Context over strings
A query such as “jaguar” is ambiguous on its own. Page title and URL resolve whether the intent is automotive, wildlife, or something else entirely.
RESEARCH NOTE / SYSTEM ARCHITECTURE
From ranking footprint to semantic signal.
Relevant evaluates what a domain consistently ranks for, not what its homepage claims to be about. The result is a defensible measure of backlink prospect quality, topical alignment, and content risk.
SYSTEM AND METHOD FOR
SEMANTIC RELEVANCE SCORING OF LINK PROSPECTS
DRAWING SHEET 1/1
REF. REL-2026-01
REF. REL-2026-01
[ FIG. 1A / SIGNAL PIPELINE ]
100
Organic search corpus
We begin with the domain’s live ranking footprint: queries, destination URLs, titles, search demand, position, and estimated traffic.
102
Representative page sampling
The system groups signals by ranking page, prioritizes the pages with the strongest organic footprint, and selects representative queries for analysis.
104
Model latent representation
Keyword, URL, and title are interpreted together. The language model forms a contextual representation of topic and intent instead of relying on literal term overlap.
106
Semantic proximity analysis
Each observed intent is evaluated against a model-built boundary of the target niche: core topics, adjacent topics, ambiguous territory, and clear outliers.
108
Dual-signal calibration
Independent relevance and content-risk signals are normalized and aggregated into decision-ready scores, with calibrated warning and failure bands.
110
Explainable evidence set
You get the score, verdict, and the keyword-level evidence behind it, so a reviewer can verify why a domain passed, warned, or failed.
[ FIG. 1B / SEMANTIC PROXIMITY MAP ]
d₂d₁A / target nicheS₁ / strong alignmentS₂ / marginal signalS₃ / topical outlierθsemantic distance
xᵢ = { queryᵢ, pathᵢ, titleᵢ }rᵢ, gᵢ = classify(xᵢ | niche) ∈ [0,1]²R = 100 · |{i : rᵢ ≥ τᵣ}| / nG = calibrate(|{i : gᵢ ≥ τg}|)Conceptual projection. Production scoring uses model-based classification and calibrated evidence aggregation rather than a single geometric distance.
FIG. 1 / SEMANTIC RELEVANCE MEASUREMENT SYSTEM
METHOD NOTES / REV 2.1
A scoring model with inspectable inputs.
This is not another backlink index. Relevant is the quality layer after prospect discovery. Import candidate domains from your outreach or backlink analysis workflow, then test whether their organic footprint actually belongs to your niche.
STAGEOBSERVED STATEINVARIANT
01 / Corpus
qᵢ, uᵢ, tᵢ, etvᵢ, rankᵢLive Google organic rankings, currently scoped to US English.
02 / Page priority
P(u) = max(etvᵤ) + λ · residual(etvᵤ)Pages are ranked by concentrated and supporting traffic signals.
03 / Context unit
xᵢ = tuple(query, path, title)One representative search intent is selected from each priority page.
04 / Classifier
yᵢ = (relevanceᵢ, riskᵢ)Both dimensions are bounded independently on the interval [0, 1].
05 / Domain signal
S = aggregate(y₁ ... yₙ | τ)Calibrated decision bands produce pass, warning, and failure states.
SOURCE live organic SERP dataUNIT ranking pageCONTEXT query + path + titleFAILURE isolated per domain
WHY IT WORKSWords match.
Words match.
Intent clusters.
Traditional filters ask whether a keyword contains the right phrase. Relevant asks whether the underlying search intent belongs inside the topical neighborhood of your niche.
Footprint over homepage copy
A polished homepage is easy to manipulate. Sustained organic rankings across representative pages are a harder, more useful signal of what a domain actually covers.
Two independent questions
A domain can be topically relevant and still carry risky content. Relevance and sensitive-content exposure are evaluated separately, then presented together.
Evidence over oracle scores
The final number is never the whole answer. Relevant returns the classified queries that shaped the verdict so humans can audit the decision.
THE BLACK BOX, WITH WINDOWSTechnical enough to trust.
Technical enough to trust.
Simple enough to use.
The production pipeline combines live organic-search observations, contextual language-model reasoning, and proprietary calibration. Exact prompts, thresholds, weighting logic, and provider orchestration remain part of the scoring system, not the marketing page.
Analyze a domain