Specific numbers and data points
What it checks
AI engines prefer answers with concrete data points over vague statements. Include specific numbers, percentages, and metrics in your content.
Why it matters
Adding concrete statistics to a source document raises the prominence of that source’s citations in a generative engine’s answer, relative to the same document carrying the same claims in vague quantitative language.
Evidence
- GEO (Aggarwal et al., KDD ’24) tested Statistics Addition as one of nine methods over GEO-bench (10,000 queries, five random seeds). Table 1: Statistics Addition scores 25.2 on the overall Position-Adjusted Word Count metric against a 19.3 no-optimization baseline (Quotation Addition 27.2, Cite Sources 24.6). The paper states its top methods “achieved a relative improvement of 30-40% on the Position-Adjusted Word Count metric and 15-30% on the Subjective Impression metric” — arxiv.org/…/2311.09735 (verified 2026-08-21)
- The abstract generalizes the result: “including citations, quotations from relevant sources, and statistics can significantly boost source visibility, with an increase of over 40% across various queries” — arxiv.org/…/2311.09735 (verified 2026-08-21)
- Confirmed on a live engine, not just a simulated one: “We also demonstrate the efficacy of Generative Engine Optimization on Perplexity.ai, a real-world generative engine and demonstrate visibility improvements up to 37%” — arxiv.org/…/2311.09735 (verified 2026-08-21)
- The benchmark discriminates rather than rewarding any edit: Keyword Stuffing scored 17.7, below the 19.3 unoptimized baseline, and the authors conclude such methods “offer little to no improvement on generative engine’s responses” (Table 1, §4) — arxiv.org/…/2311.09735 (verified 2026-08-21)
Limits
The effect is conditional on where the source already ranks. Table 2 of the same paper reports Statistics Addition’s relative visibility change by search rank: −20.6% at Rank-1, −3.9% at Rank-2, +8.1% at Rank-3, +10.0% at Rank-4 and +97.9% at Rank-5. For an already top-ranked page the measured effect is negative. This is a signal that helps low-visibility sources and can hurt leaders. Table 3 further shows the gain concentrates in “Law & Gov.”, “Debate” and “Opinion” query tags, not uniformly.
C-SEO Bench, a later independent benchmark, concludes “Most current C-SEO methods are not only largely ineffective but also frequently have a negative impact on document ranking, which is opposite to what is expected” (arxiv.org/…/2506.11097). Google states “You don’t need to write in a specific way just for generative AI search” (developers.google.com/…/ai-optimization-guide). The point that matters most for this audit is what was measured. The intervention was relevant statistics that support the document’s claims.
The detector, by contrast, fires on any unit-bearing number anywhere in main content — “30-day”, “$19”, “3-5 days”, a shipping estimate. A page can therefore pass without carrying a single supporting statistic. The grade attaches to the signal, not to this detector. All URLs verified 2026-08-21.
How it scores
This is the one signal in the answer-engine category with a controlled, peer-reviewed measurement of exactly the intervention it describes, on both a 10,000-query benchmark and a live commercial engine.
Sources
- GEO: Generative Engine Optimization (Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, Deshpande) — arXiv (Princeton University, IIT Delhi, Georgia Tech, Allen Institute for AI), study (verified 2026-08-21)
- C-SEO Bench: Does Conversational SEO Work? — Puerto, Gubri, Green, Oh, Yun (arXiv; NeurIPS 2025 Datasets & Benchmarks), study (verified 2026-08-20)
- AI features and your website — AI optimization guide (mythbusting section) — Google Search Central, vendor-doc (verified 2026-08-21)