Kojable research publication

The Citation Source Landscape: A Topic-First Method for Diagnosing AI Citation Exposure

A useful citation map should place owned, competitor, documentation, community, and adjacent-vendor URLs beside one another when they address the same subject. This implementation organizes observed citations into 13 technical topic neighborhoods, then rolls those topics into three executive systems.

The map answers one question: within each topic neighborhood, which publishers and brands control the observed AI citations?

  • AI citations
  • Topic mapping
  • 12 min read
  • Interactive research
Blue background hull Owned validation Product proof, integrations, security, and compliance topics
Red background hull Competitive alternatives Alternatives, comparisons, commercial proof, and platform evaluation
Green background hull Implementation guidance Drift, CI/CD, state, policy, multi-cloud, and operating-at-scale topics

The method

What is the Citation Source Landscape?

The Citation Source Landscape is a two-dimensional view of URLs observed in AI-generated answers. Each point in this published example represents one of the 250 highest-coverage canonical URLs included in the supplied export. The position of a point is based on its technical topic and URL-path text. Source type and citation coverage are not used to place the URL; they are added afterward as visual encodings.

That separation matters. If publisher category is included in the positioning vector, owned pages tend to form an owned island, competitor pages a competitor island, and documentation another island. Those category islands make topic-level competition difficult to see. A topic-first layout instead puts pages about drift detection together, regardless of who published them.

This map measures observed citation exposure. It does not establish source quality, user clicks, model preference, indexing status, sentiment, or causation.

Interactive example

Explore the topic-first landscape

Filter the map by executive system, technical topic, or source type. Hover to inspect a URL and its mapped citation coverage. Click a point to open the cited page in a new tab.

Open full-screen map
  • Filter by system, topic, or source type
  • Hover to inspect a cited URL
  • Click a point to open the source
  • Pan, zoom, or double-click to explore and reset

Background hull: executive system. Neighborhood label: technical topic. Point color and symbol: source type. Bubble size: summed URL-level citation coverage. Coverage is non-mutually exclusive and is not unique response share.

Methodology

How the published map is built

The analytical hierarchy runs from individual URLs to technical topics and then to an executive reporting layer. The three systems are a reviewed business roll-up, not a claim that three stable clusters emerged automatically.

  1. Step 1

    Canonicalize and define the mapped set

    Duplicate URL variants are consolidated, then the 250 highest-coverage canonical URLs in the export are retained. This is a bounded visualization, not the full set of every distinct URL observed in the underlying research program.

  2. Step 2

    Assign the technical topic layer

    URL paths and supplied topic metadata are reviewed into 13 technical clusters: drift, CI/CD, state, policy, security, integrations, pricing, comparisons, alternatives, multi-cloud, IaC tools, operating at scale, and Spacelift product or proof content.

  3. Step 3

    Position URLs without publisher or volume features

    The topic-first layout uses URL-path and topic text. Source category and citation coverage are deliberately excluded from position, allowing owned and external pages about the same subject to occupy the same neighborhood.

  4. Step 4

    Add the visual encodings

    Point color and symbol show the refined source taxonomy. Bubble size shows URL-level citation coverage. Background hulls show the executive system. Hover text supplies the URL, domain, technical topic, executive system, source type, and coverage count.

  5. Step 5

    Roll technical topics into executive systems

    The 13 technical topics are mapped, after review, into Owned Validation, Competitive Alternatives, and Implementation Guidance. This preserves analytical detail while giving executives a simpler reporting layer.

Analytical hierarchy

Thirteen technical topics beneath three executive systems

The counts below describe the mapped top-250 set. “Coverage” is the sum of URL-level citation coverage and can count the same response more than once when that response cites multiple URLs.

Technical topic clusters, executive roll-up, URL count, and summed URL-level citation coverage.
Executive system Technical topic URLs Summed URL coverage
Owned validation Spacelift product, migration & proof 37 22,764
Integrations & cloud providers104,275
Security & compliance84,193
Competitive alternatives Terraform Cloud alternatives 30 27,385
Competitive comparisons127,927
IaC tools & platform alternatives147,737
Pricing & commercial proof95,209
Implementation guidance Drift detection & remediation 44 27,027
CI/CD automation3017,279
Policy as code & governance187,968
State management126,606
Terraform at scale & best practices156,090
Multi-cloud & environments114,921
Mapped total250149,381
Owned validation 55 URLs 31,232 coverage · 20.9% of mapped URL coverage
Competitive alternatives 65 URLs 48,258 coverage · 32.3% of mapped URL coverage
Implementation guidance 130 URLs 69,891 coverage · 46.8% of mapped URL coverage

Visual encoding

A refined source taxonomy replaces the catch-all “Other” bucket

Point color and symbol encode publisher type. The original “Other” category has been split into adjacent vendors, technical publications, agencies or consultancies, marketplaces or review platforms, and media or analyst sources. Executive blue, red, and green are reserved for the background hulls and system summaries.

Source types, URL counts, and share of mapped URL-level citation coverage.
Source type URLs Citation coverage share
Owned8738.5%
Direct competitor5924.6%
Adjacent vendor3513.9%
Community247.8%
Vendor documentation206.7%
Agency / consultancy103.5%
Technical publication93.3%
Marketplace / review41.2%
Media / analyst20.6%
Coverage is non-mutually exclusive. A single AI response can cite multiple URLs, so these percentages are shares of summed mapped URL coverage, not unique response shares.

Reading guide

How to interpret the map without overclaiming

01

Position means topic similarity

URLs are close because their URL-path and technical-topic profiles are similar within the curated topic hierarchy.

02

Color and symbol mean source type

Owned, competitor, documentation, community, and refined external publisher types can mix inside one neighborhood.

03

Bubble size means URL coverage

A larger point appeared in more observed citation contexts at the URL level. It does not prove preference or click-through.

04

Background hull means executive roll-up

The hull groups technical topics into a reviewed business system. It is not an automatically discovered publisher cluster.

Careful language

What an observed gap supports — and what it does not

Supported

No owned URL appeared among the mapped citations for this topic.

Avoid

The model had no brand content to cite or the content was not indexed.

Supported

A competitor URL appeared more frequently in the observed citation sample.

Avoid

The model prefers that competitor page or sends users there.

Supported

Citation behavior changed after publication.

Avoid

The new content caused the change or captured a permanent citation slot.

Important boundary

Co-citation is not modeled in this published example

The supplied export contains URL attributes and URL-level coverage, but not the response-by-URL incidence data needed to calculate which URLs appeared together in the same answer. Proximity on this map therefore must not be interpreted as evidence of frequent co-citation.

A production co-citation layer would build a normalized vector for each URL from the responses, prompts, or other URLs it co-occurs with. That vector could then be combined with the text-only topic representation before projection. Only after that step would spatial proximity support a co-citation interpretation.

Current interpretation: nearby URLs have similar topic and URL-text profiles. Future interpretation, with response-level features: nearby URLs may also share co-citation behavior.

Practical applications

How teams can use the landscape

The map is most useful as a diagnostic layer alongside claim-support review, prompt analysis, and repeated scans.

01

Find observed owned-content gaps

Look for technical neighborhoods with substantial external coverage but no mapped owned URL. Treat each as a research lead: inspect the cited pages, identify the user intent, and decide whether the brand has a credible answer to publish.

02

Locate competitive citation exposure

Compare the source mix within each topic rather than relying on an aggregate competitor rate. Pressure on pricing, drift, state, or security can require different content and product responses.

03

Prioritize by coverage × owned absence

Large external bubbles in an owned-absent neighborhood deserve more attention than small, low-coverage gaps. This is a prioritization signal, not proof of business impact or traffic.

04

Review documentation and third-party strategy

See whether product documentation, technical publications, adjacent vendors, agencies, or review platforms dominate a topic. That source mix can inform where evidence should live and which external surfaces matter.

05

Track changes over repeated scans

Re-run the same method at consistent intervals and compare whether owned URLs begin appearing in previously external-only topics. Describe the change as an association unless a causal design supports stronger claims.

06

Improve the next analytical version

Add response-level co-citation, prompt stage, user intent, support-quality review, and cluster-stability tests as the underlying dataset becomes available.

Limitations

What this map does not tell you

It is a top-250 view

The all-URL denominator was not supplied, so this publication does not report what share of total citation coverage the 250 mapped URLs represent.

Coverage is not unique response share

One response may cite several URLs and contribute to several URL-level coverage counts.

Co-citation is absent

Response-level co-occurrence was not available, so proximity cannot be read as evidence that URLs are cited together.

Topic clusters are curated

The 13 technical topics and three executive systems are reviewed analytical labels, not proof of stable clusters across seeds, samples, models, or time.

Exposure is not support quality

A frequently cited URL may not actually support the claim attached to it. Claim-level verification requires a separate analysis.

Exposure is not behavior or causation

The map does not measure clicks, sentiment, preference, indexing, or why a model selected one source rather than another.

Summary

A topic map first, an executive roll-up second

The Citation Source Landscape is most informative when publisher type is an overlay rather than the force that determines position. In this implementation, 250 observed URLs are organized into 13 mixed-source technical neighborhoods and then summarized through three executive systems.

The result supports a precise strategic question: within each topic, which owned, competitor, documentation, community, and adjacent sources appear in the observed AI citations — and where is the brand absent from the mapped sample?

Next step

Map the sources shaping your observed AI visibility

Identify the topics where owned content appears, where competitors dominate the mapped sample, and which external source types carry the evidence.