, TL;DR).
| Component | Weight | Source |
|---|---|---|
| Content Quality | 80 % | 7-metric weighted score (Presence, Citation quality, Position, Recommendation, Sentiment, Consistency, Frequency) |
| GEO Readiness | 20 % | 10-element GEO Content Audit score (Direct Answer, TL;DR, tables, FAQ, headings, lists, JSON-LD, internal links, meta description, entity clarity) |
Score example
| Metric | Raw score | Weight | Contribution |
|---|---|---|---|
| Presence | 1.0 | × 0.25 | 0.25 |
| Citation quality | 0.6 (name mention) | × 0.20 | 0.12 |
| Position | 0.7 (2nd position) | × 0.15 | 0.105 |
| Recommendation | 0.4 (listed, not endorsed) | × 0.15 | 0.06 |
| Sentiment | 1.0 (positive) | × 0.10 | 0.10 |
| Consistency | 0.67 (2 of 3 models) | × 0.10 | 0.067 |
| Frequency | 0.7 (2 mentions) | × 0.05 | 0.035 |
| Content Quality subtotal | 0.737 → 74/100 | ||
Content Quality = 74 · GEO Readiness = 62 (page has tables and headings but lacks Direct Answer and FAQ JSON-LD)
Overall Score = 74 × 0.80 + 62 × 0.20 = 59.2 + 12.4 = 72/100
Weight calibration
Weights are re-calibrated quarterly using a correlation analysis between Visibility Score deltas and observed referral traffic and conversion outcomes from consenting Rankio customers. Key principles:
- Calibration is always prospective — historical scores are never retroactively altered
- Every calibration change is documented with a version number and effective date
- A minimum of 500 prompt–model samples is required before a weight change is applied
- The current weight set (v2.0) has been effective since January 2026
Known limitations
No measurement system is perfect. We document our constraints so you can interpret results correctly.
| Limitation | Impact | How Rankio mitigates |
|---|---|---|
| LLM non-determinism | The same prompt can produce different answers on different runs | Multi-sample averaging; score confidence intervals on repeated runs |
| Model version changes | Provider updates (e.g. GPT-4 → GPT-4o) shift response patterns | Model version is logged with every analysis; version-tagged trend charts |
| Parametric vs. retrieval knowledge | Cannot always tell if a citation comes from training data or live search | Perplexity always uses retrieval; GPT/Gemini are flagged as mixed-source |
| Prompt set coverage | Score reliability depends on the breadth of prompts tested | Minimum 30 prompts recommended; coverage warnings when set is too small |
| Geographic & language variance | AI responses vary by inferred locale | Default locale is clearly documented; multi-locale support on roadmap |
| No direct AI traffic measurement | Rankio measures visibility (citations), not downstream clicks from AI answers | Correlations with referral traffic are used for calibration; UTM integration planned |
| Correlation ≠ causation | Score changes may reflect competitor movements or model updates, not just your actions | Always track competitors in parallel via SOV benchmarking |
What Rankio does not measure
- Click-through from AI answers — AI interfaces don't expose referral data consistently. Rankio measures citation presence, not downstream clicks.
- Private/enterprise AI deployments — Custom GPTs or internal model deployments are not covered.
- Image/video visibility — Current analysis is text-only. Multimodal analysis is on the roadmap.
Why we publish our methodology
AI visibility is a new discipline. There is no Google Search Console for LLMs. When measurement is new, trust is earned through transparency — not through black-box scores.
Every Rankio analysis includes:
- The full raw AI response for every prompt × model pair
- Per-metric breakdown showing how each score was derived
- Model version, timestamp, and prompt text
- Exportable data for independent analysis
We believe that if you can't audit a score, you shouldn't trust it. This principle drives every design decision in Rankio.
| Concept | Definition | Why it matters |
|---|---|---|
| Overall Score | Two-tier composite (0–100): Content Quality (80%) + GEO Readiness (20%) | Single reliable number that captures both citation strength and structural readiness |
| Content Quality | Weighted linear model of 7 metrics (Presence, Citation quality, Position, etc.) | Measures how AI models actually respond to your brand |
| GEO Readiness | 10-element audit score from the GEO Content Audit | Measures how well your page is structured for AI extraction |
| Non-determinism | LLMs can produce different outputs for the same input | Explains why scores fluctuate slightly between runs |
| Prospective calibration | Weight changes apply going forward, never retroactively | Historical scores remain comparable over time |
| Full auditability | Every score traces to a raw AI response you can read | You never have to trust a black box |
Frequently asked questions
See the methodology in action
Run your first analysis and explore the raw data, per-metric breakdown, and competitive benchmarks behind your Visibility Score.