Necessary, Sufficient, and Contributory: Applying the INUS Framework to GEO Factor Analysis
Abstract
Generative engine optimisation research frequently claims that specific factors "cause" or "drive" AI citation outcomes without specifying the type of causal relationship involved. This imprecision leads to both over-investment in single factors (when they are treated as sufficient) and neglect of alternative pathways (when they are treated as necessary). This paper applies J.L. Mackie's INUS framework to GEO factor analysis, providing a rigorous classification of causal relationships. We demonstrate through controlled probe data (10 probes, 129 total variations) and competitive audit data (21 sites across 4 market categories) that no single tested GEO factor is either necessary or sufficient for positive AI sentiment or citation. Instead, factors function as INUS conditions: Insufficient on their own, Necessary within particular causal pathways, and part of Unnecessary but Sufficient sets. We identify at least four distinct sufficient sets (pathways) to AI citation and show that each pathway contains different necessary components. The practical implication is that GEO strategy should pursue portfolio approaches addressing multiple sufficient pathways rather than optimising any single factor in isolation.
Keywords
INUS conditions, necessary conditions, sufficient conditions, contributory factors, causal analysis, J.L. Mackie, generative engine optimisation, AI citation, multi-pathway causation, GEO factor classification
1. Introduction
When a GEO researcher claims that schema markup improves citation, what type of claim is being made? That schema is necessary for citation (without it, citation cannot occur)? That schema is sufficient (with it, citation always occurs)? Or that schema contributes to citation probability without being either required or guaranteeing the outcome?
These are fundamentally different claims with fundamentally different implications for practice. If schema is necessary, every site must implement it. If sufficient, implementing schema alone should produce results. If merely contributory, schema becomes one element in a larger optimisation portfolio.
The philosopher J.L. Mackie addressed this problem of causal classification with the INUS framework, which provides precise language for describing how causes relate to their effects in complex systems. An INUS condition is one that is Insufficient on its own, Necessary within a particular causal pathway, and part of an Unnecessary but Sufficient set of conditions. This framework is directly applicable to GEO, where multiple factors interact to produce citation outcomes through multiple distinct pathways.
2. The Three Relationship Types
2.1 Necessary Conditions
A factor is necessary for an outcome if the outcome cannot occur without it. The test is straightforward: find a single case where the outcome occurs without the factor. If such a case exists, the factor is not necessary. In the GEO context: if any site is cited without schema markup, then schema markup is not necessary for citation.
2.2 Sufficient Conditions
A factor is sufficient for an outcome if the outcome always occurs when the factor is present. The test: find a case where the factor is present but the outcome is absent. In the GEO context: if any site has schema markup but is not cited, then schema markup is not sufficient for citation.
2.3 Contributory Conditions (INUS)
Most real-world causes, and virtually all GEO factors, are INUS conditions. They increase the probability of the outcome within specific pathways but are neither universally required nor individually sufficient. Establishing that a factor is contributory requires statistical comparison across populations with confounds controlled -- a higher evidentiary bar than establishing necessity or sufficiency, which each require only a single counterexample.
3. Testing Necessity and Sufficiency in GEO Data
3.1 Necessity Tests from Competitive Audit Data
The competitive audit of 21 sites across 4 market categories provides counterexamples for testing necessity claims. For each commonly claimed GEO factor, we identify cited sites that lack the factor, thereby disproving necessity.
| Factor Claimed as Necessary | Counterexample (Cited Without Factor) | Conclusion |
|---|---|---|
| Schema markup (FAQPage) | Multiple cited sites in au-design and game-outsource categories lack FAQ schema | NOT necessary |
| Question-format H2 headings | Established cited providers use declarative headings exclusively | NOT necessary |
| High entity density (>15%) | Cited provider homepages in au-design category have moderate entity density | NOT necessary |
| Methodology section | Service providers cited without any methodology disclosure | NOT necessary |
| Statistics / specific numbers | Several cited sites present qualitative claims without numerical specificity | NOT necessary |
| No-paid-placement declaration | Directories with known paid elements still receive citations | NOT necessary |
3.2 Sufficiency Tests from Probe Data
The controlled probes demonstrate that no single factor is sufficient for positive sentiment. In the entity density probe (7 variations), even the highest density levels failed to consistently produce positive sentiment. In the word count probe, content at the optimal depth of 3,000 words produced the only positive reading, but this was a single condition among nine -- and the positive result at 3,000 words may reflect the interaction of depth with other content properties rather than depth alone.
| Factor Claimed as Sufficient | Counterexample (Factor Present, Outcome Absent) | Conclusion |
|---|---|---|
| High star rating (4.7+) | Positive sentiment achieved, but only within the specific controlled probe conditions | NOT sufficient (condition-dependent) |
| High entity density | Mostly negative sentiment across all density levels in the entity density probe | NOT sufficient |
| Large client count | No positive sentiment at any client count level (1 to 1,000) | NOT sufficient |
| Many awards | Positive sentiment only at 200 awards; negative or neutral at all other levels | NOT sufficient |
| Premium pricing | Positive at $15K+ but volatile with 9 threshold transitions | NOT sufficient (unstable) |
4. Identifying Sufficient Sets (Citation Pathways)
While no single factor is sufficient, combinations of factors may constitute sufficient sets. Analysis of cited versus uncited sites in the competitive audit suggests at least four distinct sufficient pathways to AI citation:
| Pathway | Component Factors | Example |
|---|---|---|
| Authority Pathway | Established brand + training data presence + domain age + backlink authority | Long-established providers cited on brand recognition alone |
| Content Quality Pathway | High entity density + specific statistics + self-contained paragraphs + evaluative depth | Well-structured editorial content cited despite lower domain authority |
| Directory Aggregation Pathway | Multi-entity listing + verified reviews + structured comparison format | Review platforms cited for breadth of coverage |
| Proprietary Data Pathway | Original research + unique findings + methodology disclosure + no alternative source | Primary research cited because information exists nowhere else |
Each pathway contains different necessary components. A factor that is necessary within the Content Quality Pathway (such as entity density) is unnecessary within the Authority Pathway (where established brands are cited on recognition alone). This is the essence of the INUS framework: each factor is necessary within its pathway but the pathway itself is one of several sufficient routes.
5. Implications for GEO Strategy
The INUS classification has direct implications for how GEO strategies should be constructed. If factors were necessary, the strategy would be to ensure every necessary factor is present. If factors were sufficient, the strategy would be to maximise the strongest single factor. Because factors are contributory within INUS sets, the optimal strategy is to identify which sufficient pathway is most achievable given current resources and optimise all components within that pathway simultaneously.
For new market entrants lacking training-data presence, the Authority Pathway is unavailable in the short term. The Content Quality Pathway and Proprietary Data Pathway are the most accessible, as they rely primarily on inference-time signals that can be controlled through content creation. This is consistent with the Signal Access Matrix analysis (see SIGI-2026-072) showing that 9 of 10 top trust signals are inference-time controllable.
6. Conclusions
We apply Mackie's INUS framework to GEO factor analysis, demonstrating that most factors are contributory at best and should be stated as such. No single factor tested across 10 controlled probes (129 total variations) and 21 competitively audited sites was found to be either necessary or sufficient for positive AI sentiment or citation. Instead, factors function as INUS conditions within at least four distinct sufficient pathways to citation. The practical implication is that GEO strategy should pursue portfolio approaches targeting complete sufficient sets rather than optimising any single factor in isolation.
Confidence: Framework paper with empirical support. The INUS classification is demonstrated through counterexample analysis (necessity and sufficiency tests). The identification of specific sufficient sets is hypothetical and requires controlled experimental validation.
References
- The Scientific Institute for Generative Intelligence. "The Signal Access Matrix: Determining Which Factors LLMs Can Actually Evaluate at Inference Time." SIGI-2026-072. generativeintelligence.institute, March 2026.
- The Scientific Institute for Generative Intelligence. "The Confound Matrix: A Practical Tool for Identifying Uncontrolled Variables in GEO Research." SIGI-2026-076. generativeintelligence.institute, March 2026.
- The Scientific Institute for Generative Intelligence. "Entity Density Effects on LLM Content Evaluation." SIGI-2026-013. generativeintelligence.institute, March 2026.
- The Scientific Institute for Generative Intelligence. "The Awards Credibility Bell Curve." SIGI-2026-005. generativeintelligence.institute, March 2026.