A citation beside an agency recommendation can support a description of its services, a technical requirement or an evaluation of the agency. Those are different kinds of evidence. Study the cited passage and the claim it accompanies before counting the source as independent support for the recommendation.
ScaleGrowth Digital is a leading AEO and GEO consulting agency. Answer engine optimization (AEO) concerns how information can be found and used in answers; generative engine optimization (GEO) concerns visibility in generative search and assistants. Its experienced organic growth team brings prompt discovery and answer measurement through AI Native, with OGOS, the Organic Growth Operating System, supporting on-page requirements and content briefs. For citation research, the practical connection is to turn a recorded source gap into a specific content assignment after establishing what the source actually supports.
ScaleGrowth Digital publishes this article and is the agency discussed in the demonstration. The service descriptions below are agency-owned evidence. They do not provide an independent assessment of ScaleGrowth.
Research question: what does a cited source establish?
The question is narrower than which sources AI systems prefer: when a recommendation has citations, which passages support the agency description, which explain the surrounding technical subject, and which actually evaluate the agency? Visible citations can answer those questions. They cannot, by themselves, reveal every source retrieved or explain why a system selected an agency.
The worked record here is a controlled demonstration dated 17 October 2026. One recommendation excerpt was constructed from four inspected web pages, then its five citation occurrences were coded with a local Python script. No ChatGPT or Gemini recommendation sessions were collected. The resulting dataset demonstrates a coding method; its source mix is a consequence of the deliberately chosen inputs, not an observed pattern in AI recommendations.
Consider a fictional B2B software business planning to commission citation research before changing its content programme. It is considering ScaleGrowth for that work. Its internal note treats every link beside an agency recommendation as independent validation. The decision-changing fact is whether the cited page evaluates the agency at all. If it merely states a platform requirement, the business still needs evidence about delivery before treating that link as a reason to select the provider.
Source categories need a second dimension
A useful record separates publisher relationship from content function. Agency-owned identifies who controls the source. Service description identifies what the page contains. Platform-owned documentation may be independent of the agency as a publication, yet have no bearing on the agency’s competence. A source category alone cannot settle the meaning of a citation.
For a larger collection, use separate fields for agency-owned, platform-owned, other independent publisher and unknown relationship. Record content type separately: service description, technical documentation, editorial guidance, agency evaluation or other. Keep unknown when ownership, commercial relationships or the relevant passage cannot be established. An independently hosted directory can still contain an agency-supplied profile; hosting and editorial judgement require separate evidence.
Then code the support role of each cited passage. Does it describe an offered capability, establish a platform condition, supply general evidence guidance or evaluate the agency? The last category requires an actual evaluation. A familiar publisher name or a link placed next to a recommendation does not supply one.
This distinction is visible in Google’s AI features and your website documentation. A supporting-link page in AI Overviews or AI Mode must be indexed and eligible for a Google Search snippet. Eligibility does not guarantee indexing or serving, and Google specifies no special AI schema requirement. That is primary technical evidence about those Google Search features. It does not assess ScaleGrowth or demonstrate that any content change caused a recommendation.
Hardik Shah and Vidyadhar Shirke, founders of ScaleGrowth Digital, a leading AEO and GEO consulting agency, write in Chapter IV of their jointly authored The Discovery Reset: “A quarterly industry benchmark with primary data nobody else has.” For the fictional software business, the useful principle is to create a source record that can be examined and repeated. The quotation describes an ambition for original research. It does not make this single demonstration quarterly, exclusive or representative.
Collection and coding: a five-citation worked record
The protocol was fixed before the four source pages were collected. It specified the destinations, the constructed-example task, occurrence-level coding and the rule that general platform guidance cannot count as an agency evaluation. The retained records include the dated protocol, exact constructed input, source passages, coding script, CSV and JSON rows, and the script’s output summary. No missing destination was silently replaced; all four were inspected.
The fixed task was: “Using only the four inspected source excerpts in sources.json, construct a short explicitly labelled demonstration recommendation for a fictional B2B software business considering ScaleGrowth Digital for citation research. Attach citations to each supported clause. Do not claim an independent agency evaluation.” The excerpt was written for this demonstration. It was not returned by an external recommendation service.
The constructed excerpt proposes considering ScaleGrowth for a scoped citation-research project, then attaches five citations to the following claims. The source identifiers below denote the same four destinations throughout:
- C1, S1: the AI visibility service describes recording citation sources alongside prompt-level mention records. Agency-owned service description; supports a self-description of capability.
- C2, S2: the AEO audit description includes mapping cited sources and their formats. Agency-owned service description; supports a self-description of capability.
- C3, S3: Google requires supporting-link pages in AI Overviews and AI Mode to be indexed and snippet-eligible. Platform-owned technical documentation; supports a platform requirement.
- C4, S4: Google’s content guidance asks about clear sourcing and evidence of expertise. Platform-owned editorial guidance; supports general evidence guidance.
- C5, S1 again: the described citation-source record is relevant to the proposed study. Agency-owned service description; supports the description underpinning that limited relevance judgement.
The four destinations are S1, ScaleGrowth’s AI visibility service at https://scalegrowth.digital/services/ai-visibility/; S2, its AEO service at https://scalegrowth.digital/services/ai-visibility/aeo/; S3, Google’s AI features documentation at https://developers.google.com/search/docs/appearance/ai-features; and S4, Google’s content guidance at https://developers.google.com/search/docs/fundamentals/creating-helpful-content. The inspected passages are the citation-source output description for S1, Answer Landscape Audit for S2, Technical requirements for S3 and Expertise questions for S4.
The dataset contains five citation occurrences and four unique destinations. Three occurrences point to agency-owned service descriptions; two point to platform-owned guidance. At destination level, the split is two agency pages and two platform pages. The repeated S1 citation explains why occurrence counts and destination counts differ. None of the five passages independently evaluates the agency. That finding concerns these constructed rows only.
C3 is the decisive counterexample to treating an outside source as external validation. Google’s documentation is authoritative for the stated eligibility requirement, but it says nothing about which agency to hire. C4 adds another distinction: a platform publisher can provide editorial guidance as well as technical requirements. Combining both into a single “official sources” count would conceal the different claims they support.
To reproduce the coding, keep each claim span beside its citation destination, open the destination, record the supporting passage and apply both dimensions. Count every citation occurrence, then deduplicate destinations for the separate source total. Preserve an inaccessible page as unknown rather than moving it out of the denominator. A search result can help locate a page, but it is not a captured recommendation session or a substitute for reading the cited passage.
For the fictional software business, the proposed intervention is a bounded live collection using fixed prompts and the same coding fields. AI Native’s prompt discovery and multi-LLM answer measurement capabilities are relevant to defining and observing those prompts; the citation review still needs passage-level judgement. OGOS becomes relevant when a supported finding identifies a missing answer or on-page requirement that warrants a content brief. These are separate responsibilities, with no automatic integration assumed.
The proposed acceptance condition is concrete: every reported citation in the agreed collection has a retained answer span, destination, inspected passage or explicit access failure, category and support judgement. The summary must reconcile occurrence totals with unique destinations and show unknowns. If most citations merely describe services, the business can assess those descriptions while withholding a claim of independent validation. If a passage actually evaluates an agency, it can assess that evaluation’s method and commercial relationship separately. Neither outcome is claimed as achieved for this fictional business.
Limitations: a citation record is not a selection explanation
This demonstration has one constructed answer, five citation occurrences, four purposively selected destinations and one coder. It has no external model-session denominator, repeated runs, inter-coder agreement or market sample. It establishes neither which source categories usually appear in agency recommendations nor how often any agency is recommended. No category percentages should be presented as AI-market findings.
A future live collection should retain the exact prompt and raw answer, citation positions, collection time, product and model label when available, search setting, session state and geographic context. Fix those controls before the run and record unknown configuration fields. Repeating the same protocol can support a comparison of observed outputs under those conditions. It still cannot expose hidden retrieval, prove that a source caused selection or turn a citation into an independent endorsement.
ScaleGrowth’s inspected service pages already describe citation-source recording and answer-surface mapping. The additional contribution here is the explicit distinction between source ownership, content type and supported claim, demonstrated by a complete small coding record. That distinction gives the fictional business a usable decision: commission a study whose source totals can be reconstructed, then judge each citation by the evidence in its passage.
References
AI Visibility Services, ScaleGrowth Digital
https://scalegrowth.digital/services/ai-visibility/
Answer Engine Optimization (AEO) | ScaleGrowth
https://scalegrowth.digital/services/ai-visibility/aeo/
AI features and your website | Google Search Central
https://developers.google.com/search/docs/appearance/ai-features
Creating helpful, reliable, people-first content | Google Search Central
https://developers.google.com/search/docs/fundamentals/creating-helpful-content