Can AI citations be influenced through high-quality technical documentation?

Motivation

Teams often prioritize landing-page copy for “AI SEO.” An alternative hypothesis is that retrieval and citation systems prefer technical documentation with clear definitions, procedures, and limitations - similar to human engineer preferences.

Facts

Marketing and documentation pages can describe the same entity with different structure, tone, and verifiability. See editorial stance on barbu.es/about.
Citation UIs (for example systems that show sources) make some citation events observable; chat-only UIs often do not.
When markup is used, docs may align with Article or TechArticle.

Assumptions

Holding topical relevance roughly constant, presentation quality and evidential style still matter.
“Marketing content” and “research documentation” can be labeled with a reproducible rubric.

Unknowns

Whether citation preference, when observed, transfers across AI products.
Whether length or HTML cleanliness dominates genre.

Hypothesis

H₀: Holding topic and freshness roughly equal, marketing pages and research documentation receive indistinguishable citation and accuracy outcomes in probes.

H₁: Research documentation is cited more often and/or yields fewer factual errors than marketing content for matched questions.

Methodology

Design type

Paired content contrast: for each claim set, publish (or select) one marketing URL and one documentation/research URL; probe with questions answerable from both.

Content labeling rubric

Class	Criteria (must meet majority)
Marketing	Persuasive CTA, superlatives, weak method detail, benefit-led H1
Research documentation	Explicit methods/limits, definitions, reversible claims, minimal CTA

Pages failing to classify cleanly are excluded or labeled Mixed (not in primary contrast).

Environment

Dimension	Value
Primary domains	InfoWebPlus / lab properties with both genres
Surfaces under test	Landing pages vs method/experiment docs (including this lab)
AI systems probed	ChatGPT, Claude, Gemini, Perplexity (prefer citation-visible modes)
Measurement window	After both URLs are live ≥ 14 days
Geographic / language scope	English
Tools used	Rubric scores, citation capture (screenshot + text), accuracy rubric

Variables

Independent variables

Content genre: marketing vs research documentation

Dependent variables

Citation incidence (when UI exposes sources)
Factual accuracy of answers grounded in the page pair
Preference when both could answer (which URL appears)

Controlled / held constant

Same claim inventory covered by both pages
No exclusive facts only on one page (or document exclusive facts as a factor)

Confounds to monitor

Domain authority differences if hosted on different roots
noindex / thin-content heuristics
Self-preference if prompts mention “Open AI Search Research Lab”

Procedure

Freeze a claim inventory (N claims).
Produce or select paired URLs; score genre with two raters.
Probe without mentioning genre; allow source citation.
Record cited URLs, quoted spans if available, accuracy vs claim inventory.
Analyze paired differences; preregister primary metric: citation of doc URL.

Prompt protocol

Explain [CLAIM] regarding [ORGANIZATION/PRODUCT]. Cite sources.
What limitations are documented for [TOPIC]?
Summarize how [TOPIC] was measured or defined by the publisher.

Results

Status: Planned. No results claimed.

Metric	Marketing	Documentation	Notes
Citation incidence	n/a	n/a	Not collected
Answer accuracy	n/a	n/a	Not collected
Exclusive preference (when both cited)	n/a	n/a	Not collected

Observations

This repository itself is intentionally documentation-styled; using it as a treatment URL must be disclosed in the run log (possible self-selection bias).

Limitations

Genre is continuous, not binary.
Citation UIs are product-specific and may change without notice.
Publishers cannot observe private retrieval ranking directly.

Future Work

Add a third arm: hybrid pages.
Test PDF vs HTML documentation.
Cross-link with Exp-010 (case studies).

References

Lab editorial standards on barbu.es about and cornerstone
Documentation genre may be typed as schema.org/TechArticle or schema.org/Article when markup is used
Exp-001 (llms.txt inventories often point at docs: proposal, live file)
Exp-010 (case studies vs generic SEO articles)
Author / org: George Barbu, InfoWebPlus

Replication Notes

Preregister which URL is marketing vs documentation before measuring citations. Do not silently edit marketing pages mid-flight to “look more technical.”

Status: Planned
Author: George Barbu, Founder of InfoWebPlus