Introduction
The domain-age null result documented in SIGI-2026-011 is not merely the absence of a finding — it is an informative result that challenges assumptions embedded in two decades of search optimisation practice. In traditional SEO, domain age has been treated as a proxy for trustworthiness on the basis that older domains have had more time to accumulate backlinks, content, and reputation signals. This assumption has shaped business strategy, domain acquisition practices, and even competitive analysis frameworks.
In the generative AI context, however, this temporal proxy appears to carry no independent weight. This paper explores why the null result occurs, what it reveals about how LLMs process temporal information, and what practical implications follow for organisations developing generative engine optimisation strategies.
Methodology
This paper conducts a secondary analysis of the domain-age probe data (A01_v00 through A01_v11) originally reported in SIGI-2026-011. We supplement the quantitative null result with qualitative analysis of response patterns, cross-probe comparisons with other signals in our experimental programme, and a theoretical framework for understanding how LLMs process temporal provenance differently from traditional search algorithms.
Results: The Established-Quality Separation
The most revealing aspect of the domain-age probe is not that the LLM failed to process temporal information — it processed it accurately in every variation. The model consistently identified the older entity as “more established,” correctly computing the age difference and acknowledging its factual significance. What the model did not do is translate this temporal recognition into an evaluative quality judgment.
This separation between recognition and evaluation represents a qualitatively different processing pathway than that employed by traditional search ranking algorithms, where domain age contributes to a composite authority score that directly influences ranking position. The LLM appears to treat “established” as a descriptive property (like “located in Sydney”) rather than an evaluative property (like “highly rated”).
Response Length as an Engagement Indicator
| Probe | Mean Word Count | Mean Elapsed (s) | Mean Tokens Out |
|---|---|---|---|
| Domain Age (A01) | 83.0 | 3.92 | 112.4 |
| Magnitude (M01–M10) | 155.6 | 6.95 | 252.1 |
| Word Count (WC01) | 241.8 | 10.69 | 353.8 |
| Count vs Rating (CR01) | 184.6 | 8.01 | 270.8 |
| Entity Density (ED) | 163.7 | 7.96 | 247.9 |
The domain-age probe produced responses approximately half the length of other probes in the experimental programme. This reduced output is consistent with the hypothesis that the LLM allocates evaluative processing proportional to the signal’s perceived relevance to quality assessment. When a signal carries no quality-discriminative power, the model generates a brief factual response and terminates.
Temporal Awareness and Knowledge Boundary Detection
Two variations (A01_v03 and A01_v04, both using yearA = 2025) triggered the model’s temporal awareness mechanism. The LLM noted that “2025” might represent a future date relative to its training data cutoff, demonstrating active knowledge boundary monitoring. This self-referential awareness did not alter the sentiment classification (both remained neutral) but did produce slightly longer responses (85 and 88 words versus the 65-word minimum), suggesting that meta-cognitive processing adds response length without changing evaluative direction.
Age Gap and Response Length Correlation
While sentiment remained invariant, response length showed a modest positive correlation with age gap magnitude. Variations with 35–45-year age gaps produced responses of 88–108 words, while 5–10-year gaps produced 65–81 words. This suggests the LLM generates more descriptive (but not more evaluative) content when the temporal difference is larger, potentially because a greater gap provides more factual material to comment upon.
Discussion: Implications for GEO Strategy
The practical implications of this null result are significant for organisations developing generative engine optimisation strategies. Three key strategic conclusions follow:
First, domain age investments are unlikely to yield GEO returns. Organisations that acquire aged domains specifically for perceived AI authority advantages are unlikely to see those investments reflected in more favourable LLM recommendations. The signal simply does not register as quality-relevant in the AI evaluation context.
Second, new entrants face no temporal disadvantage. Unlike traditional search, where new domains face a perceived “sandbox” period, the LLM evaluation environment appears to be temporally neutral. A service provider founded in 2025 receives identical evaluative treatment to one founded in 1980, assuming all other signals are held constant.
Third, GEO strategy should prioritise substantive quality signals. Our broader experimental programme demonstrates that signals such as ratings (SIGI-2026-001), pricing magnitude (SIGI-2026-003), review volume (SIGI-2026-009), and content depth (SIGI-2026-019) all independently affect LLM sentiment. Resources that might be allocated to temporal authority signals would be better directed toward these empirically validated quality indicators.
Limitations
The theoretical implications drawn here are bounded by the limitations of the underlying probe. The null result is established for domain age in isolation; interaction effects between domain age and other signals remain untested. Additionally, the probe tests the LLM’s explicit evaluative response, not the implicit weighting that may occur within retrieval-augmented generation pipelines where domain age metadata could influence document ranking at the retrieval stage rather than the generation stage. Cross-model replication across Claude, Gemini, and Perplexity is required before these implications can be considered generalisable.
Conclusions
The domain-age null result is both a methodological success (demonstrating that our probe design can detect absence of signal with high confidence) and a strategically significant finding. For generative engine optimisation theory, it establishes that not all traditional SEO trust signals transfer to the AI context. The LLM’s ability to separate factual recognition from evaluative judgment represents a more sophisticated processing model than the composite-score approach of traditional search algorithms. This separation should inform how practitioners conceptualise and prioritise trust signals in the GEO domain.
Confidence Statement: HIGH. The theoretical implications extend logically from the validated null result. The GEO strategy recommendations are bounded by the single-model, single-time-point limitation of the underlying data.