We only do what the evidence supports
GEO is full of practices with nothing behind them. Below is how we decide: what research supports, we do; what it doesn't, we say so plainly — including things other vendors are billing you for.
Evidence-backed
| Tactic | Effect | Source | We do it |
|---|---|---|---|
| Server-side rendering | Binary prerequisite — client-side rendered sites have zero AI visibility | Vercel / MERJ，5 億次 GPTBot 抓取 | Yes |
| Statistics and citation density | +37% to +41% | KDD 2024 同儕審查（Princeton / Georgia Tech / IIT Delhi / Allen AI） | Yes |
| Third-party brand mentions | 0.664 correlation (backlinks only 0.218) | Ahrefs，75,000 個品牌研究 | No |
| Answer-first structure | Direct statements ~2x more likely cited; start/end outperform the middle by 30%+ | Stanford TACL 2024；Kevin Indig，120 萬則 ChatGPT 引用 | Yes |
| Schema markup | +22%; FAQPage nearly doubles ChatGPT citations | Semrush，304,805 個 URL；SE Ranking | Yes |
| Topical clusters / fan-out coverage | +161% citation likelihood | Surfer SEO，10,000 關鍵字研究 | Yes |
| Original research and proprietary data | 4.31x more citations than directory listings | Yext，1,720 萬次引用分析 | No |
| Content freshness | Half of citations from content under 13 weeks; 76.4% of ChatGPT's most-cited under 30 days | Ahrefs，1,700 萬次引用 | Yes |
| E-E-A-T signals and named authors | +31%; author credentials add 16% weight | Semrush，304,805 個 URL | Yes |
| Page speed (first contentful paint) | FCP under 0.4s is 3x more likely to be cited than over 1.13s | SE Ranking，129,000 網域 / 216,524 頁面 | Yes |
| Per-platform optimization | Only 11% domain overlap between ChatGPT and Perplexity sources | Search Atlas，550 萬則回應 | Yes |
| Citations reach beyond the top 10 | 62% of citations come from outside the top 10 organic (was 24% in July 2025) | Ahrefs，863K SERP / 400 萬 URL（2026 年 3 月） | Yes |
No supporting evidence — we don't do these, and you shouldn't pay for them
| Tactic | Effect | Source | We do it |
|---|---|---|---|
| llms.txt | Zero correlation. No platform confirms reading it, yet 844,000+ sites have added one. | SE Ranking，300,000 網域研究 | No |
| Vector embedding optimization | Theoretically sound but unmeasurable in practice, and indistinguishable from ordinary topical relevance advice | 無可複現的對照研究 | No |
| Speakable schema | A single unverified claim; no platform confirmation | 原設計用途為語音朗讀，非引用訊號 | No |
| Serving crawlers different content (shadow pages) | Classic cloaking; carries penalty risk | 我們不會做，也建議你拒絕任何提出這個做法的廠商 | No |
| Adversarial content injection | Legal and reputational risk; platform detection keeps improving | 明確避免 | No |
A line worth remembering from the industry: if a GEO vendor does not openly tell you that success in AI visibility is 80 percent solid fundamental SEO, they are selling snake oil. We agree — which is why we position this as an engineering service, not marketing magic.