Methodology
Our scoring rubric, including the parts that did not survive contact with evidence.
Everyone in this category claims a methodology. Almost nobody publishes the weights. Here are ours, generated from the same configuration the scorer reads, with an honest label on the strength of the evidence behind each factor.
10 scored factors · 100 points
The weights
17 industry study
Direct answer opening
Large-scale citation analyses consistently find the opening section of a document carries a disproportionate share of citations. The specific figures circulating for this are industry measurements, not refereed work.
15 industry study
Question intent match
Overlap between a page title and the question asked correlates with citation in vendor studies. We weight it highly because it is cheap to get right, not because it is settled science.
14 industry study
Structured format
Tables, lists and question-shaped headings are repeatedly associated with citation across vendor datasets. Corroborated across several independent studies, none refereed.
13 preprint, not refereed
Metadata freshness
The strongest single correlation in the GEO-16 analysis (r=0.68). That analysis is an unrefereed preprint over 1,100 English B2B SaaS URLs at one point in time, and its authors state the design cannot establish causation.
13 peer-reviewed
Citation hooks
Citing sources, adding quotations and adding statistics improved visibility 30-40% on a position-adjusted word-count metric (Aggarwal et al., ACM SIGKDD 2024). The effect is rank-dependent: +115.1% for a page ranked fifth, -30.3% for one already ranked first.
9 our own reasoning
Confidence grounding
Our own requirement, not an external ranking signal. A study of four generative search engines found only 51.5% of generated sentences were fully supported by their citations, which is the failure we are guarding against.
7 our own reasoning
Entity coverage
Held on a single source and treated as a hypothesis rather than an established lever. Weighted accordingly.
6 preprint, not refereed
Internal linking
Correlates at r=0.57 in the same unrefereed GEO-16 analysis. Weighted modestly for that reason.
5 peer-reviewed
Schema markup ready
Demoted from 9 points to 5. A matched difference-in-differences study of 1,885 pages that added JSON-LD, against 4,000 controls, found no meaningful citation lift. We still generate and validate it, because it is hygiene — it is simply not a lever.
1 our own reasoning
Conversational tone
A minor signal. Weighted at the bottom because we cannot defend more.
0 our own reasoning
Multilingual consistency
Dropped to zero. We could find no evidence of an independent effect, so it earns no points, even though we believed in it initially.
What we will not claim
The evidence is thinner than this industry pretends.
An independent benchmark tested ten content-optimisation tactics across two tasks, six domains, 1,921 queries and 16,360 documents. Of 54 method-domain combinations, only three produced a statistically significant improvement in citation ranking.
That benchmark also found that adding statistics reduced citation ranking in 19 of 24 settings, and that where a document sits in the model's context matters more than any rewriting tactic tested. Its authors conclude that content optimisation "must be considered as a complement and not a replacement for traditional SEO". We agree, and we would rather tell you that than sell you a number.
Source: Puerto, Gubri, Green, Oh & Yun, "C-SEO Bench: Does Conversational SEO Work?", NeurIPS 2025 Datasets & Benchmarks Track.
See how your pages score today.
The same rubric, run against your own domain, free.
✓ Free✓ No card✓ Two minutes✓ GDPR
