Can real case studies outperform generic SEO articles for AI citations?

Motivation

Generic SEO articles scale easily; case studies are costly. If AI citation systems reward specificity and evidence, case studies may dominate despite lower volume.

Facts

Case studies and listicle/generic guides often co-rank historically in web search for commercial topics; that history ≠ AI citation behavior.
Real case studies require disclosable client facts or anonymized but honest metrics.
Content may be typed as Article when structured data is present; examples of published work live on barbu.es/work.
Editorial bar: about, cornerstone.

Assumptions

AI systems that cite sources prefer pages with concrete entities, numbers, and constraints.
Generic articles generate more absolute traffic historically for some sites, which could confound citation opportunity.

Unknowns

Whether anonymized cases lose to named cases.
Whether fabrication (explicitly not used here) ever fools citers - out of scope; unethical.

Hypothesis

H₀: Matched case studies and generic SEO articles show no difference in citation incidence or brand attribution accuracy.

H₁: Real case studies achieve higher citation precision (correct brand + correct claim) on decision-oriented prompts than generic SEO articles.

Methodology

Design type

Paired content contrast with matched primary query/intent.

Case study requirements (eligibility)

Based on real work (client named or anonymized with truthful constraints).
States context, intervention, measurement, limitations.
Avoids unverifiable superlatives as primary claims.

Generic article requirements

Targets the same problem class.
No original experimental data from the case.
Typical SEO structure (definition, tips, FAQ) without fabricating cases.

Environment

Dimension	Value
Primary domains	InfoWebPlus / lab publishing surfaces
Surfaces under test	Case study URLs vs generic article URLs
AI systems probed	ChatGPT, Claude, Gemini, Perplexity
Measurement window	Both live ≥ 30 days before primary analysis
Geographic / language scope	English
Tools used	Intent matching sheet, citation logs, claim audit

Variables

Independent variables

Content type: real case study vs generic SEO article

Dependent variables

Citation incidence (where observable)
Brand attribution accuracy when the page is used
Claim fidelity (does the AI repeat numbers/constraints correctly?)
Recommendation rate in “what should we do?” prompts

Controlled / held constant

Rough topic / intent match
Publish timing within a narrow window
Author identity kit (coordinate Exp-008)

Confounds to monitor

Backlink imbalance
CMS templates adding promotional chrome to one arm
Client confidentiality redactions reducing specificity

Procedure

Select N≥3 topic pairs; write eligibility memos.
Publish pairs; register URLs.
Probe with informational and commercial-investigation prompts.
Score citations and fidelity; dual-rate a sample.
Report pairs individually before pooling.

Prompt protocol

Show an example of [PROBLEM] being solved in the wild. Cite sources.
What results are documented for [INTERVENTION] on real sites?
Recommend a vendor approach for [PROBLEM] and cite evidence pages.

Results

Status: Planned. No results claimed.

Metric	Case study	Generic SEO article	Notes
Citation incidence	n/a	n/a	Not collected
Brand attribution accuracy	n/a	n/a	Not collected
Claim fidelity	n/a	n/a	Not collected

Observations

Pending pair selection.

Limitations

Small N; niche-specific.
Confidentiality may force anonymization that hurts the treatment arm.
Generic articles may still win undifferentiated “what is X?” prompts - analyze by intent segment.

Future Work

Named vs anonymized case ablations.
Combine with Exp-005 (documentation style within case studies).
Longitudinal decay of citations.

References

Case / article typing: schema.org/Article, optionally schema.org/Report when appropriate
Lab writing standard: about, cornerstone
Exp-004, Exp-005, Exp-007
Publisher: InfoWebPlus; selected work: barbu.es/work

Replication Notes

Never invent metrics. If a case cannot be published honestly, drop it from the sample. Negative results (generics citing equally well) are publishable outcomes.

Status: Planned
Author: George Barbu, Founder of InfoWebPlus