What Semantyra measures
Semantyra is an SEO and AI Search Intelligence platform for semantic SEO, topical authority, technical SEO, competitor intelligence, internal linking, search performance and AI visibility / recommendation intelligence.
Every number Semantyra shows is derived from a real crawl of the site and, where connected, from real Google Search Console and SERP-provider data and real calls to AI answer engines. Nothing is scraped from a third-party authority index and nothing is invented. When a signal cannot be observed, it is labelled "not connected" or "no data", never scored as zero.
The crawl
Semantyra fetches pages starting from the domain root and its sitemaps, respecting robots rules, and extracts the title, headings, body text, links, images and structured data from each one. JavaScript-rendered pages are rendered with a headless browser when needed. The free scan is bounded to 150 pages; paid plans crawl up to the plan limit.
Only pages that returned a 2xx and are indexable become nodes in the analysis. Pages that error, are blocked, or are marked noindex are recorded but excluded from the scores they would distort.
Semantic clustering and topics
Each page gets a semantic embedding. Pages are grouped by meaning into topic clusters using agglomerative clustering with an adaptive cut, so each cluster represents one coherent subject rather than a URL folder. When no embedding provider is available the analysis falls back to lexical vectors and lowers the confidence it reports.
For each cluster, Semantyra estimates the sub-topics a comprehensive source would cover and checks which of those the site has a real page for. That ratio is the cluster completeness.
The Authority Score
The Authority Score is a weighted composite of nine on-site factors, renormalised over only the factors that could actually be observed. It is an on-site estimate of whether a site is built to earn authority, not a prediction of rank and not a third-party metric.
Weights (v1): semantic coverage 16%, cluster completeness 14%, technical health 14%, content depth 12%, entity coverage 12%, internal connectivity 12%, search performance 8%, competitive position 6%, AI visibility 6%.
Each factor carries a data state: observed (measured this scan), not connected (needs a provider or integration), or unavailable (not enough data). Factors that are not observed are excluded from the mean and lower the reported data completeness rather than the score.
Search performance
When Google Search Console is connected, search performance is computed from impression-weighted average position and blended click-through rate over the last 90 days, and brand and non-brand queries are separated. When a SERP provider is configured, ranking coverage across tracked keywords is blended in. Without either, search performance is reported as not connected.
AI visibility
A stable set of non-brand prompts derived from the site’s category, offerings and competitors is run against ChatGPT, Claude and Perplexity with live web search. For each prompt Semantyra records whether the brand is mentioned unprompted, whether a page is cited, which competitors appear, and every cited URL and domain.
A brand token that already appears in the question does not count as a mention. Only unprompted mentions on category questions are counted. Brand-understanding prompts ("what is X") are measured separately as a diagnostic and are not part of the recommendation rate.
Normal monitoring runs a small core prompt set to keep cost proportionate; a full measurement of every prompt runs periodically. The prompt-set hash is always computed on the core set so trends stay comparable.
History and comparison
Every scan writes one metric snapshot. Comparing two scans produces a verdict per metric (improved, worse, unchanged, no data) against a per-metric threshold. A recommendation that stops being detected is marked resolved and its target metric is snapshotted before and after, so the report can show whether fixing it actually moved the number.
What is an estimate
The Authority Score and its factor scores are estimates, labelled as such. Topical authority per cluster is an on-site estimate. Search performance and AI visibility are direct measurements of real data when their sources are connected. Expected-impact figures on recommendations are directional, not guarantees.