The Seven Logic Gates for GEO Research: A Framework for Validating Claims About AI Citation Behavior

1. Introduction

The GEO industry regularly publishes findings that would fail basic logical scrutiny. Claims such as "websites with FAQ schema rank higher in AI results" typically derive from comparing websites that differ on dozens of variables simultaneously. The observed correlation between FAQ schema and higher ranking may be driven by any of those other variables, yet the claim is stated in causal language and treated as actionable advice.

The seven logic gates presented here provide a systematic method for identifying such errors before they propagate. Each gate tests a specific logical requirement, and failure at any gate indicates a flaw that must be addressed before the finding can be stated at its claimed confidence level.

2. The Seven Gates

2.1 Gate 1: Logical Form

Gate 1 evaluates whether the argument follows a valid logical form. Valid forms include modus ponens (if P then Q; P; therefore Q) and modus tollens (if P then Q; not Q; therefore not P). Invalid forms include affirming the consequent (if P then Q; Q; therefore P) and denying the antecedent (if P then Q; not P; therefore not Q).

GEO example of failure: "If high entity density causes citation, and this site was cited, then it must have high entity density." This affirms the consequent -- citation could have been caused by other factors.

2.2 Gate 2: Confound Check

Gate 2 requires listing every variable that differs between the groups being compared. If more than one variable differs, no causal conclusion is possible -- only correlational statements are permitted.

GEO example of failure: Comparing a cited site (3,000 words, FAQ schema, 47 named entities, 12 external links) with an uncited site (500 words, no schema, 3 named entities, 0 external links) and attributing the citation difference to any single variable. At least four variables differ; no causal attribution is valid.

2.3 Gate 3: Necessary/Sufficient/Contributory

Gate 3 applies J.L. Mackie's INUS framework to classify the claimed causal relationship. A factor may be necessary (citation never occurs without it), sufficient (citation always occurs with it), or contributory (increases the probability of citation without guaranteeing it). Most GEO factors are contributory under the INUS framework -- they are Insufficient but Necessary parts of an Unnecessary but Sufficient condition set.

GEO application: Entity density is likely contributory rather than necessary or sufficient. High entity density increases citation probability but does not guarantee citation, and citation can occur with low entity density if other conditions are met.

2.4 Gate 4: Counterfactual

Gate 4 asks: what would happen if the proposed cause were absent? The gate ranks counterfactual approximation methods from weakest to strongest:

2.5 Gate 5: Alternative Explanations

Gate 5 requires the researcher to generate at least three alternative explanations for the observed result and explicitly describe how each was ruled out. If alternatives cannot be ruled out, the finding must acknowledge them as potential confounds.

2.6 Gate 6: Replication

Gate 6 defines replication requirements for different confidence levels:

2.7 Gate 7: Mechanism

Gate 7 requires a plausible causal pathway. In GEO research, this specifically means verifying that the AI system has access to the proposed signal at inference time. A page-level signal cannot influence citation if the AI system never fetches the page. A schema markup cannot influence citation if the AI system does not parse structured data. Mechanism verification is essential for distinguishing genuine causal factors from coincidental correlates.

3. Application to SIGI Research

All 10 probe experiments in the SIGI Category A research pass all 7 gates because they use single-variable isolation (passing Gate 2), include the variable directly in the LLM prompt (passing Gate 7), and compare 15-19 variations of a single variable (providing counterfactual approximation for Gate 4). Category B-D findings that fail one or more gates are explicitly stated at reduced evidence levels with the failed gate identified.

4. Limitations

• Framework paper: The gates themselves are not empirically testable -- they are logical standards applied to empirical findings.
• Adapted from philosophy of science: The gates draw from established epistemology (Mackie, Popper, Hempel) adapted for AI behaviour research. Alternative logical frameworks exist.
• Application overhead: Rigorous gate checking requires significant analytical effort per finding, which may limit adoption in fast-moving industry contexts.

5. Conclusions

We present seven sequential validation gates that GEO research findings must pass before classification as validated. The gates are designed to catch the most common errors in GEO research: multi-variable confounding (Gate 2), over-claiming causal necessity (Gate 3), absence of counterfactual reasoning (Gate 4), failure to consider alternatives (Gate 5), and asserting mechanism without verification (Gate 7).

Note: This is a methodological framework paper. The seven gates provide a logical standard; their adoption would elevate the evidentiary quality of GEO research as a field.