SpyderBot Documentation
Citations and sources
A citation is an explicit reference supporting an observed answer. A retrieved or inspected page is not automatically a cited page. A citation does not prove that a model trained on a page, that the page caused an answer, or that SpyderBot could read the entire source.
Answer evidence, source access and source portrayal are separate. Evidence links are available only where the report, plan and your access permissions allow them.
On this page
Citation Presence · Owned Citation Presence · Citation Domain Presence · Domain Citation Allocation Share · Citations, Unique URLs and Unique Domains · Source Type Composition · Source Access Status · Source Brand Portrayal · Citation Set Agreement · Observed Source Prompt Reach and Domain Coverage · Share of Cited Domains by Owner · Citation-Capable Evidence and Unique Owned URLs
Citation Presence
Unit: Percent of eligible answers
How often an eligible answer includes at least one qualifying citation.
How it is calculated
Within the declared citation-capable surface/protocol and natural or requested-citation cohort, measure whether each complete eligible observable answer has a citation and apply balanced aggregation. An answer with ten citations still contributes one positive citation-presence outcome. An unknown or incomplete citation extraction is missing evidence, not a verified absence; a provider name alone does not establish citation observability.
How to interpret it
This measures citation behaviour, not brand mentions or source quality.
What to do next
Inspect which answers and systems provide usable references.
Measurement scope and aggregation ↗ · Evidence and availability ↗
Owned Citation Presence
Unit: Percent of eligible answers
How often an eligible answer cites a source that belongs to the declared owned-domain set.
How it is calculated
For each eligible citation-capable answer, check for at least one qualifying citation to the configured owned domains, count the answer once, and apply balanced aggregation. Domain ownership follows the report’s declared mapping; a marketplace listing is not automatically an owned domain. A confirmed owned citation establishes a positive outcome. Without one, unresolved ownership is unknown rather than a proven negative; complete capture with no citations is an observed negative.
How to interpret it
A mention without a link is not an owned citation. Citation presence does not measure visits or click-through.
What to do next
Compare cited pages with relevant pages that are accessible but not cited.
Related concepts: Citation Domain Presence ↗ · AI Referral Sessions ↗
Measurement scope and aggregation ↗ · Evidence and availability ↗
Citation Domain Presence
Unit: Percent of eligible answers
How often an eligible answer cites a specified domain.
How it is calculated
Count an answer once when it cites the domain, regardless of the number of URLs or repeated references from that domain; apply balanced aggregation. Several domains can be present in one answer, so domain-presence percentages can sum above 100%.
How to interpret it
Domain Presence is not Domain Citation Allocation Share. Neither is web traffic market share.
What to do next
Use presence to find sources repeatedly appearing across prompts and systems.
Related concepts: Domain Citation Allocation Share ↗
Measurement scope and aggregation ↗ · Evidence and availability ↗
Domain Citation Allocation Share
Unit: Percent of allocated cited-answer credit
A domain’s share of equal credit allocated across the distinct cited domains in each cited answer.
How it is calculated
Within each answer with citations, split one unit equally among its distinct cited domains. Combine these domain credits using conditional balanced aggregation over cited answers. A domain receives no credit in a cited answer that omits it. Repeated URLs from one domain do not add domain credit.
How to interpret it
Allocation shares sum to 100% across the complete included domain set. A displayed top list can sum below 100%. The metric describes observed citation allocation, not an influence or authority score.
What to do next
Compare allocation with domain presence to distinguish frequent inclusion from crowded citation lists.
Illustrative example
An answer citing A, A and B gives domains A and B one half each, rather than giving A two thirds.
Measurement scope and aggregation ↗ · Evidence and availability ↗
Citations, Unique URLs and Unique Domains
Unit: Observed counts
The volume and diversity of citation evidence in the selected scope.
How it is calculated
Total Citations counts distinct accepted answer-to-source support occurrences after deduplication. Duplicate records of the same supporting citation count once; the same URL supporting distinct answer passages can contribute more than once when those support occurrences are identified. If separate support occurrences cannot be identified, count that URL once per answer. Unique URLs counts distinct normalized cited URLs; Unique Domains counts distinct cited domains. Citation markers, support occurrences and unique URLs are therefore different units.
How to interpret it
These are observed counts, not balanced rates. URL normalization and deduplication can make them differ from a visual count of footnote markers.
What to do next
Use counts to understand support breadth; use presence and allocation rates for comparable behaviour.
Measurement scope and aggregation ↗ · Evidence and availability ↗
Source Type Composition
Unit: Percent or count of distinct domains
The types of sources represented in the observed cited-domain set.
How it is calculated
Each distinct cited domain contributes once to its primary category in the displayed source taxonomy. Category counts divided by all distinct cited domains give the composition shares, with unknown classifications retained. A separately labelled citation-exposure chart uses a different denominator and is not this inventory composition.
How to interpret it
This does not measure traffic, recommendation influence or the relative quality of source types.
What to do next
Inspect the actual sources before prioritising outreach.
Measurement scope and aggregation ↗ · Evidence and availability ↗
Source Access Status
Unit: Full / Partial / Metadata only / Unavailable
How much source content was available for inspection at collection time.
How it is assessed
The status records the source content available at inspection time: Full for the available complete content, Partial for a readable subset, Metadata only for descriptive metadata without the article body, or Unavailable when content could not be inspected. Citation presence is assessed independently of this access status.
How to interpret it
A cited URL can be inaccessible to SpyderBot. Source access does not establish truth, ownership or model training use.
What to do next
Review available evidence and access limitations before accepting a content-level conclusion.
Measurement scope and aggregation ↗ · Evidence and availability ↗
Source Brand Portrayal
Unit: Labels and distributions in source content
How accessible source content represents the selected brand.
How it is assessed
Target-linked portrayal is assessed only in source content available for inspection. Source-content labels describe that content, while answer sentiment describes the AI answer. The access status and observation time limit what can be concluded about the source.
How to interpret it
An AI answer can use a source without repeating its tone. Missing source text is not neutral portrayal.
What to do next
Open the source evidence and compare it with the answer evidence.
Related concepts: Sentiment Distribution ↗ · Source Access Status ↗
Measurement scope and aggregation ↗ · Evidence and availability ↗
Citation Set Agreement
Unit: Percent of comparable answer pairs
How consistently repeated cited answers use the same set of normalized citation URLs.
How it is calculated
Within the same measurement configuration, compare unordered pairs of answers that each contain citations. A pair agrees only if the complete normalized URL sets match. Divide agreeing pairs by comparable pairs. Fewer than two cited answers provides insufficient pair evidence. Domain-level variation is a different diagnostic.
How to interpret it
Agreement is not source accuracy. Two agreeing answers may share the same omission.
What to do next
Review citation substitutions when recommendations or portrayal also change.
Illustrative example
Three cited answers with sets {A,B}, {A,B}, {A,C} create three pairs; one agrees, so pair agreement is 33.3%.
Measurement scope and aggregation ↗ · Evidence and availability ↗
Observed Source Prompt Reach and Domain Coverage
Unit: Percent of full-depth prompt or system units
How widely citations or a particular cited domain appear across the declared prompt and system panel.
How it is calculated
Source Prompt Reach counts full-depth eligible Core Prompts with at least one citation divided by full-depth eligible planned Core Prompts. Domain Prompt Coverage counts those with at least one citation to the selected domain. Domain System Coverage counts full-depth planned systems citing that domain at least once divided by full-depth planned systems. A full-panel claim requires the complete declared panel; partial diagnostics are labelled separately with the original planned support visible.
How to interpret it
These at-least-once measures depend on sampling depth and are not average answer citation probability. Missing coverage must not become 100%.
What to do next
Compare compatible depths and inspect the prompts and systems supporting each source.
Related concepts: Citation Presence ↗ · Citation Domain Presence ↗
Measurement scope and aggregation ↗ · Evidence and availability ↗
Share of Cited Domains by Owner
Unit: Percent of cited-answer domain allocation
How citation allocation is divided among the declared domain ownership categories.
How it is calculated
Within each cited answer, divide the number of unique domains in each ownership category by all unique cited domains. Combine these vectors with conditional balanced aggregation. Use exclusive categories for tracked owners, Other and Unknown; keep unknown ownership visible. This differs from whether any owned domain appears in an answer.
How to interpret it
A top list retains the full denominator and does not renormalise visible rows to 100%. The result is not causal influence or traffic share.
What to do next
Check ownership coverage and source access before interpreting the allocation.
Related concepts: Owned Citation Presence ↗ · Domain Citation Allocation Share ↗
Measurement scope and aggregation ↗ · Evidence and availability ↗
Citation-Capable Evidence and Unique Owned URLs
Unit: Planned, delivered, observable or distinct URL counts
The available citation evidence under the actual collection surface and the distinct confirmed owned URLs cited.
How it is calculated
Count planned, delivered and citation-observable samples separately for the actual surface/protocol and natural or requested-citation question group. Unique Owned Cited URLs counts distinct normalized cited URLs with confirmed target ownership. A provider name alone does not establish citation capability, and retrieved-only pages are not answer citations.
How to interpret it
Counts depend on scope and depth. Unknown ownership is not confirmed nonownership, and more unique URLs do not imply more influence.
What to do next
Check citation capture completeness and domain ownership before comparing citation metrics.
Measurement scope and aggregation ↗ · Evidence and availability ↗