Executive abstract & research goals
QLABS Rank Algorithm™ specifies two separate weighted research models: GEO Intelligence for heuristic generative-search readiness and AISEO™ PageRank for independently evidenced SEO authority and readiness signals.
The aim is a reproducible diagnostic with visible data gaps. Neither model reports an actual search ranking, measured AI citation frequency, or a probability of being mentioned. Weights are research assumptions that require independent validation.
Weighted coverage is the weight share of dimensions for which eligible input evidence exists, expressed as a percentage. It measures data availability, not statistical certainty.
Author: SEOSTUDIO Research & Innovation. Publisher: SEOSTUDIO, Germany. This English edition shares the mathematical model and version with the German paper.
An independent model, not Google's PageRank
The historical Google PageRank ranks pages from a directed link graph using a random-surfer model. AISEO™ PageRank instead combines six specified, evidenced dimensions into a weighted sum. There is no affiliation with Google and no equivalence to the SISTRIX Visibility Index.
| Aspect | Historical PageRank | AISEO™ PageRank V1.0 |
|---|---|---|
| Basis | Link graph and random surfer | Six weighted dimensions |
| Calculation | Iterative link analysis | Evidence-gated weighted sum |
| Scale | Not reconstructed here | 0–10 research scale |
| Source | Original Google research | QLABS / SEOSTUDIO |
Mathematical foundation
Every input xi is normalized to 0–1. Fixed, nonnegative weights wi sum to one. A public aggregate is released only after the evidence decision. An indicator mi equals one for eligible evidence and zero for unavailable or unverified evidence; a missing input itself is never assigned a score of zero.
Weighted sum S equals the sum of weight times normalized signal; the weights sum to one.
GEO score G equals one hundred times the complete weighted GEO sum, rounded to one decimal.
PageRank P equals ten times the complete weighted rank sum, rounded to one decimal.
The Python core accepts dimension scores from 0–100; it divides each by 100 before weighting. The companion calculation lab reproduces Python's one-decimal binary floating-point rounding and fixed accumulation order.
QLABS GEO Intelligence · seven dimensions
The same seven fixed weights are used by the integrated GEO core. They describe a readiness rubric, not established search-engine ranking factors.
AI Citability
Extractable answers and clear content structure.
Brand & Entity
Documented identity and entity signals.
Content & E-E-A-T
Expertise, experience, and source quality.
Technical SEO
Technical accessibility and page properties.
Schema Intelligence
Structured data beyond mere syntax.
AI Platform Readiness
Accessibility for AI search systems under review.
LONG TAIL GEO™
Questions, answers, and follow-up paths.
Citability carries 20%; brand and entity, content and E-E-A-T, technical SEO, and LONG TAIL GEO™ each carry 15%; schema and platform readiness each carry 10%. These choices have not been empirically calibrated.
What does the current HTML core actually observe?
It computes four simple heuristics: technical signals (title 30, description 20, exactly one H1 20, HTTPS 15, at least three headings 15); readable JSON-LD syntax (100 or 0, without semantic validation); question headings (25 each up to 100); and a citability heuristic using word count, headings, and questions. Brand and entity, content and E-E-A-T, and platform readiness remain unknown. The observed four dimensions yield only 60% weighted coverage and no public overall GEO score.
More headings or valid JSON-LD do not prove better answers or real citations. The core's score describes the sampled home page, while extra crawled pages are separate evidence.
AISEO™ PageRank · six dimensions
This executable research formula does not itself collect backlink exports, Search Console data, or independent authority evidence. Its inputs need documented sources and normalization before a real domain value can be released.
| Dimension | Weight | Evidence required |
|---|---|---|
| Link AuthorityRequired signal | 25% | A documented backlink dataset with source and date. |
| Organic VisibilityRequired signal | 20% | Authorized Search Console data or a defined visibility dataset. |
| Technical Quality | 15% | HTTP crawl and documented technical measurements. |
| Content / Topic Authority | 15% | A defined rubric for expertise and topical coverage. |
| Entity / Trust | 15% | Verifiable company, author, and source information. |
| Information Architecture | 10% | Internal links, crawl depth, and topical relationships. |
The 25% link-authority and 20% organic-visibility weights emphasize external evidence. Technical quality, content and topic authority, and entity and trust each carry 15%; information architecture carries 10%. Link authority and organic visibility are critical inputs and cannot be compensated by other dimensions. The weights are proposals, not empirically validated optima.
Normalizing input signals
For nonnegative, highly skewed counts, the research model proposes logarithmic normalization. A positive reference value must be justified and versioned before comparing domains.
Normalized N of x equals the smaller of one and natural log of one plus x divided by natural log of one plus the reference value.
Link data require a dated authorized export or documented API. Organic visibility requires a specified source, geography, device, period, and metric. Technical and content checks need time-stamped snapshots, measurement definitions, and a review rubric. The current crawler does not supply independent backlink or search data; no universal reference dataset has been defined. Negative, infinite, and invalid inputs are rejected.
Evidence classes & weighted coverage
Observation
A feature found directly in a documented page snapshot.
Verified measurement
Independent data with provenance, date, and an appropriate definition.
Heuristic
A rule-based estimate, marked with low confidence in the HTML core.
Hypothesis or unknown
Generated search paths are hypotheses; missing measurements remain unknown.
Weighted coverage in percent equals one hundred times the sum of weight times eligibility indicator divided by the sum of the weights.
In the GEO HTML core, coverage means a dimension heuristic can be calculated. PageRank applies the stricter rule of explicitly verified dimensions. Coverage is not the reliability of any individual source.
Confidence, missing data & release
Current confidence labels are qualitative: high, medium, low, and none. They are not calibrated probability percentages. Coverage is reported separately.
No aggregate or partial score is released.
Only an explicitly labeled internal partial diagnostic may be calculated when model-specific conditions hold.
The threshold alone is insufficient; all dimensions and critical requirements must be met.
Conservative V1.0 interpretation: A complete sum needs all factors. At 85–99.9% coverage, a missing dimension still blocks a public full score. In practice this implementation requires 100% eligible dimension coverage and critical evidence for release. Missing factors are never filled with zero or extrapolated.
Internal partial diagnostic only: divide the weighted sum of available factors by the covered weight share c, provided c is greater than zero.
GEO may calculate an internal partial diagnostic at 70% coverage; the public full score additionally requires all seven eligible dimensions. PageRank's internal diagnostic additionally requires both verified critical inputs. Its full release requires all six verified dimensions. The current HTML core does not meet the GEO release rules, and independent rank data are not connected.
LONG TAIL GEO™ and query fan-out
A conversation intent graph connects discovery, problem, comparison, commercial, local, and follow-up questions. Query fan-out and multi-path discovery suggest related subquestions; they do not claim to reveal an AI provider's internal queries.
Observed question headings from HTML are recorded separately from synthetic paths. Clear answers, entity information, source evidence, and contextual links may help people navigate a topic. Their effect on retrieval or citations requires separate measurement.
Limits, bias & reproducibility
- Research assumptions: Weights and thresholds are proposals, with no claimed validation against rankings or AI citations.
- Selection bias: Industry, language, domain age, site size, and dataset provider may shift signal distributions.
- Measurement gaps: Static HTML and a bounded crawl cannot see all JavaScript content, protected pages, actual AI citations, or an entire link graph.
- Gaming: More headings or JSON-LD can raise a simple heuristic without improving answer quality. Human review remains necessary.
- Reproducibility: Record domain, UTC time, source, snapshot hash, raw inputs, reference values, evidence status, normalization version, and algorithm version together.
A case report or correlation does not prove causation. A missing detection of an AI-native technology must not be penalized as a failed measure.
Synthetic calculation lab
Every number below is invented for demonstration and belongs to no real domain. Change values and the simulated evidence checkboxes; weights remain fixed. The lab uses the same model specification and release decision as the German edition.
Documented backlink dataset and measurement date.
Authorized Search Console data or a defined visibility dataset.
HTTP crawl, index access, and documented technical measurements.
A documented rubric for expertise and topical coverage.
Verifiable company, author, and source information.
Internal links, crawl depth, and topical relationships.
Synthetic PageRank example: weighted sum 74.75 from inputs on a 0–100 scale, divided by ten and rounded to one decimal equals 7.5 on a 0–10 scale.
If information architecture alone is missing, coverage is 90%. Despite exceeding 85%, the full PageRank remains blocked. If both required signals are present, only an internal partial diagnostic may be shown. If link authority is missing, that diagnostic is blocked too.
Versioning and changelog
The V1.8 package adds a separate rank module. The four original HTML GEO heuristics retain their rules; the aggregator has explicit 0–1 normalization and a separate internal partial-score field. The package number does not imply a new GEO scoring formula.
V1.0 research model and engine alignment 1.0-alignment.1
Six rank dimensions, seven GEO dimensions, fixed weights, evidence tiers, Python-compatible rounding, and a synthetic lab. No empirical validation and no real PageRank without verified sources.
Calculation, normalization, or release-policy changes require a versioned specification and comparison tests. Download the shared machine-readable manifest ↓
Methodology questions
What does the QLABS GEO Score measure?
It describes readiness under a defined seven-dimension research rubric on a 0–100 scale. It does not measure actual AI citations or the probability of a brand mention.
Is AISEO™ PageRank Google's PageRank?
No. It is an independent QLABS research model using six weighted dimensions on a 0–10 scale. It is not affiliated with Google.
Can PageRank be derived from the GEO Score?
No. The dimensions and evidence requirements differ. In particular, the PageRank model needs independently verified link authority and organic visibility. Dividing GEO by ten would be invalid.
Why is there no overall score when data is missing?
Unknown is not zero. A complete score requires every eligible dimension and the critical evidence rules. A partial internal diagnostic never substitutes for a public complete score.
How do coverage and confidence differ?
Weighted coverage measures the share of model weight with eligible inputs. Confidence describes the strength of a particular claim. The current core uses qualitative high, medium, low, and none labels, not calibrated probabilities.
Are query fan-out questions measured searches?
No. Generated subquestions and discovery paths are editorial hypotheses. Question headings observed in HTML also do not prove search demand.
Does a higher score guarantee rankings or AI citations?
No. The weights are research assumptions without claimed empirical validation. Actual search and AI answers depend on systems, competitors, sources, and the query context.
What is needed to reproduce a calculation?
Record the engine and whitepaper versions, input values, provenance, measurement time, snapshot, normalization references, missing-data status, evidence class, and rounding rule. A newly retrieved page may already have changed.
From methods to application
GEO Check (German) ↗
Review the interface and its current live-analysis availability. The public domain analysis must not fabricate results while the backend is unavailable.
CONSULTINGGEO Analysis & Consulting ↗
Four defined analysis packages and a free inquiry through the existing SEOSTUDIO delivery process.
Sources and model basis
- SEOSTUDIO Research & Innovation: QLABS Rank Algorithm™ Whitepaper V1.0 and provided Core V1.8
core/pagerank.py, October 30, 2026. - QLABS GEO Core V1.6:
core/scoring.py,core/audit.py, andcore/evidence.py. - Page, Brin, Motwani & Winograd, The PageRank Citation Ranking ↗, for the historical distinction.