What Distinguishes Negative Signals from Low-Scoring Positive Signals?
The model drew a clear distinction between signals that simply fail to contribute to citation probability (low-scoring positive signals like sitemap priority at 1.0) and signals that actively reduce citation probability (negative signals). A page without sitemap priority loses nothing; a page with detectable sponsored labels loses trust that would otherwise be present. This distinction has practical significance: while low-scoring positive signals can be safely ignored in optimisation efforts, negative signals require active avoidance or remediation.
What Are the Eight Negative Trust Signals?
| Signal | Score | Layer | Mechanism |
|---|---|---|---|
| JavaScript Dependency | 7.5 | Inference | Content invisible to AI crawlers; binary visibility gate |
| Sponsored/Paid Label Detection | 7.5 | Inference | Inference-time trust discount; additive with training knowledge |
| Syndicated Content on Low-DA Domains | 7.5 | Training + Inference | Manufactured breadth; zero validation |
| Known Paid Directory Status | 7.0 | Training | Permanent baked-in trust discount |
| Self-Ranking at #1 | 7.0 | Inference | Contaminates entire ranking credibility |
| Uniform Positive Descriptions | 6.5 | Inference | Paid-directory pattern; real evaluation includes differentiation |
| Detectable AI-Generated Content | 6.0 | Inference | Consensus content; zero information gain |
| Affiliate Link Presence | 6.0 | Inference | Commercial incentive via URL parameters (utm_*, ref=, aff=) |
How Does Syndicated Content Create a Negative Signal?
The model described syndicated press release content, where the same press release appears across multiple low-authority domains (anonymised as "Syndication Network Alpha," "Syndication Network Beta"), as creating an appearance of media coverage breadth that carries zero actual validation. The model reported that it can detect syndication patterns: identical or near-identical content across multiple low-domain-authority sites is recognised as purchased distribution rather than earned coverage. One genuine editorial placement in a recognised publication was described as outweighing all syndicated placements combined.
Why Is Detectable AI-Generated Content a Negative Signal?
The model reported that AI-generated content tends to produce consensus content with zero information gain. In a RAG re-ranking process evaluating 20-50 candidate chunks, content that says the same thing as every other source scores near-zero on information gain. The model described strong human editorial voice with original opinion as a competitive advantage because it provides the novel perspective that re-rankers prioritise. The negative signal is not that the content was generated by AI per se, but that it lacks the originality and unique perspective that drives citation selection.
What Methodology Was Employed in This Research?
This paper consolidates all negative-direction signals from the 77 Trust Signal Taxonomy (see SIGI-2026-021). The asymmetry claim (negative signals outweighing positive signals) is a qualitative self-report that has not been quantified. PR syndication services are anonymised as "Syndication Network Alpha," "Syndication Network Beta." The introspective elicitation methodology and its limitations are detailed in the parent paper.
What Are the Limitations of This Research?
The asymmetric weighting claim is unquantified and based solely on the model's self-report. The actual magnitude of negative signal impact relative to positive signals has not been measured through controlled experiments. The AI-generated content detection mechanism is not fully characterised: the model did not specify whether it detects AI-generated content through stylistic patterns, lack of originality, or both. The syndication detection mechanism may vary across AI platforms with different training data compositions.
Conclusions
Eight negative signals were identified that the model reports as actively reducing citation probability. The asymmetric weighting claim suggests that avoiding these patterns may yield greater citation improvements than adding additional positive signals. The practical implication is that GEO audits should begin by identifying and remediating negative signals before investing in positive signal optimisation. These findings generate specific testable hypotheses about the comparative impact of negative signal removal versus positive signal addition.